This study leveraged the rapid evolution of female pheromones across the Drosophila genus to gain insight into how males coordinately adapt their detection and interpretation of these chemical cues to hone their mating strategies. While in some Drosophila species females produce unique pheromones that act to attract and arouse their conspecific males, the pheromones of most species are sexually monomorphic such that females possess no distinguishing chemosensory signatures that males can use for mate recognition. By comparing several close and distantly-related Drosophila species, it was revealed that D. yakuba males have evolved the distinct ability to use a sexually-monomorphic pheromone, 7-tricosene (7-T), as an excitatory cue to promote courtship, a sensory innovation that enables D. yakuba males to court in the dark thereby expanding their reproductive opportunities. To gain insight into the neural adaptations that enable 7-T to act as an excitatory cue, the functional properties were compared of two key nodes within the pheromone circuits of D. yakuba and a subset of its closest relatives. The instructive role of 7-T in D. yakuba arises from concurrent peripheral and central circuit changes: a distinct subpopulation of sensory neurons has acquired sensitivity to 7-T which in turn selectively signals to a distinct subset of P1 neurons in the central brain that trigger courtship behaviors. Such a modular circuit organization, in which different sensory inputs can independently couple to multiple parallel courtship control nodes, may facilitate the evolution of mate recognition systems by allowing males to take advantage of novel sensory modalities to become aroused. Together, these findings suggest how peripheral and central circuit adaptations can be flexibly linked to underlie the rapid evolution of mate recognition and courtship strategies across species (Coleman, 2023).

Sequenced behaviours, including locomotion, reaching and vocalization, are patterned differently in different contexts, enabling animals to adjust to their environments. How contextual information shapes neural activity to flexibly alter the patterning of actions is not fully understood. Previous work has indicated that this could be achieved via parallel motor circuits, with differing sensitivities to context. This study demonstrated that a single pathway operates in two regimes dependent on recent sensory history. The Drosophila song production system was leveraged to investigate the role of several neuron types in song patterning near versus far from the female fly. Male flies sing 'simple' trains of only one mode far from the female fly but complex song sequences comprising alternations between modes when near her. Ventral nerve cord (VNC) circuits are shaped by mutual inhibition and rebound excitability between nodes driving the two song modes. Brief sensory input to a direct brain-to-VNC excitatory pathway drives simple song far from the female, whereas prolonged input enables complex song production via simultaneous recruitment of functional disinhibition of VNC circuitry. Thus, female proximity unlocks motor circuit dynamics in the correct context. A compact circuit model was constructed to demonstrate that the identified mechanisms suffice to replicate natural song dynamics. These results highlight how canonical circuit motifs can be combined to enable circuit flexibility required for dynamic communication (Roemschied, 2023).

Accepting or rejecting a mate is one of the most crucial decisions a female will make, especially when faced with food shortage. Previous studies have identified the core neural circuity from sensing male courtship or mating status to decision-making for sexual receptivity in Drosophila females, but how hunger and satiety states modulate female receptivity is poorly understood. This study identified the neural circuit and its neuromodulation underlying the hunger modulation of female receptivity. Adipokinetic hormone receptor (AkhR)-expressing neurons inhibit sexual receptivity in a starvation-dependent manner. AkhR neurons are octopaminergic and act on a subset of Octβ1R-expressing LH421 neurons. Knocking down Octβ1R expression in LH421 neurons eliminates starvation-induced suppression of female receptivity. It was further found that LH421 neurons inhibit the sex-promoting pC1 neurons via GABA-resistant to dieldrin (Rdl) signaling. pC1 neurons also integrate courtship stimulation and mating status and thus serve as a common integrator of multiple internal and external cues for decision-making (Sun, 2023).

Vision is critical for the regulation of mating behaviors in many species. This study discovered that the Drosophila ortholog of human GABA (A) -receptor-associated protein (GABARAP; Atg8a) is required to fine-tune male courtship by modulating the activity of visual feedback neurons, lamina tangential cells (Lat). GABARAP is a ubiquitin-like protein that regulates cell-surface levels of GABA (A) receptors. Knocking down GABARAP or GABA (A) receptors in Lat neurons or hyperactivating them induces male courtship toward other males. Inhibiting Lat neurons, on the other hand, delays copulation by impairing the ability of males to follow females. Remarkably, the human ortholog of Drosophila GABARAP restores function in Lat neurons. Using in vivo two-photon imaging and optogenetics, Lat neurons were shown to be functionally connected to neural circuits that mediate visually-guided courtship pursuits in males. This work reveals a novel physiological role for GABARAP in fine-tuning the activity of a visual circuit that tracks a mating partner during courtship (Mabuchi, 2023).

Female sexual behavior as an innate behavior is of prominent biological importance for survival and reproduction. However, molecular and circuit mechanisms underlying female sexual behavior is not well understood. This study identified the Cholecystokinin-like peptide Drosulfakinin (DSK) to promote female sexual behavior in Drosophila. Loss of DSK function reduces female receptivity while overexpressing DSK enhances female receptivity. Two pairs of Dsk-expressing neurons in the central brain were found to promote female receptivity. DSK peptide acts through one of its receptors, CCKLR-17D3, to modulate female receptivity. Manipulation of CCKLR-17D3 and its expressing neurons alters female receptivity. It was further revealed that the two pairs of Dsk-expressing neurons receive input signal from pC1 neurons that integrate sex-related cues and mating status. These results demonstrate how a neuropeptide pathway interacts with a central neural node in the female sex circuitry to modulate sexual receptivity (Wang, 2022).

This study systematically investigated DSK-mediated neuromodulation of female sexual receptivity. At the molecular level, it was revealed that DSK neuropeptide and its receptor CCKLR-17D3 are crucial for modulating female sexual receptivity. At the neuronal circuit level, it was identified that DSK neurons are the immediate downstream targets of sex-promoting pC1 neurons in controlling female sexual receptivity. Moreover, intersectional tools were employed to subdivide DSK neurons into medial DSK neurons (DSK-MP1) and lateral DSK neurons (DSK-MP3) and uncovered that DSK-MP1 neurons rather than DSK-MP3 neurons play essential roles in modulating female receptivity. Collectively, these findings illuminate a pC1-DSK-MP1-CCKLR-17D3 pathway that modulates female sexual behaviors in Drosophila (Wang, 2022).

The female sexual behavior is a complex innate behavior. The decision for the female to accept a courting male or not depends on not only sensory stimulation but also internal states. If the female is willing to mate, she slows down, pauses, and opens her vaginal plates to accept a courting male, if not, she extrudes her ovipositor to deter a courting male or flies away. Tbe results of this study show that DSK signaling is crucial for virgin female receptivity but has no effect on ovipositor extrusion behavior. How exactly does the DSK signaling regulate virgin female receptivity is still not clear. One possibility is that DSK signaling regulates pausing behavior in response to male courtship, as DSK receptor CCKLR-17D3 expresses in the central complex that has been found to be crucial for locomotor behaviors. However, no change was observed in locomotor activity in DSK or CCKLR-17D3 mutant females. Note only locomotor behavior was studied in single females but not in females paired with courting males due to technical limit for analysis, and it is possible that the DSK signaling does not affect general locomotor behavior but regulates courtship-stimulated pausing behavior (Wang, 2022).

The four pairs of DSK neurons are classified into two types (DSK-MP1 and DSK-MP3) based on the location of the cell bodies, and DSK-MP1 neurons extend descending fibers to ventral nerve cord. In this study, it was also found that activating DSK-MP1 neurons enhance female receptivity whereas inactivating DSK-MP1 neurons reduce female receptivity. Silencing adult Abd-B neurons and SAG neurons located in the abdominal ganglion inhibits female sexual receptivity. It has been found that Abd-B neurons control female pausing behavior, and it would be interesting to further investigate whether DSK-MP1 neurons relay information from Abd-B neurons to regulate pausing and receptivity in females. It is also noted that DSK-MP1 neurons extend projections to suboesophageal ganglion (SOG), and the SOG is the terminus of ascending projections from a subset of female reproductive tract sensory neurons labeled by pickpocket (ppk), fruitless (fru), and doublesex (dsx). It is also possible that DSK-MP1 neurons may integrate information directly from these sensory neurons to regulate female receptivity. Another study further classified the DSK-MP1 neurons into two types (MP1a and MP1b) based on the morphology of their neuritis. Future studies would further build genetic tools to uncover the function of each subset of DSK neurons in regulating distinct innate behaviors, such as male courtship, aggression, and female sexual behavior (Wang, 2022).

Previous studies have revealed that pC1 neurons extend projections to lateral protocerebral complex (LPC) and this neural cluster responds to courtship song and cVA. Moreover, recent works have shown that DSK-MP1 neurons project to the region of LPC. This study used GRASP, trans-Tango, and patch-clamp techniques and revealed that DSK-MP1 neurons are direct downstream target of R71G01-GAL4 neurons that include pC1. EM reconstruction revealed that pC1 neurons have intense synaptic input on MP1b but not MP1a neurons, suggesting a crucial role of the single pair of MP1b neurons in female receptivity. Based on these findings, it is propose that: (1) pC1 neurons act as a central node for female sexual receptivity by integrating sex-related sensory cues (courtship song and cVA) and mating status; and (2) DSK-MP1 neurons may integrate internal states and pC1-encoded information to modulate female sexual behavior. Thus, it is of prime importance to further investigate how a neuropeptide pathway modulate a core neural node in the sex circuitry to fine-tune the female's willingness for sexual behavior in the future (Wang, 2022).

Newborns and hatchlings can perform incredibly sophisticated behaviors, but many animals abstain from sexual activity at the beginning of life. Hormonal changes have long been known to drive both physical and behavioral changes during adolescence, leading to the largely untested assumption that sexuality emerges from organizational changes to neuronal circuitry. This study shows that the transition to sexuality in male Drosophila is controlled by hormonal changes, but this regulation is functional rather than structural. In very young males, a broadly acting hormone directly inhibits the activity of three courtship-motivating circuit elements, ensuring the complete suppression of sexual motivation and behavior. Blocking or overriding these inhibitory mechanisms evokes immediate and robust sexual behavior from very young and otherwise asexual males. Similarities to mammalian adolescence suggest a general principle in which hormonal changes gate the transition to sexuality not by constructing new circuitry but by permitting activity in otherwise latent motivational circuit elements (Zhang, 2021).

The identification and evaluation of a potential mating partner require multisensory integration, but, in the experimental paradigm used in this study, the decision to court is ultimately triggered by a male tapping a female with his pheromone receptor-bearing leg. The tap delivers parallel excitatory and inhibitory inputs to the male's P1 courtship command neurons, which initiate and maintain courtship when sufficiently stimulated. The sensitivity of P1 neurons to the inhibitory input from a tap is decreased by a local dopamine signal, which, in mature males, is tuned to reflect recent mating history. If a male has not mated for several days, the dopamine tone is high, increasing the probability that a tap will lead to courtship. Once courtship has commenced, the same dopamine signal maintains courtship bouts for the tens of seconds to several minutes required before mating starts. This motivating dopaminergic activity is maintained for days in the absence of females, is decremented by each mating, and slowly recovers over 3 to 4 days (Zhang, 2021).

Matings promote satiety by activating a set of copulation reporting neurons (CRNs). These neurons, as a population, project dendrites to the external genitalia and send axons to the brain, where they reduce the activity of fruitless-positive neuropeptide F (NPF) producing neurons. This decrease is relayed through the NPF receptor to the motivation-promoting dopamine neurons, reducing their activity. After a few matings, substantial satiety is induced, and reproductive motivation remains low for several days. Mating drive has an intrinsic tendency to recover because of recurrent excitation: NPF neurons excite and are excited by Doublesex-positive pCd neurons, forming a loop that holds and gradually accumulates activity during periods of abstinence. Loop activity is prevented from immediately rebounding by the activity-dependent transcription factor CREB2, which, during the period of high motivation that precedes mating, transcribes inhibitory genes (e.g., potassium channels) that will sustain the decremented activity state, forcing recovery to proceed on a biochemical, rather than electrical, time scale (Zhang, 2021).

This circuit architecture suggests several hypotheses for the prevention of courtship in newly eclosed males. For example, a juvenile male might not recognize or tap females; the CRNs may be constantly active, inducing overwhelming satiety; or, as in the organizational hypothesis, some or all of the courtship circuitry may not be fully developed. This study presents evidence refuting all these hypotheses. Instead, this study found that a spike in juvenile hormone levels at eclosion directly and selectively imposes long-lasting activity suppression on all known motivation-promoting circuit elements: both populations of loop neurons and the downstream dopaminergic neurons. Overriding any of these repressive mechanisms evokes robust courtship from extremely young males, providing clear evidence that much of the reproductive circuitry is developed and functional but lays dormant in juveniles. The multitiered suppression of motivational circuitry appears to be the key difference between the complete inactivation of mating drive in early life and the fluctuations experienced in adulthood (Zhang, 2021).

The story of Eden indicates that the diverse behavioral changes coinciding with reproductive maturity had been fascinating and frustrating humans thousands of years before they were known to involve the brain. The hormonal signaling pathways that trigger these changes in mammals are now understood, but little is known about their mechanistic impact on behavioral and motivational circuitry. Teleological explanations for delays in the onset of sexual behavior likely vary across species, but the pervasiveness of this phenomenon suggests the possibility of a core mechanistic conservation that transcends idiosyncrasies of duration, purpose, and even the molecular nature of the hormonal trigger. The system described in this paper for analyzing the transition to sexuality allows rapid insight, with principles that suggest molecular and circuit hypotheses in other animals and other late-emerging brain functions (Zhang. 2021).

One obvious but likely superficial difference in the control of behavioral maturation between Drosophila and mammals is the sign of the regulation by circulating hormones. Although some gonadal hormones can suppress mating drive, in juvenile male rodents, experimental elevation of testosterone causes precocious mating behavior, whereas this study found that juvenile hormone suppresses mating behavior. This discordance is reconciled early in signal transduction, as loss of the Met receptor for juvenile hormone decreases sexual behavior, similar to the effects of androgen (testosterone) receptor antagonists in humans. The receptors for testosterone, estrogen, and progesterone are all transcription factors, as are the Met and Gce receptors for juvenile hormone. Unlike ablating the source of juvenile hormone or overriding its action at the level of neuronal activity or CREB activation, removing the suppressive receptor, Gce, does not cause substantial juvenile courtship. This points to the likely existence of yet another juvenile hormone receptor, the identification of which will provide a deeper understanding of the suppressive mechanisms used to completely, selectively, and transiently inactivate this fundamental drive (Zhang. 2021).

In mature animals, the recurrent loop that drives the male to court also primes itself for satiety through activity-induced, CREB-mediated production of suppressive TASK7-containing potassium channel complexes. This motivation-suppressing module is also used in juveniles, with hormonal signaling causing CREB2 activation. Although additional suppressive effectors have not yet been identified in juveniles, evidence is seen of fast-acting suppression by juvenile hormone and direct suppression of the dopaminergic neurons, neither of which can be accounted for by CREB activity in the pCd/NPF loop. Parallel fast and slow responses have also been found in the response to mammalian gonadal hormones. Multiple repressive mechanisms are likely necessary to completely inactivate mating drive, since individually removing or overriding individual suppressive effectors allows juvenile courtship. This is again similar to findings in mammals, where administering testosterone to either the preoptic area of the hypothalamus or to the medial amygdala suffices to restore mating behaviors in castrated males (Zhang. 2021).

A second clear difference between Drosophila and vertebrate juvenile stages is the time scale. The fly system requires suppression of sexual behavior for days, not months or years [even 150 years as suggested in Greenland sharks]. However, several nonneuronal structures are still maturing in juvenile flies, such as the abdominal musculature, the cuticle, and the ejaculatory bulb, indicating that the neuronal circuitry that will eventually animate these features may require structural development as well. Although the kind of detailed anatomical analysis required to argue for or against fine-scale structural rearrangements, courtship circuitry appears grossly normal in juveniles, and the stimulation of several circuit elements (P1, dopaminergic, NPF, and pCd) can rapidly drive coherent courtship. While it is not yet known whether the courtship performed by juvenile males is identical to that of mature males (e.g., in song production),the current findings serve as strong evidence against a strict developmental ontogeny for the posteclosion emergence of reproductive motivation and behavior (Zhang. 2021).

A recent report found structural changes in a key hypothalamic population during estrus and showed that these neurons could not drive sexual behaviors when activated in ovariectomized mice. It is noted that it was not possible to restore sexual behavior in male flies if subpopulations of individually necessary dopaminergic neurons are stimulated, a consideration that leaves open the possibility that the motivational and behavioral circuitry may be at least partially functional, but suppressed, in nonestrus female mice. The most parsimonious explanation may be that both structural and activational changes take place between asexual and sexual periods of life. Increased connectivity in motivational circuitry could be a result, rather than a cause, of increased activity, as has been reported in a growing number of systems. No reports were found of neuronal stimulation eliciting juvenile sexual behavior in the mammalian literature, but stimulation of hypothalamic neurons can produce coherent and robust parenting behaviors from otherwise nonparental mice, demonstrating latent but functional circuitry for these late-emerging behaviors (Zhang. 2021).

Recently, new functions for pCd and NPF neurons have been reported in male-specific behaviors. In the Anderson laboratory, pCd neurons were found to sustain sexual and aggressive behaviors after P1 activation. This sustained excitation is remarkably similar to the role of pCd in the accrual and maintenance of sexual motivation in mature male flies. Since P1 activity is constantly required to sustain courtship, it is suggested that subsets of the eight pCd neurons may be specialized for maintaining recurrent activity over minutes (together with P1 neurons in sustaining courtship and aggression) and over days (together with NPF neurons in sustaining reproductive motivation). In work from the Montell laboratory, male-specific NPF neurons were reported to decrease sexual motivation generally and to prevent male-male courtship in particular. These conclusions appear to be in direct contrast with the courtship-promoting roles for these same NPF neurons described in earlier work. The results obtained across these studies are derived from a variety of assays and genetic manipulations, complicating direct comparisons, although they clearly point to multiple roles for NPF in courtship target selection and motivation. While several experiments have argued for various forms of hypersexuality resulting from decreased NPF signaling, the lone NPF neuronal stimulation experiment to yield a reduction in courtship behavior assessed courtship toward decapitated females and the effect was modest. The robust effects seen in this study make the authors confident that increasing the output of NPF neurons promotes a net increase in sexual motivation in otherwise asexual juveniles and in satiated males recovering from ad libitum access to virgin females (Zhang. 2021).

These results provide an alternative explanation for what has been the strongest evidence in favor of the organizational hypothesis for the maturation of reproductive behaviors in mammals: the delay between hormonal changes and the behaviors they induce. In the fly, the long-lasting, CREB-imposed effects of an earlier suppressive hormonal state must decay away to allow loop activity to ramp up and promote sexual behavior. Enduring and distributed suppressive effects may explain why the awakening of sexuality triggered by hormonal changes is gradual and halting in both flies and mammals (Zhang. 2021).

Acoustic communication signals diversify even on short evolutionary time scales. To understand how the auditory system underlying acoustic communication could evolve, this study conducted a systematic comparison of the early stages of the auditory neural circuit involved in song information processing between closely-related fruit-fly species. Male Drosophila melanogaster and D. simulans produce different sound signals during mating rituals, known as courtship songs. Female flies from these species selectively increase their receptivity when they hear songs with conspecific temporal patterns. This study firstly confirmed interspecific differences in temporal pattern preferences; D. simulans preferred pulse songs with longer intervals than D. melanogaster. Primary and secondary song-relay neurons, JO neurons and AMMC-B1 neurons, shared similar morphology and neurotransmitters between species. The temporal pattern preferences of AMMC-B1 neurons were also relatively similar between species, with slight but significant differences in their band-pass properties. Although the shift direction of the response property matched that of the behavior, these differences are not large enough to explain behavioral differences in song preferences. This study enhances understanding of the conservation and diversification of the architecture of the early-stage neural circuit which processes acoustic communication signals (Ohashi, 2023).

Animals perform a series of actions in a fixed order during ritualistic innate behaviors. Although command neurons and sensory pathways responding to external stimuli that trigger these behaviors have been identified, how each action is induced in a fixed order in response to multimodal sensory stimuli remains unclear. In this study, the sexually dimorphic lateral antennal lobe tract projection neuron 4 (lPN4) in male Drosophila melanogaster mediates the expression of a fixed behavioral action pattern at the beginning of the courtship ritual, in which a male taps a female body and then extends a wing unilaterally to produce a courtship song. Blocking the synaptic output of lPN4 caused an increase in the ratio of male flies that extended a wing unilaterally without tapping the female body, whereas excitation of lPN4 suppressed the transition from the tapping phase to the unilateral wing extension phase. Real-time calcium imaging showed that lPN4 is activated by a volatile pheromone, palmitoleic acid, whose responses were inhibited by simultaneous gustatory stimulation with female cuticular hydrocarbons, showing the existence of an "AND-gate" for multimodal sensory inputs during male courtship behaviors. These results suggest that the function of lPN4 is to suppress unilateral wing extension while responding to a female smell, which is released by appropriate contact chemosensory inputs received when tapping a female. As the female smell also promotes male courtship behaviors, the olfactory system is ready to simultaneously promote and suppress the progress of courtship actions while responding to a female smell (Tanaka, 2021).

Choosing a mate is one of the most consequential decisions a female will make during her lifetime. A female fly signals her willingness to mate by opening her vaginal plates, allowing a courting male to copulate. Vaginal plate opening (VPO) occurs in response to the male courtship song and is dependent on the mating status of the female. How these exteroceptive (song) and interoceptive (mating status) inputs are integrated to regulate VPO remains unknown. This study characterize the neural circuitry that implements mating decisions in the brain of female Drosophila melanogaster. VPO was shown to be controlled by a pair of female-specific descending neurons (vpoDNs). The vpoDNs receive excitatory input from auditory neurons (vpoENs), which are tuned to specific features of the D. melanogaster song, and from pC1 neurons, which encode the mating status of the female. The song responses of vpoDNs, but not vpoENs, are attenuated upon mating, accounting for the reduced receptivity of mated females. This modulation is mediated by pC1 neurons. The vpoDNs thus directly integrate the external and internal signals that control the mating decisions of Drosophila females (Wang, 2020).

Drosophila males woo potential mates by vibrating their wings to produce a species-specific courtship song. The male song induces deflections of the female aristae, thereby activating auditory sensory neurons that project to the central brain. Several types of song-responsive neurons have been identified in the female brain, but it is unknown whether and how these neurons regulate sexual receptivity. How a female responds to the song of a male is highly dependent on whether or not she has previously mated. Once mated, females store sperm for days to weeks, and during this time are reluctant to mate again. A male seminal fluid peptide (Sex peptide) binds to sperm and signals the presence of sperm in the female reproductive tract through an ascending pathway from the sex peptide sensory neurons (SPSNs) in the uterus via the sex peptide abdominal ganglion (SAG) neurons in the ventral nerve cord to the pC1 neurons in the brain. Sex peptide attenuates neuronal activity in the SPSN, SAG and pC1 neurons, thereby reducing sexual receptivity after mating. This study sought to investigate how these distinct external and internal signals are integrated in the female brain to control VPO, the motor output that signals the willingness of the female to mate (Wang, 2020).

Female receptivity is impaired by blocking the activity of the approximately 2,000 neurons that express either of the two sex-determination genes, fruitless (fru) or doublesex (dsx). This class of neurons includes the fru+ dsx+ SPSNs, the dsx+ SAGs13 and the dsx+ pC1 cells3. To search for other fru+ or dsx+ neurons that contribute to female receptivity, a collection of 234 sparse driver lines specific for various fru+ or dsx+ cell types were screened. These driver lines were used to genetically silence each of these cell types, and virgin females were assayed for their frequency of copulation within 10 min of being individually paired with naive wild-type males. Of the seven lines with the strongest reduction in receptivity, two labelled the SPSNs, one labelled the SAGs and one labelled the pC1 cells. The other three lines targeted a pair of female-specific descending neurons, which were named vpoDNs. These neurons are dsx+, fru- and cholinergic. Their dendrites arborize primarily in the lateral protocerebrum and their axons project to multiple regions of the ventral nerve cord, including the abdominal ganglion (Wang, 2020).

Acute optogenetic silencing or genetic ablation of the vpoDNs rendered virgin females unreceptive, markedly reducing the frequency of VPO but not the intensity of male courtship. Conversely, photoactivation of vpoDNs reliably triggered VPO in isolated virgin females. No peripheral expression was detected that was driven by the vpoDN lines, and by severing the abdominal nerve, it was confirmed that the VPO response is indeed due to activation of central neurons (Wang, 2020).

Mated females are less receptive than virgins, which was found to correlate with a lack of VPO. To assess whether the failure to perform VPO accounts for the low receptivity of mated females, the vpoDNs were photoactivated in mated females as they were being courted by wild-type males. Whereas control females never copulated during a 1-h assay, approximately 30-50% of the vpoDN-activated females did remate. A similar remating frequency was observed upon vpoDN activation in mated females paired with wingless males, which court but cannot sing. Thus, direct activation of vpoDNs bypasses the need for both the internal state (virginity) and the external cue (song) that normally combine to elicit VPO (Wang, 2020).

Wing extension was found to be the most frequent male action just before female VPO, and both VPO and copulation rates were found to be reduced if males are muted by removing their wings or females deafened by removing their aristae. These results suggested that the vpoDNs might be activated by male song. Indeed, in two-photon calcium-imaging experiments, a robust increase of calcium levels was detected in the neurites of vpoDNs in virgin females upon playback of male courtship song, but not in response to white noise. The response to courtship song was lost when the aristae were immobilized to deafen the female. Song responses have also been reported for the pMN2 neurons, which are morphologically similar to vpoDNs and also dsx+, although their reported functions differ (Wang, 2020).

The vpoDN dendrites lie mostly in the superior lateral protocerebrum, with no obvious arborizations within the antennal mechanosensory centre (AMMC), the primary auditory neuropil, or in the wedge region, a secondary auditory neuropil known to include song-responsive neuron. Therefore attempts were made to trace potential pathways from these auditory centres to the vpoDNs within the electron microscopy volume of a full adult female brain (FAFB). A single vpoDN was detected in each hemisphere and the vpoDN was extensively traced in the right hemisphere as well as its presynaptic partners, identifying a total of 45 neurons with at least 10 synapses impinging onto vpoDN (Wang, 2020).

None of the vpoDN input neurons innervate the AMMC, but at least two cell types have extensive arborizations within the wedge. Multiple split-GAL4 driver lines specific for these two cell types WERE obtained. Fluorescence in situ hybridization predicted, and whole-cell recording confirmed, that one of these cell types is excitatory and the other is inhibitory. Accordingly, these two cell types were named the vpoENs and vpoINs, respectively. Within FAFB there are two vpoEN cells and 14 vpoIN cells in each hemisphere (Wang, 2020).

Optogenetic silencing and activation experiments were performed to examine the roles of vpoENs and vpoINs in VPO and receptivity. Acute inhibition of the vpoENs significantly reduced the frequency of copulation and VPO when virgin females were paired with males. Conversely, strong optogenetic activation of vpoENs elicited VPO in isolated females, mimicking activation of vpoDNs. In virgin females paired with males, activating vpoINs suppressed mating and VPO, whereas silencing vpoINs had no effect. Thus, vpoENs and vpoINs promote and suppress, respectively, both VPO and receptivity (Wang, 2020).

Using two-photon calcium imaging, it was found that both vpoENs and vpoINs, as with vpoDNs, responded to playback of male courtship songs. The courtship song varies considerably between different Drosophila species and is the primary cue the female uses for species recognition. To test whether the vpoDNs, vpoENs and vpoINs are specifically tuned to the D. melanogaster courtship song, natural courtship songs were presented from seven other Drosophila species, selecting two representative audio clips from each species. The vpoDNs showed little or no response to any of these songs, the vpoENs responded to one or two clips from five species, and the vpoINs responded to all but one clip from one species (Wang, 2020).

Artificial pulse songs were generated for each of the other species, again systematically altering the IPI from 10 ms to 300 ms. Notably, the vpoDNs responded to the pulse songs of five other species once their IPI was shifted to match that of the D. melanogaster song. Together, these data establish that the vpoDNs are finely tuned to the D. melanogaster pulse song, owing to their selectivity for an IPI of about 35 ms. This narrow tuning may arise through a combination of strong excitation from highly selective vpoENs and weak inhibition from broadly responsive vpoINs (Wang, 2020).

Having determined how auditory input controls VPO and sexual receptivity, how this response is modulated by the mating status of the female was examined. VPO may be attenuated after mating either because the vpoENs and vpoDNs are less potent at eliciting VPO, or because they are less excited by song. In optogenetic activation and calcium-imaging experiments, it was found that vpoDNs are equally potent in mated and virgin females, whereas both the basal calcium levels and the response to courtship song were lower in mated females than in virgins. By contrast, the vpoENs were significantly less potent at eliciting VPO in mated than in virgin females. Although basal fluorescence of vpoENs was slightly higher in mated females than in virgins, their song responses were indistinguishable. Calcium levels were imaged in vpoINs, and it was found that the basal fluorescence and song responses of these cells were similar in mated and virgin females. Thus, these data show that vpoENs have a similar response to song in mated females as they do in virgins, but they are less able to excite vpoDNs in mated females (Wang, 2020).

It is concluded that the decision of the female fly to mate or not to mate is largely determined by how the vpoDNs integrate signals from two direct synaptic inputs: the vpoENs, which are selectively tuned to the conspecific male courtship song, and the pC1 cells, which encode the mating status of the female. When the male sings, female vpoENs are activated; whether or not this leads to vpoDN activation and hence VPO depends on the level of pC1 activity, which is higher in virgins than in mated females. The neural computation that underlies this state-dependent sensorimotor transformation remains to be determined; this will require methods for simultaneously manipulating and recording from all three cell types. One possibility is that the pC1 inputs gate the vpoEN inputs in a nonlinear fashion. No obvious spatial segregation of vpoEN and pC1 synapses onto the vpoDN dendrites was noted, as might be expected if these inputs are indeed processed hierarchically. Alternatively, vpoDNs might simply use a sum-to-threshold mechanism, in which the combined input from vpoENs and pC1s must exceed a certain level to elicit action potentials in vpoDNs. In this scenario, the lower pC1 activity after mating would necessitate a stronger vpoEN input to activate the vpoDNs. This model may account for the observation that wild-caught females are often multiply mated, consistent with the prediction from evolutionary theory that a mated female would increase her reproductive fitness by remating when she is courted by a male of higher quality than her first partner (Wang, 2020).

The many other, as yet uncharacterized, inputs to pC1, vpoEN and vpoDN cells may convey additional signals that modulate female receptivity. For example, pC1 cells are reported to respond to a male pheromone, which may serve to enhance the receptivity of both virgin and mated females. The persistent enhancement of vpoDN song responses upon transient activation of pC1 cells resembles the persistent state of courtship arousal induced in males by transient activation of the male pC1 counterparts. The female pC1 cells may therefore encode both mating status and, as with their male counterparts, a lasting state of mating arousal induced by sensory cues from potential mates. The ensuing interaction between the two sexes involves a coordinated sequence of signals and responses, as exemplified by the male singing to elicit female VPO. In both sexes, these sensorimotor transformations may not be directly mediated by pC1 cells, as commonly thought, but rather modulated by the arousal states they encode. The neural architecture reported in this study for the control of Drosophila female sexual receptivity may thus also serve as a paradigm for understanding male sexual behaviour, and perhaps more generally for the state-dependent signal processing that underlies behavioural decisions across a range of species (Wang, 2020).

Endocrine state is an important determinant of learning and memory in animals. In Drosophila, rejection of male courtship overtures by mated females leads to an aversive response manifested as courtship memory. This study reports that ecdysis triggering hormone (ETH) is an obligatory enabler of long-term courtship memory (LTM). ETH deficiency suppresses LTM, whereas augmented ETH release reduces the minimum training period required for LTM induction. ETH receptor knockdown either in the mushroom body (MB) γ lobe or in octopaminergic dorsal-anterior-lateral (DAL) neurons impairs memory performance, indicating its direct action in these brain areas. Consistent with these findings, brain exposure to ETH mobilizes calcium in MB γ lobe neuropils and DAL neurons. ETH receptor (ETHR) knockdown in the corpus allatum (CA) to create juvenile hormone (JH) deficiency also suppresses LTM, as does knockdown of the JH receptor Met in the MB γ lobe, indicating a convergence of ETH and JH signaling in this region of the brain. These findings identify endocrine-enabled neural circuit components in the brain that are critical for persistent behavioral changes resulting from aversive social experience (Lee, 2021).

Innate behaviors involve both reflexive motor programs and enduring internal states, but how these responses are coordinated by the brain is not clear. In Drosophila, male-specific P1 interneurons promote courtship song, as well as a persistent internal state that prolongs courtship and enhances aggressiveness. However, P1 neurons themselves are not persistently active. This study identified pCd neurons as persistently active, indirect P1 targets that are required for P1-evoked persistent courtship and aggression. Acute activation of pCd neurons alone is inefficacious but enhances and prolongs courtship or aggression promoted by female cues. Brief female exposure induces a persistent increase in male aggressiveness, an effect abrogated by interruption of pCd activity. pCd activity is not sufficient but necessary for persistent physiological activity, implying an essential role in a persistence network. Thus, P1 neurons coordinate both command-like control of courtship song and a persistent internal state of social arousal mediated by pCd neurons (Jung, 2020).

Neuromodulators such as monoamines are often expressed in neurons that also release at least one fast-acting neurotransmitter. The release of a combination of transmitters provides both "classical" and "modulatory" signals that could produce diverse and/or complementary effects in associated circuits. This study establish that the majority of Drosophila octopamine (OA) neurons are also glutamatergic and identify the individual contributions of each neurotransmitter on sex-specific behaviors. Males without OA display low levels of aggression and high levels of inter-male courtship. Males deficient for dVGLUT solely in OA-glutamate neurons (OGNs) also exhibit a reduction in aggression, but without a concurrent increase in inter-male courtship. Within OGNs, a portion of VMAT and dVGLUT puncta differ in localization suggesting spatial differences in OA signaling. These findings establish a previously undetermined role for dVGLUT in OA neurons and suggests that glutamate uncouples aggression from OA-dependent courtship-related behavior. These results indicate that dual neurotransmission can increase the efficacy of individual neurotransmitters while maintaining unique functions within a multi-functional social behavior neuronal network (Sherer, 2020).

Addressing the functional complexities of 'one neuron, multiple transmitters' is critical to understanding how neuron communication, circuit computation, and behavior can be regulated by a single neuron. Over many decades, significant progress has been made elucidating the functional properties of neurons co-expressing neuropeptides and small molecule neurotransmitters, where the neuropeptide acts as a co-transmitter and modulates the action of the neurotransmitter. Only recently have studies begun to examine the functional significance of co-transmission by a fast-acting neurotransmitter and a slow-acting monoamine (Sherer, 2020).

This study has demonstrated that OA neurons express dVGLUT and has utilized a new genetic tool to remove dVGLUT in OA-glutamate neurons. Quantifying changes in the complex social behaviors of aggression and courtship revealed that dVGLUT in brain OGNs is required to promote aggressive behavior and a specific behavioral pattern, the lunge. In contrast, males deficient for dVGLUT function do not exhibit an increase in inter-male courtship. These results establish a previously undetermined role for dVGLUT in OA neurons located in the adult brain and reveal glutamate uncouples aggression from inter-male courtship. It has been suggested that classical neurotransmitters and monoamines present in the same neuron modulate each other's packaging into synaptic vesicles or after release via autoreceptors. For example, a reduction of dVGLUT in DA-glutamate neurons resulted in decreased AMPH-stimulated hyperlocomotion in Drosophila and mice suggesting a key function of dVGLUT is the mediation of vesicular DA content. In this study, the independent behavioral changes suggests enhancing the packaging of OA into vesicles is not the sole function of dVGLUT co-expression and suggests differences in signaling by OA from OGNs on courtship-related circuitry (Sherer, 2020).

Co-transmission can generate distinct circuit-level effects via multiple mechanisms. One mechanism includes spatial segregation; the release of two neurotransmitters or a neurotransmitter and monoamine from a single neuron occurring at different axon terminals or presynaptic zones. Recent studies examining this possible mechanism have described; (i) the release of GLU and DA from different synaptic vesicles in midbrain dopamine neurons and (ii) the presence of VMAT and VGLUT microdomains in a subset of rodent mesoaccumbens DA neurons. This study expressed a new conditionally expressed epitope-tagged version of VMAT in OGNs and visualized endogenous dVGLUT via antibody labeling. Within OGNs, the colocalization of VMAT and dVGLUT puncta was not complete suggesting the observed behavioral phenotype differences may be due to spatial differences in OA signaling (Sherer, 2020).

A second mechanism by which co-transmission may generate unique functional properties relies on activating distinct postsynaptic receptors. In Drosophila, recent work has identified a small population of male-specific neurons that express the alpha-like adrenergic receptor, OAMB, as aggression-promoting circuit-level neuronal targets of OA modulation independent of any effect on arousal and separately knockdown of the Rdl GABAa receptor in a specific doublesex+ population stimulated male aggression. Future experiments identifying downstream targets that express both glutamate and octopamine receptors would be informative, as well as using additional split-Gal4 lines to determine if segregation of transporters is a hallmark of the majority of OGNs. Finally, a third possible mechanism is Glu may be co-released from OGNs and act on autoreceptors to regulate presynaptic OA release (Sherer, 2020).

Deciphering the signaling complexity that allows neural networks to integrate external stimuli with internal states to generate context-appropriate social behavior is a challenging endeavor. Neuromodulators including monoamines are released to signal changes in an animal's environment and positively or negatively reinforce network output. In invertebrates, a role for OA in responding to external chemosensory cues as well as promoting aggression has been well-established. In terms of identifying specific aggression circuit-components that utilize OA, previous results determined OA neurons directly receive male-specific pheromone information and the aSP2 neurons serve as a hub through which OA can bias output from a multi-functional social behavior network towards aggression. The ability of OA to bias behavioral decisions based on positive and negative reinforcement was also recently described for food odors. In vertebrates, it has been proposed that DA-GLU cotransmission in the NAc medial shell might facilitate behavioral switching. The finding that the majority of OA neurons are glutamatergic, suggests that the complex social behavior of aggression may rely on small subsets of neurons that both signal the rapid temporal coding of critical external stimuli as well as the frequency coding of such stimuli resulting in the enhancement of this behavioral network. One implication of the finding regarding the separable OA-dependent inhibition of inter-male courtship is the possibility of identifying specific synapses or axon terminals that when activated gate two different behavioral outcomes. A second implication is that aggressive behavior in other systems may be modified by targeting GLU function in monoamine neurons (Sherer, 2020).

Finally, monoamine-expressing neurons play key roles in human behavior including aggression and illnesses that have an aggressive component such as depression, addiction, anxiety, and Alzheimer's. While progress is being made in addressing the functional complexities of dual transmission, the possible pathological implications of glutamate co-release by monoamine neurons remains virtually unknown. Analyzing the synaptic vesicle and release properties of monoamine-glutamate neurons could offer new possibilities for therapeutic interventions aimed at controlling out-of-context aggression (Sherer, 2020).

Diffuse neuromodulatory systems such as norepinephrine (NE) control brain-wide states such as arousal, but whether they control complex social behaviors more specifically is not clear. Octopamine (OA), the insect homolog of NE, is known to promote both arousal and aggression. A systematic, unbiased screen identified OA receptor-expressing neurons (OARNs) that control aggression in Drosophila. The results uncover a tiny population of male-specific aSP2 neurons that mediate a specific influence of OA on aggression, independent of any effect on arousal. Unexpectedly, these neurons receive convergent input from OA neurons and P1 neurons, a population of FruM(+) neurons that promotes male courtship behavior. Behavioral epistasis experiments suggest that aSP2 neurons may constitute an integration node at which OAergic neuromodulation can bias the output of P1 neurons to favor aggression over inter-male courtship. These results have potential implications for thinking about the role of related neuromodulatory systems in mammals (Watanabe, 2017).

A rich behavioral literature has implicated OA in the control of invertebrate aggression, although the direction of its effects differs between species. Classic studies in lobsters have shown that injection of OA into the hemolymph promotes a subordinate-like posture, while injection of serotonin (5HT) produces a dominant-like posture. In contrast, hemolymph injections of OA in crickets restore aggressiveness to subordinated animals, mimicking the arousing effects of episodes of free flight. OA has also been suggested to play a role in aggressive motivation restored to defeated crickets by residency in a shelter. In Drosophila, null mutations of TβH strongly suppressed aggressiveness, suggesting a positive-acting role for OA in flies as in crickets. Interestingly, intra-hypothalamic infusion of NE in mammals can also enhance aggression. However, little is known about the neurons on which these amines act directly to influence aggression, in any organism (Watanabe, 2017).

This study applied a novel, unbiased approach to identify OARNs relevant to aggression in Drosophila. Importantly this screen was based not on mutations in OAR genes, but rather upon genetic silencing of neurons that express GAL4 under the control of different OAR gene cis-regulatory modules (CRMs). This screen was agnostic with respect to which OAR gene is involved, or in which neurons that OAR is expressed. It yielded a small population of male-specific, FruM+ OA-sensitive neurons, called aSP2, the activity of which isrequired for normal levels of aggressiveness. No significant change in UWEs (male-male courtship) was observed when these neurons were activated or silenced. Nevertheless, neuronal silencing in the parental R47A04-GAL4 line increased male-male courtship, perhaps reflecting an inhibitory role for non-aSP2 neurons in that line. Therefore, while it is not possible to completely exclude a role for aSP2 neurons to suppress male-male courtship, the evidence does not strongly support it (Watanabe, 2017).

Multiple lines of evidence suggest that R47A04^aSP2 neurons are indeed OA responsive, likely via OAMB. First, these neurons are labeled by a CRM from the Oamb gene. Second, RNAi-mediated knockdown of Oamb in R47A04 neurons reduced aggression, phenocopying the effects of an Oamb null allele. (However, knockdown using the split-GAL4 R47A04^aSP2driver only yielded a trend to reduced aggression that did not reach significance, perhaps reflecting a floor effect in this assay.) Third, overexpression of Oamb cDNAs in these neurons using R47A04-GAL4 rescued the Oamb null mutant and enhanced the effect of OA feeding to promote aggression. Fourth,R47A04^aSP2 neurons were activated by bath-applied OA in brain explants, and this effect was also blocked by RNAi-mediated knockdown of Oamb. Taken together, these data strongly suggest that aSP2 neurons respond directly to OA to mediate its effects on aggression, although they do not exclude a role for other OA-responsive non-aSP2 neurons in line R47A04. While it was not possible to definitively establish which of the 27 different classes of OANs in Drosophila provide functional OA input to aSP2 cells, some candidate OA neurons labeled in retrograde PA-GFP experiments (VUM and VPM) have previously been implicated in aggression (Watanabe, 2017).

In Drosophila OA, like NE in vertebrates, is thought to promote arousal. Consistent with such a function, OAergic fibers are broadly distributed across the entire Drosophila CNS, as are NE fibers in vertebrates. Thus OARNs could enhance aggression by increasing arousal, and there is evidence for such a function in crickets. However, manipulations of R47A04^aSP2neurons that increased or decreased aggression did not affect locomotion, circadian activity, or sleep. This suggests that these neurons influence aggression directly and specifically, rather than by increasing generalized arousal. Other classes of OARNs not investigated in this study have been implicated in sleep-wake arousal (Watanabe, 2017).

Does OA promote aggression in a permissive or instructive manner? While it is clear that OA synthesis and release are required for aggression in Drosophila, whether increasing OA suffices to promote aggression is less clear. It was reported that NaChBac-mediated activation of Tdc2-GAL4 neurons enhanced aggression, but the current study neither this manipulation, nor activation of Tdc2 neurons using dTrpA1 or Chrimson, yielded consistent effects. Thus, while OA is essential for normal levels of aggression, it is not clear whether it plays an instructive role to promote this behavior (Watanabe, 2017).

In principle, OA RNs could act directly in command-like neurons that mediate aggression, or rather in cells that play a modulatory role. It was found that aggression was increased by tonically enhancing the excitability of R47A04^aSP2 neurons using NaChBac, but not by phasically activating them optogenetically, arguing against a command-like function. Furthermore, the influence of TK FruM neurons, which do promote aggression when phasically activated, was not dependent on the activity of R47A04^aSP2 neurons, indicating that the latter are not functionally downstream of the former. Together, these data argue against a role for R47A04^aSP2 cells as command-like neurons, or as direct outputs of command neurons, for aggression. Rather, these cells exert a modulatory influence on agonistic behavior (Watanabe, 2017).

In searching for neurons that may interact with R47A04^aSP2 cells in their modulatory capacity, P1 neurons, a FruM+ population of 20 neurons/hemibrain was identified that controls male courtship, but which can also promote aggression when activated. It has been argued that this aggression-promoting effect is due to a subset of FruM neurons in the GAL4 line used in these studies, R71G01-GAL4. However, this study shows that conditional expression of FLP-ON Chrimson in a subset of neurons within the R71G01-GAL4 population using Fru-FLP. Nevertheless, these data do not exclude that the aggression-promoting neurons in the P1 cluster expressed Fru-FLP only transiently during development, nor do they exclude the possibility that different subpopulations of neurons within line R71G01 control courtship versus aggression; further studies will be required to resolve these issues (Watanabe, 2017).

The P1 cluster is known to project to downstream cells that are specific for courtship . The present study provides the first evidence that cells in this cluster also functionally activate (and physically contact) aggression-specific neurons. However, due to limitations of the genetic reagents employed, it is not certain that the behavioral, physiological, and anatomical interactions with aSP2 cells demonstrated in this study are all mediated by the same subset of neurons in the P1 cluster. With this caveat in mind, these data suggest that aSP2 neurons are functionally downstream of both a subset(s) of P1 neurons, as well as of OA neurons (Watanabe, 2017).

Feeding flies OA potentiated the activation of R47A04^aSP2 neurons by P1 neuron stimulation, in brain explants. Furthermore, activation of aggression by P1 stimulation was enhanced and suppressed by pharmacologically increasing or decreasing OA signaling, respectively. While some off-target effects of the drugs, or an action on non-aSP2 neurons expressing OARs, cannot be excluded these pharmacologic effects were overridden by opposite-direction genetic manipulations of R47A04^aSP2neuronal activity. Whether P1 neurons and OANs normally activate aSP2 neurons in vivo, simultaneously or sequentially, is not yet clear. Nevertheless it is striking that P1 and Tdc2 putative inputs occupy adjacent regions of aSP2 dendrites. Taken together, these findings suggest that aSP2 cells may serve as a node through which OA can bias output from a multifunctional social behavior network involving P1 neurons, in a manner that favors aggression. However, aSP2 neurons themselves do not appearto control directly the choice between mating and fighting (Watanabe, 2017).

Male-specific cuticular hydrocarbons such as 7-tricosene (7-T) are known to be required for aggression. Interestingly, it has recently been shown that gustatory neurons expressing Gr32a, which encodes a putative 7-T receptor, innervate OANs in the SEZ; these OANs are activated by 7-T in a Gr32a-dependent manner. SEZ-innervating OANs include the VPM/VUM subsets seen in PA-GFP retrograde labeling experiments. These data raise the possibility that R47A04^aSP2 neurons might be targets of VPM/VUM OANs activated by 7-T. If so, then they could provide a potential link between the influence of male-specific pheromones, OA, and central aggression circuitry. Studies of NE neurons in vertebrates have led to a prevailing view that this neuromodulator is released in a diffuse, 'sprinkler system'-like manner to control brain-wide states like arousal. Recent studies in Drosophila indicate that the broad, brain-wide distribution of OAergic fibers reflects the superposition of close to 30 anatomically distinct subclasses of OANs). The data presented in this study reveal a high level of circuit specificity for OARNs that mediate the effects of OA on aggression, mirroring the anatomical and functional specificity of OANs reported to control this behavior. If this anatomical logic is conserved, then such circuit specificity may underlie the actions of NE in mammals to a greater extent than is generally assumed (Watanabe, 2017).

How brains are hardwired to produce aggressive behavior, and how aggression circuits are related to those that mediate courtship, is not well understood. This large-scale screen for aggression-promoting neurons in Drosophila identifies several independent hits that enhance both inter-male aggression and courtship. Genetic intersections reveal that P1 interneurons, previously thought to exclusively control male courtship, are responsible for both phenotypes. The aggression phenotype is fly-intrinsic, and requires male-specific chemosensory cues on the opponent. Optogenetic experiments indicate that P1 activation promotes aggression vs. wing extension at low vs. high thresholds, respectively. High frequency photostimulation promotes wing extension and aggression in an inverse manner, during light ON and OFF, respectively. P1 activation enhances aggression by promoting a persistent internal state, which could endure for minutes prior to social contact. Thus P1 neurons promote an internal state that facilitates both aggression and courtship, and can control these social behaviors in a threshold-dependent manner (Hoopfer, 2015).

This study describes the first large-scale neuronal activation screen for aggression neurons in Drosophila. Using the thermosensitive ion channel dTrpA1, a collection of over 3,000 GAL4 lines was screened for flies that exhibited increased fighting following thermogenetic neuronal activation. Among ~20 hits obtained, three exhibited both increased aggression and male-male courtship behavior. Intersectional refinement of expression patterns using split-GAL4 indicated that both social behaviors are controlled, in all three hits, by a subpopulation of ~8-10 P1 neurons per hemibrain. P1 cells are male-specific, Fru^M+ interneurons that integrate pheromonal and visual cues to promote male courtship behavior. The results indicate, surprisingly, that at least a subset of P1 neurons, previously thought to control exclusively courtship, can promote male aggression as well. Moreover, it was shown that they exert this influence by inducing a persistent fly-intrinsic state, lasting for minutes, that enhances these behaviors. These data define a sexually dimorphic neural circuit node that may link internal states to the control of mating and fighting, and identify a potentially conserved circuit 'motif' for the control of social behaviors (Hoopfer, 2015).

In insects and mammals, olfactory experience in early life alters olfactory behavior and function in later life. In the vinegar fly Drosophila, flies chronically exposed to a high concentration of a monomolecular odor exhibit reduced behavioral aversion to the familiar odor when it is reencountered. This change in olfactory behavior has been attributed to selective decreases in the sensitivity of second-order olfactory projection neurons (PNs) in the antennal lobe that respond to the overrepresented odor. However, since odorant compounds do not occur at similarly high concentrations in natural sources, the role of odor experience-dependent plasticity in natural environments is unclear. This study investigated olfactory plasticity in the antennal lobe of flies chronically exposed to odors at concentrations that are typically encountered in natural odor sources. These stimuli were chosen to each strongly and selectively excite a single class of primary olfactory receptor neuron (ORN), thus facilitating a rigorous assessment of the selectivity of olfactory plasticity for PNs directly excited by overrepresented stimuli. Unexpectedly, it was found that chronic exposure to three such odors did not result in decreased PN sensitivity but rather mildly increased responses to weak stimuli in most PN types. Odor-evoked PN activity in response to stronger stimuli was mostly unaffected by odor experience. When present, plasticity was observed broadly in multiple PN types and thus was not selective for PNs receiving direct input from the chronically active ORNs. The DL5 olfactory coding channel was further investigated and it was found that chronic odor-mediated excitation of its input ORNs did not affect PN intrinsic properties, local inhibitory innervation, ORN responses or ORN-PN synaptic strength; however, broad-acting lateral excitation evoked by some odors was increased. These results show that PN odor coding is only mildly affected by strong persistent activation of a single olfactory input, highlighting the stability of early stages of insect olfactory processing to significant perturbations in the sensory environment (Gugel, 2023).

In Drosophila, a dedicated olfactory channel senses a male pheromone, cis-vaccenyl acetate (cVA), promoting female courtship while repelling males. This study shows that separate cVA-processing streams extract qualitative and positional information. cVA sensory neurons respond to concentration differences in a 5-mm range around a male. Second-order projection neurons encode the angular position of a male by detecting inter-antennal differences in cVA concentration, which are amplified through contralateral inhibition. At the third circuit layer, 47 cell types were identified with diverse input-output connectivity. One population responds tonically to male flies, a second is tuned to olfactory looming, while a third integrates cVA and taste to coincidentally promote female mating. The separation of olfactory features resembles the mammalian what and where visual streams; together with multisensory integration, this enables behavioral responses appropriate to specific ethological contexts (Taisz, 2023).

Johnston's organ (Jo) acts as an antennal wind-sensitive and/or auditory organ across a spectrum of insect species and its axons universally project to the brain. In the locust, this pathway is already present at mid-embryogenesis but the process of fasciculation involved in its construction has not been investigated. Terminal projections into the fine neuropilar organization of the brain also remain unresolved, information essential not only for understanding the neural circuitry mediating Jo-mediated behavior but also for providing comparative data offering insights into its evolution. In this study, neuron-specific, axon-specific, and epithelial domain labels were employed to show that the pathway to the brain of the locust is built in a stepwise manner during early embryogenesis as processes from Jo cell clusters in the pedicel fasciculate first with one another, and then with the two tracts constituting the pioneer axon scaffold of the antenna. A comparison of fasciculation patterns confirms that projections from cell clusters of Jo stereotypically associate with only one axon tract according to their location in the pedicellar epithelium, consistent with a topographic plan. At the molecular level, all neuronal elements of the Jo pathway to the brain express the lipocalin Lazarillo, a cell surface epitope that regulates axogenesis in the primary axon scaffold itself, and putatively during fasciculation of the Jo projections to the brain. Central projections from Jo first contact the primary axon scaffold of the deutocerebral brain at mid-embryogenesis, and in the adult traverse mechanosensory/motor neuropils similar to those in Drosophila. These axons then terminate among protocerebral commissures containing premotor interneurons known to regulate flight behavior (Boyan, 2023).

Animals communicate using sounds in a wide range of contexts, and auditory systems must encode behaviorally relevant acoustic features to drive appropriate reactions. How feature detection emerges along auditory pathways has been difficult to solve due to challenges in mapping the underlying circuits and characterizing responses to behaviorally relevant features. This paper describes auditory activity in the Drosophila melanogaster brain and investigated feature selectivity for the two main modes of fly courtship song, sinusoids and pulse trains. Twenty-four new cell types of the intermediate layers of the auditory pathway were identified, and using a new connectomic resource, FlyWire, all synaptic connections between these cell types, in addition to connections to known early and higher-order auditory neurons were mapped-this represents the first circuit-level map of the auditory pathway. The sign (excitatory or inhibitory) was determined of most synapses in this auditory connectome. Auditory neurons were found to display a continuum of preferences for courtship song modes and that neurons with different song-mode preferences and response timescales are highly interconnected in a network that lacks hierarchical structure. Nonetheless, this study found that the response properties of individual cell types within the connectome are predictable from their inputs. This study thus provides new insights into the organization of auditory coding within the Drosophila brain (Baker, 2022).

Sensory pathways are typically studied by starting at receptor neurons and following postsynaptic neurons into the brain. However, this leads to a bias in analyses of activity toward the earliest layers of processing. This paper presents new methods for volumetric neural imaging with precise across-brain registration to characterize auditory activity throughout the entire central brain of Drosophila and make comparisons across trials, individuals and sexes. It was discovered that auditory activity is present in most central brain regions and in neurons responsive to other modalities. Auditory responses are temporally diverse, but the majority of activity is tuned to courtship song features. Auditory responses are stereotyped across trials and animals in early mechanosensory regions, becoming more variable at higher layers of the putative pathway, and this variability is largely independent of ongoing movements. This study highlights the power of using an unbiased, brain-wide approach for mapping the functional organization of sensory activity (Pacheco, 2021).

Flies detect sound using a feathery appendage of the antenna, the arista, which vibrates in response to near-field sounds. Antennal displacements activate mechanosensory Johnston's organ neurons (JONs) housed within the antenna. Three major populations of JONs (A, B and D) respond to vibratory stimuli at frequencies found in natural courtship song. These neurons project to distinct areas of the antennal mechanosensory and motor center (AMMC) in the central brain. Recent studies suggest that the auditory pathway continues from the AMMC to the wedge (WED), then to the ventrolateral protocerebrum (VLP) and to the lateral protocerebral complex (LPC). However, knowledge of the fly auditory pathway remains incomplete, and the functional organization of regions downstream of the AMMC and WED are largely unexplored. Moreover, nearly all studies of auditory coding in Drosophila have been performed using female brains, even though both males and females process courtship song information (Pacheco, 2021).

To address these issues, methods were developed to investigate the representation of behaviorally relevant auditory signals throughout the central brain of Drosophila and to make comparisons across animals. Two-photon microscopy was used to sequentially target the entirety of the Drosophila central brain in vivo, combined with fully automated segmentation of regions of interest (ROIs). In contrast to recent brain-wide imaging studies of Drosophila, temporal speed for was traded off for enhanced spatial resolution. Imaging at high spatial resolution facilitates automated ROI segmentation, with each ROI covering subneuropil structures, including cell bodies and neurites. ROIs were accurately registered into an in vivo template brain to compare activity across trials, individuals and sexes, and to build comprehensive maps of auditory activity throughout the central brain. The results reveal that the representation of auditory signals is broadly distributed throughout 33 out of 36 major brain regions, including in regions known to process other sensory modalities, such as all levels of the olfactory pathway, or to drive various motor behaviors, such as the central complex. The representation of auditory stimuli is diverse across brain regions, but focused on conspecific features of courtship song. Auditory activity is more stereotyped (across trials and individuals) at early stages of the putative mechanosensory pathway, becoming more variable and more selective for particular aspects of the courtship song at higher stages. This variability cannot be explained by simultaneous measurements of ongoing fly behavior. Meanwhile, auditory maps are largely similar between male and female brains, despite extensive sexual dimorphisms in neuronal number and morphology. These findings provide the first brain-wide description of sensory processing and feature tuning in Drosophila (Pacheco, 2021).

Sensory systems are typically studied starting from the periphery and continuing to downstream partners guided by anatomy. This has limited the understanding of sensory processing to early stages of a given sensory pathway. This study used a brain-wide imaging method to unbiasedly screen for auditory responses beyond the periphery and, via precise registration of recorded activity, to compare auditory representations across brain regions, individuals and sexes. Auditory activity was found to be widespread, extending well beyond the canonical mechanosensory pathway, and is present in brain regions and tracts known to process other sensory modalities (that is, olfaction and vision) or to drive motor behaviors. The representation of auditory stimuli diversified, in terms of both temporal responses to stimuli and tuning for stimulus features, from the AMMC to later stages of the putative pathway, becoming more selective for particular aspects of courtship song (that is, sine or pulse song, and their characteristic spectrotemporal patterns). Auditory representations were more stereotypic across trials and individuals in early stages of mechanosensory processing, and more variable at later stages. By recording neural activity in behaving flies, this study found that fly movements accounted for only a small fraction of the variance in neural activity, which suggests that across-trial auditory response variability stems from other sources. These results have important implications for how the brain processes auditory information to extract salient features and guide behavior (Pacheco, 2021).

Understanding of the Drosophila auditory circuit thus far has been built up from targeted studies of neural cell types that innervate particular brain regions close to the auditory periphery. Altogether, these studies have delineated a pathway that starts in the Johnston's organ and extends from the AMMC to the WED, the VLP and the LPC. By imaging pan-neuronally, widespread auditory responses that spanned brain regions beyond the canonical pathway, which suggests that auditory processing was found to be more distributed. However, for neuropil signals, it was challenging to determine the number of neurons that contribute to the described ROI responses. Although the diverse set of temporal and tuning types sampled per neuropil suggests that many neurons per neuropil, restricting GCaMP to spatially restricted genetic enhancer lines will assist with linking broad functional maps with the cell types constituting them (Pacheco, 2021).

Findings of widespread auditory activity are likely not unique to audition. So far, in adult Drosophila, only taste processing has been broadly surveyed. While that study did not map activity onto neuropils and tracts, nor did it make comparisons across individuals, it suggested that taste processing was distributed throughout the brain. Similarly, in vertebrates, widespread responses to visual and nociceptive stimuli have been observed throughout the brain. The findings are consistent with anatomical studies that found connections from the AMMC and the WED to several other brain regions, and with the analysis of the hemibrain connectome. While it is not yet known what role this widespread auditory activity plays in behavior, this study shows that ROIs that respond to auditory stimuli do so with mostly excitatory (depolarizing) responses and that activity throughout the brain is predominantly tuned to features of the courtship song. During courtship, flies evaluate multiple sensory cues (olfactory, auditory, gustatory and visual) to inform mating decisions and to modulate their mating drive. Although integration of multiple sensory modalities has been described in higher-order brain regions, the results suggest that song representations are integrated with olfactory and visual information at earlier stages. In addition, song information may modulate the processing of non-courtship stimuli. Song representations in the MB may be useful for learning associations between song and general olfactory, gustatory or visual cues, while diverse auditory activity throughout all regions of the LH may indicate an interaction between song processing and innate olfactory behaviors. Finally, auditory activity was found in brain regions involved in locomotion and navigation (the central and lateral complex, and the superior and ventromedial neuropils). Auditory activity in these regions is diverse, which suggests that pre-motor circuits receive information about courtship song patterns and could therefore underlie stimulus-specific locomotor responses (Pacheco, 2021).

D. melanogaster songs are composed of pulses and sines that differ in their spectral and temporal properties; however, it is unclear how and where selectivity for the different song modes arises in the brain. Since neurons in the LPC are tuned for pulse song across all time scales that define that mode of song, neurons upstream must carry the relevant information to generate such tuning. This study found many ROIs that are selective for either sine or pulse stimuli throughout the entire central brain. A more detailed systematic examination of tuning in the neuropils that carry the most auditory activity (the AMMC, the SAD, the WED, the AVLP and PVLP, the PLP and the LH) revealed that most ROIs are tuned to either pulse or sine features, with few ROIs possessing intermediate tuning. This suggests that sine and pulse information splits early in the pathway. It was also found that sine-selective responses dominate throughout the brain. Although some of this selectivity may simply reflect preference for continuous versus pulsatile stimuli, this investigation of feature tuning revealed that many of these ROIs preferred frequencies that are specifically present in courtship songs. Previous studies indicated that pulse song is more important for mating decisions, with sine song purported to play a role in only priming females. However the results, in combination with the fact that males spend a greater proportion of time in courtship singing sine versus pulse song, suggest a need for reevaluation of the importance and role of sine song in mating decisions. This study therefore lays the foundation for exploring how song selectivity arises in the brain (Pacheco, 2021).

The results also revealed that early mechanosensory brain areas contain ROIs with less variable auditory activity across trials and animals. The results for across-animal variability have parallels to the Drosophila olfactory pathway, whereby third-order MB neurons are not stereotyped, while presynaptic neurons in the AL, the PNs, are. Similarly, a lack of stereotypy beyond early mechanosensory brain areas may reflect stochasticity in synaptic wiring. The amount of variation observed in some brain areas was large, and follow-up experiments with sparser driver lines will be needed to validate whether what is reported here applies to variation across individual identifiable neurons (Pacheco, 2021).

A wide range of across-trial variability was observed throughout neuropils with auditory activity. Imaging from a subset of brain regions in behaving flies revealed that trial-to-trial variance in auditory responses is not explained by spontaneous movements, which suggests that variance is driven by internal dynamics. This result differs from recent findings in the mouse brain, which showed that a large fraction of activity in sensory cortices corresponds to non-task-related or spontaneous movements. This may indicate an important difference between invertebrate and vertebrate brains and the degree to which ongoing movements shape activity across different brains. However, it should be pointed out that while motor activity is known to affect sensory activity in flies, this modulation is tied to movements that are informative for either optomotor responses or steering. In these experiments, although flies walked abundantly, they did not produce reliable responses to auditory stimuli, although playback of the same auditory stimuli can reliably change walking speed in freely behaving flies. Adjusting the paradigm to drive such responses might uncover behavioral modulation of auditory activity. Alternatively, behavioral modulation of auditory responses may occur primarily in motor areas, such as the central complex, or areas containing projections of descending neurons. Further dissection of the sources of this variability would require the simultaneous capture of more brain activity in behaving animals while not significantly compromising spatial resolution (Pacheco, 2021).

This paper provides tools for characterizing sensory activity registered in common atlas coordinates for comparisons across trials, individuals and sexes. By producing maps for additional modalities and stimulus combinations and by combining these maps with information on connectivity between and within brain regions, the logic of how the brain represents the myriad stimuli and their combinations present in the world should emerge (Pacheco, 2021).

Several features of the adult nervous systems develop in a "critical period," (CP) during which high levels of plasticity allow neural circuits to be tuned for optimal performance. Through an analysis of long-term olfactory habituation (LTH) in female Drosophila, this study provides new insight into mechanisms by which CPs are regulated in vivo LTH manifests as a persistently reduced behavioural response to an odorant encountered for four continuous days and occurs together with the growth of specific, odorant-responsive glomeruli in the antennal lobe. The CP for behavioral and structural plasticity induced by ethyl butyrate (EB) or carbon dioxide (CO(2)) closes within 48 hours after eclosion. The elaboration of excitatory projection neuron (PN) processes likely contribute to glomerular volume increases: both occur together and similarly require cAMP signalling in the antennal lobe inhibitory local interneurons (iLNs). Further, the CP for structural plasticity could be extended beyond 48 hours if EB- or CO(2)-responsive olfactory sensory neurons (OSNs) are silenced after eclosion; thus, OSN activity is required for closing the CP. Strikingly, silencing of glomerulus-selective OSNs extends the CP for structural plasticity only in respective target glomeruli. This indicates existence of a local, short-range mechanism for regulating CP closure (Chodankar, 2020).

Many animals use acoustic signals to attract a potential mating partner. In fruit flies (Drosophila melanogaster), the courtship pulse song has a species-specific interpulse interval (IPI) that activates mating. Although a series of auditory neurons in the fly brain exhibit different tuning patterns to IPIs, it is unclear how the response of each neuron is tuned. This study examined the neural circuitry regulating the activity of antennal mechanosensory and motor center (AMMC)-B1 neurons, key secondary auditory neurons in the excitatory neural pathway that relay song information. By performing Ca(2+) imaging in female flies, it as found that the IPI selectivity observed in AMMC-B1 neurons differs from that of upstream auditory sensory neurons [Johnston's organ (JO)-B]. Selective knock-down of a GABAA receptor subunit in AMMC-B1 neurons increased their response to short IPIs, suggesting that GABA suppresses AMMC-B1 activity at these IPIs. Connection mapping identified two GABAergic local interneurons that synapse with AMMC-B1 and JO-B. Ca(2+) imaging combined with neuronal silencing revealed that these local interneurons, AMMC-LN and AMMC-B2, shape the response pattern of AMMC-B1 neurons at a 15 ms IPI. Neuronal silencing studies further suggested that both GABAergic local interneurons suppress the behavioral response to artificial pulse songs in flies, particularly those with a 15 ms IPI. Altogether, this study identified a circuit containing two GABAergic local interneurons that affects the temporal tuning of AMMC-B1 neurons in the song relay pathway and the behavioral response to the courtship song. These findings suggest that feedforward inhibitory pathways adjust the behavioral response to courtship pulse songs in female flies (Yamada, 2018).

Acoustic communication between insects serves as an excellent model system for analyzing the neuronal mechanisms underlying auditory information processing. To understand the central auditory pathways a large-scale analysis was performed of the interneurons associated with the primary Drosophila auditory center. By screening expression driver strains and performing single-cell labeling of these strains, 44 types of interneurons were identified innervating the primary auditory center - 5 types were local interneurons whereas the other 39 types were projection interneurons connecting the primary auditory center with other brain regions. The projection neurons comprised three frequency-selective pathways and two frequency-embracive pathways. Mapping of their connection targets revealed that five neuropils in the brain - the wedge, anterior ventrolateral protocerebrum, posterior ventrolateral protocerebrum (PVLP), saddle (SAD), and gnathal ganglia (GNG) - were intensively connected with the primary auditory center. In addition, several other neuropils, including visual and olfactory centers in the brain, were directly connected to the primary auditory center. The distribution patterns of the spines and boutons of the identified neurons suggest that auditory information is sent mainly from the primary auditory center to the PVLP, WED, SAD, GNG, and the thoracico-abdominal ganglia. Based on these findings, this study has established the first comprehensive map of secondary auditory interneurons, which indicates the downstream information flow to parallel ascending pathways, multimodal pathways, and descending pathways (Matsuo, 2016).

Animals perform many stereotyped movements, but how nervous systems are organized for controlling specific movements remains unclear. This study used anatomical, optogenetic, behavioral, and physiological techniques to identify a circuit in Drosophila melanogaster that can elicit stereotyped leg movements that groom the antennae. Mechanosensory chordotonal neurons detect displacements of the antennae and excite three different classes of functionally connected interneurons, which include two classes of brain interneurons and different parallel descending neurons. This multilayered circuit is organized such that neurons within each layer are sufficient to specifically elicit antennal grooming. However, differences were found in the durations of antennal grooming elicited by neurons in the different layers, suggesting that the circuit is organized to both command antennal grooming and control its duration. As similar features underlie stimulus-induced movements in other animals, there is a possibility of a common circuit organization for movement control that can be dissected in Drosophila (Hampel, 2015).

Sleep is essential, but animals may forgo sleep to engage in other critical behaviors, such as feeding and reproduction. Previous studies have shown that female flies exhibit decreased sleep after mating, but understanding of the process is limited. This study reports that postmating nighttime sleep loss is modulated by diet and sleep deprivation, demonstrating a complex interaction among sleep, reproduction, and diet. It was also found that female-specific pC1 neurons and sleep-promoting dorsal fan-shaped body (dFB) neurons are required for postmating sleep plasticity. Activating pC1 neurons leads to sleep suppression on standard fly culture media but has little sleep effect on sucrose-only food. Published connectome data suggest indirect, inhibitory connections among pC1 subtypes. Using calcium imaging, it was shown that activating the pC1e subtype inhibits dFB neurons. It is proposed that pC1 and dFB neurons integrate the mating status, food context, and sleep drive to modulate postmating sleep plasticity (Duhart, 2023).

Sleep is required to maintain physiological functions and is widely conserved across species. To understand the sleep-regulatory mechanisms, sleep-regulating genes and neuronal circuits are studied in various animal species. In the sleep-regulatory neuronal circuits in Drosophila melanogaster, the dorsal fan-shaped body (dFB) is a major sleep-promoting region. However, other sleep-regulating neuronal circuits were not well identified. It was recently found that arousal-promoting T1 dopamine neurons, interneurons of protocerebral bridge (PB) neurons, and PB neurons innervating the ventral part of the FB form a sleep-regulatory circuit, which was named "the PB-FB pathway". In the exploration of other sleep-regulatory circuits, it was found that activation of FB interneurons, also known as pontine neurons, promoted arousal. FB interneurons had possible connections with the PB-FB pathway and dFB neurons. Ca2+ imaging revealed that FB interneurons received excitatory signals from the PB-FB pathway. The possible role of FB interneurons to regulate dFB neurons was demonstrated. These results suggested the role of FB interneurons in sleep regulation (Kato, 2022).

This study reports a novel sleep-regulatory pathway that promotes arousal. This study first focused on FB interneurons and found that cholinergic FB interneurons promoted arousal. The arousal-promoting effect of FB interneurons was confirmed by using more specific drivers. These drivers label FB interneurons which receive input from P-FN neurons (that project from the protocerebral bridge to the ventral FB and the NO) and send output to dFB neurons. There should be other FB interneurons that do not have a connection with P-FN neurons or dFB neurons. It means that FB interneurons labeled by two split drivers are only a part of FB interneurons. Therefore we considered that the weaker effects were due to the smaller number of neurons labeled by split-Gal4 lines than NP2320, not the effect of neurons other than FB interneurons. No aclear sleep rebound was observed after neuronal activation. In the previous study, R52B10-Gal4 was used which is reported to drive sleep rebound in the female fly. A clear sleep rebound was observed in female flies but not in male flies. These results indicated that there is a sex difference in the regulation of sleep rebound, at least, in R52B10 neurons. The current study used only male flies and this could be one of the reasons why no clear sleep rebound was observed. It was next asked about the relationship between FB interneurons and known sleep-regulatory circuits. GRASP and Ca2+ experiments were performed, and FB interneurons and R52B10 neurons which label the output neurons of the PB-FB pathway were shown to be anatomically and functionally connected . Although the possibility of other pathways downstream to R52B10 cannot be excluded, this study demonstrated clearly that FB interneurons are one of the downstream to R52B10. Further study will show the impact size of the connection between them in sleep regulation. To investigate the postsynaptic partners of FB interneurons, a GRASP experiment was conducted. FB interneurons and dFB neurons labeled by R23E10 were found to form close associations. Besides, according to the connectome paper and connectome dataset, vDeltaB, C, D, and hDeltaC receive input from P-FN neurons and send output to FB tangential neurons which arborize in layers 6 and 7. This information also supported the idea that R52B10 neurons including P-FN neurons, FB interneurons like vDeltaB, C, D, and hDeltaC neurons, and dFB neurons that are consisted of FB tangential neurons which arborize in layers 6 and 7 form a neuronal circuit. Future study will clarify their functional connection and the role of this circuit in sleep regulation. Furthermore, a previous study showed that neurons that project to the ventral FB (vFB neurons) promote sleep and mediate consolidation of long-term memory. Since axon terminals and dendrites of FB interneurons arborize in both the dorsal and ventral FB, there would be interactions between dFB and vFB neurons via FB interneurons. Further research will clarify the functional relationship between these neurons (Kato, 2022).

According to previous reports, FB interneurons regulate optomotor behavior and express tachykinin, a neuropeptide that regulates aggression. Additionally, T1 dopamine neurons, which are upstream of R52B10 neurons, regulate aggression as well. Besides, P2 neurons, which include FB interneurons, regulate chronic isolation evoked sleep loss. Moreover, courtship-regulator P1 neurons activate T1 neurons and modulate sleep/courtship balance based on the nutritional status. Taking all the information mentioned above into account, it is considerd that arousal signals related to aggression, courtship, nutrition, and vision converge into the PB-FB pathway including FB interneurons to regulate arousal. Further studies should clarify the role of these arousal signals on the PB-FB pathway and FB interneurons in sleep regulation (Kato, 2022).

In conclusion, the results provided possible sleep-regulatory neurons that may connect with the PB-FB pathway and dFB neurons. It is hypothesized that arousal signals are sent from the PB-FB pathway to FB interneurons, inhibit dFB neurons via inhibitory signals, and regulate sleep (Kato, 2022).

Understanding how neural circuits underlie behaviour is challenging even in the connectome era because it requires a combination of anatomical and functional analyses. This is exemplified in the circuit underlying the light avoidance behaviour displayed by Drosophila melanogaster larvae. While this behaviour is robust and the nervous system relatively simple, the circuit is only partially delineated with some contradictions among studies. This study devised trans-Tango MkII, an offshoot of the transsynaptic circuit tracing tool trans-Tango, and implement it in anatomical tracing together with functional analysis. Neuronal inhibition was used to test necessity of particular neuronal types in light avoidance and selective neuronal activation to examine sufficiency in rescuing light avoidance deficiencies exhibited by photoreceptor mutants. These studies reveal a four-order circuit for light avoidance connecting the light-detecting photoreceptors with a pair of neuroendocrine cells via two types of clock neurons. This approach can be readily expanded to studying other circuits (Sorkac, 2022).

The study revealed a circuit consisting of four orders of neurons that connect the Rh5 photoreceptors to PTTH neurons via the 5th-LaN and DN2s. While this circuit mediates the response to bright light, the observation that DN1s are necessary for photophobic response only to low light intensity indicates the existence of an additional pathway for dim light. It is noteworthy that a third, independent system has been reported in which a gustatory receptor mediates photophobic response to high-intensity light in class IV multidendritic neurons. This study has no information as to at which level these pathways might meet, if at all, upstream of the motor neurons (Sorkac, 2022).

The results clarify several earlier studies regarding the role of Pdf-LaNs in light avoidance. In the current experiments, Pdf-LaNs are dispensable for light avoidance, yet their activation is attractive. A potential explanation is that Pdf-LaNs may modulate larval photophobia via inhibition, especially since adult Pdf-LaNs are glycinergic. In addition, the results contradict a previous study reporting that Pdf-LaNs are presynaptic to PTTH neurons. This study relied on a version of GRASP that is in fact not synaptic. Thus, the proposed connection could have been the result of a non-synaptic reconstitution of GFP due to proximity especially since reconstituted GFP was not observed using a synaptic version of GRASP. This result, however, does not rule out a Pdf-LaN-mediated inhibition of the light avoidance circuit from the Rh5 photoreceptors to PTTH neurons. It is conceivable that, alongside their roles in alternative circuits that mediate this behaviour, Pdf-LaNs play inhibitory roles in this circuit as well. Indeed, ablating Pdf-LaNs increases the activity in PTTH neurons as revealed by the GCaMP signal (Sorkac, 2022).

Our analysis of the robust light avoidance response in larvae exemplifies the importance of employing a comprehensive approach combining circuit tracing together with neuronal inhibition and activation to test necessity and sufficiency. Circuit epistasis analysis was made possible by trans-Tango MkII, a new version of trans-Tango that allows researchers to trace and manipulate neural circuits in Drosophila larvae. The combination of a robust and user-friendly genetic tool such as trans-Tango MkII with careful functional analysis constitutes a powerful approach that can be readily expanded to studying other circuits and behaviours (Sorkac, 2022).

Sleep is a universally conserved physiological state which contributes toward basic organismal functions, including cognitive operations such as learning and memory. Intriguingly, organisms can sometimes form memory even without sleep, such that Drosophila display sleep-dependent and sleep-independent memory in an olfactory appetitive training paradigm. Sleep-dependent memory can be elicited by the perception of sweet taste, and this study now shows that a mixed-sex population of flies maintained on sorbitol, a tasteless but nutritive substance, do not require sleep for memory consolidation. Consistent with this, silencing sugar-sensing gustatory receptor neurons in fed flies triggers a switch to sleep-independent memory consolidation, whereas activating sugar-sensing gustatory receptor neurons results in the formation of sleep-dependent memory in starved flies. Sleep-dependent and sleep-independent memory relies on distinct subsets of reward signaling protocerebral anterior medial dopaminergic neurons (PAM DANs) such that PAM-β'2mp DANs mediate memory in fed flies whereas PAM-α1 DANs are required in starved flies. Correspondingly, a feeding-dependent calcium increase was observed in PAM-β'2mp DANs, but not in PAM-α1 DANs. Following training, the presence of sweet sugars recruits PAM-β'2mp DANs, whereas tasteless medium increases calcium in PAM-α1 DANs. Together, this work identifies mechanistic underpinnings of sleep-dependent memory consolidation, in particular demonstrating a role for the processing of sweet taste reward signals (Chouhan, 2022).

Circadian clocks align various behaviors such as locomotor activity, sleep/wake, feeding, and mating to times of day that are most adaptive. How rhythmic information in pacemaker circuits is translated to neuronal outputs is not well understood. This study used brain-wide, 24-h in vivo calcium imaging in the Drosophila brain and searched for circadian rhythmic activity among identified clusters of dopaminergic (DA) and peptidergic neurosecretory (NS) neurons. Such rhythms were widespread and imposed by the PERIOD-dependent clock activity within the ∼150-cell circadian pacemaker network. The rhythms displayed either a morning (M), evening (E), or mid-day (MD) phase. Different subgroups of circadian pacemakers imposed neural activity rhythms onto different downstream non-clock neurons. Outputs from the canonical M and E pacemakers converged to regulate DA-PPM3 and DA-PAL neurons. E pacemakers regulate the evening-active DA-PPL1 neurons. In addition to these canonical M and E oscillators, evidence is for a third dedicated phase occurring at mid-day: the l-LNv pacemakers present the MD activity peak, and they regulate the MD-active DA-PPM1/2 neurons and three distinct NS cell types. Thus, the Drosophila circadian pacemaker network is a polyphasic rhythm generator. It presents dedicated M, E, and MD phases that are functionally transduced as neuronal outputs to organize diverse daily activity patterns in downstream circuits (Liang, 2023).

Drosophila's lateral posterior neurons (LPNs) belong to a small group of circadian clock neurons that is so far not characterized in detail. Thanks to a new highly specific split-Gal4 line, this study describes LPNs' morphology in fine detail, their synaptic connections, daily bimodal expression of neuropeptides, and a putative role of this cluster in controlling daily activity and sleep patterns is proposed. The three LPNs were found to be heterogeneous. Two of the neurons with similar morphology arborize in the superior medial and lateral protocerebrum and most likely promote sleep. One unique, possibly wakefulness-promoting, neuron with wider arborizations extends from the superior lateral protocerebrum toward the anterior optic tubercle. Both LPN types exhibit manifold connections with the other circadian clock neurons, especially with those that control the flies' morning and evening activity (M- and E-neurons, respectively). In addition, they form synaptic connections with neurons of the mushroom bodies, the fan-shaped body, and with many additional still unidentified neurons. Both LPN types rhythmically express three neuropeptides, Allostatin A, Allostatin C, and Diuretic Hormone 31 with maxima in the morning and the evening. The three LPN neuropeptides may, furthermore, signal to the insect hormonal center in the pars intercerebralis and contribute to rhythmic modulation of metabolism, feeding, and reproduction. These findings are discussed in the light of anatomical details gained by the recently published hemibrain of a single female fly on the electron microscopic level and of previous functional studies concerning the LPN.

Animal circadian rhythms persist in constant darkness and are driven by intracellular transcription-translation feedback loops. Although these cellular oscillators communicate, isolated mammalian cellular clocks continue to tick away in darkness without intercellular communication. To investigate these issues in Drosophila, behavior as well as molecular rhythms were assayed within individual brain clock neurons while blocking communication within the ca. 150 neuron clock network. CRISPR-mediated neuron-specific circadian clock knockouts were also generated. The results point to two key clock neuron groups: loss of the clock within both regions but neither one alone has a strong behavioral phenotype in darkness; communication between these regions also contributes to circadian period determination. Under these dark conditions, the clock within one region persists without network communication. The clock within the famous PDF-expressing s-LNv neurons however was strongly dependent on network communication, likely because clock gene expression within these vulnerable sLNvs depends on neuronal firing or light (Schlichting, 2019).

Neuronal networks make myriad contributions to behavior and physiology. By definition, individual neurons within a network interact, and different networks also interact to coordinate specialized functions. For example, the visual cortex and motor output centers must coordinate to react properly to environmental changes. In a less immediate fashion, sleep centers and circadian clocks are intertwined to properly orchestrate animal physiology. The brain clock is of special interest: it not only times and coordinates physiology within neuronal tissues but also sends signals to the body to keep the entire organism in sync with the cycling external environment (Schlichting, 2019).

The small, circumscribed Drosophila clock network is ideal to address circadian communication issues. The comparable region in mammals, the suprachiasmatic nucleus, is composed of thousands of cells depending on the species. There are in contrast only 75 clock neurons per hemisphere in Drosophila. These different clock neurons can be divided into several subgroups according to their location within the fly brain. There are 4 lateral and three dorsal neuron clusters, which have different functions in controlling fly physiology (Schlichting, 2019).

The four small ventro-lateral neurons (sLNvs) are arguably the most important of the 75 clock neurons. This is because ablating or silencing these neurons abolishes rhythms in constant darkness (DD). They reside in the accessory medulla region of the fly brain, an important pacemaker center in many insects, and express the neuropeptide PDF. In addition, they are essential for predicting dawn. A very recent study suggests that the sLNvs are also able to modulate the timing of the evening (E) peak of behavior via PDF. The other ventral-lateral group, the four large-ventro-lateral neurons (lLNvs), also express PDF and send projections to the medulla, the visual center of the fly brain; they are important arousal neurons. Consistent with the ablation experiments mentioned above, the absence of pdf function or reducing PDF levels via RNAi causes substantial arrhythmic behavior in DD (Schlichting, 2019).

Other important clock neurons include the dorso-lateral neurons (LNds), which are essential for the timing of the E peak and adjustment to long photoperiods. Two other clock neuron groups, the lateral-posterior neurons (LPN and a subset of the dorsal neurons (DN1s), were recently shown to connect the clock network to sleep centers in the fly central complex. The DN2 neurons are essential for temperature preference rhythms, whereas no function has so far been assigned to the DN3s (Schlichting, 2019).

Despite these distinct functions, individual clock neuron groups are well-connected to each other. At the anatomical level, all lateral neuron clusters and even DN1 dorsal neurons send some of their projections into the accessory medulla, where they can interact. A second area of common interaction is the dorsal brain; only the lLNvs do not project there (Schlichting, 2019).

Several studies have investigated interactions between different clock neurons. Artificially expressing kinases within specific clock neurons causes their clocks to run fast or slow and also changes the overall free-running period of the fly, indicating that network signaling adjusts behavior. Similarly, speeding up or slowing down individual neurons is able to differentially affect behavioral timing in standard light-dark (LD) cycles. A high level of neuronal plasticity within the network also exists: axons of individual cells undergo daily oscillations in their morphology, and neurons change their targets depending on the environmental condition (Schlichting, 2019).

How neuronal communication influences the fly core feedback loop is not well understood. The latter consists of several interlocked transcriptional-translational feedback loops, which probably underlie rhythms in behavior and physiology. A simplified version of the core feedback loop consists of the transcriptional activators Clock (CLK) and Cycle (CYC) and the transcriptional repressors Period (PER) and Timeless (TIM). CLK and CYC bind to E-boxes within the period (per) and timeless (tim) genes (among other clock-controlled genes) and activate their transcription. After PER and TIM synthesis in the cytoplasm, they form a heterodimer and enter the nucleus toward the end of the night. There they interact with CLK and CYC, release them from their E-box targets and thereby inhibit their own transcription. All 75 pairs of clock neurons contain this canonical circadian machinery, which undergoes daily oscillations in level. Indeed, the immunohistochemical cycling of PER and TIM within these neurons is a classic assay to visualize these molecular oscillations (Schlichting, 2019).

Silencing PDF neurons stops their PER cycling, indicating an important role of neuronal firing in maintaining circadian oscillations. However, only two time points were measured, and the results were possibly confounded by developmental effects. PDF neuron silencing also phase advances PER cycling in downstream neurons, suggesting that PDF normally serves to delay cycling in target neurons. This is consistent with experiments showing that PDF signaling stabilizes PER. In addition, neuronal activation is able to mimic a light pulse and phase shift the clock due to firing-mediated TIM degradation (Schlichting, 2019).

To investigate more general features of clock neuron interactions on the circadian machinery, the majority of the fly brain clock neurons were silenced, and behavior and clock protein cycling within the circadian network was examined in a standard light-dark cycle (LD) as well as in constant darkness (DD). Silencing abolished rhythmic behavior but had no effect on clock protein cycling in LD, indicating that the silencing affects circadian output but not oscillator function in a cycling light environment. Silencing similarly abolished rhythmic behavior in DD but with very different effects on clock protein cycling. Although protein cycling in the LNds was not affected by neuronal silencing in DD, the sLNvs dampened almost immediately. Interestingly, this differential effect is under transcriptional control, suggesting that some Drosophila clock neurons experience activity-regulated clock gene transcription. Cell-specific CRISPR/Cas9 knockouts of the core clock protein Per further suggests that network properties are critical to maintain wild-type activity-rest rhythms. The data taken together show that clock neuron communication and firing-mediated clock gene transcription are essential for high amplitude and synchronized molecular rhythms as well as rhythmic physiology (Schlichting, 2019).

The central clock of animals is essential for dictating the myriad diurnal changes in physiology and behavior. Knocking out core clock components such as period or Clock severely disrupts circadian behavior as well as molecular clock properties in flies and mammals. This study show that similar behavioral effects occur when the central clock neurons are silenced and thereby abolish communication within this network and with downstream targets, that is fly behavior becomes arrhythmic in LD as well as DD conditions and resembles the phenotypes of core clock mutant strains (Schlichting, 2019).

Despite the loss of all rhythmic behavior, silencing did not impact the molecular machinery in LD conditions: PER and PDP1 protein cycling was normal. These findings suggest that (1) rhythmic behavior requires clock neuron output, which is uncoupled from the circadian molecular machinery by network silencing, and (2) synchronized molecular rhythms of clock neurons do not require neuronal activity. These findings are in agreement with previous work showing that silencing the PDF neurons had no effect on Per cycling within these neurons. The results presumably reflect the strong effect of the external light-dark cycle on these oscillators (Schlichting, 2019).

In DD however, the individual neurons change dramatically: the different neurons desynchronize, and their protein cycling damps to different extents. Interestingly, sLNv cycling relies most strongly on neuronal communication: these neurons cycle robustly in controls but apparently not at all in the silenced state. sLNvs were previously shown to be essential for DD rhythms. Unfortunately, the sensitivity of immunohistochemistry precludes determining whether the molecular clock has actually stopped or whether silencing has only (dramatically) reduced cycling amplitude. However, a simple interpretation of the adult-specific silencing experiment favors a stopped clock: decreasing the temperature to 18 degrees after a week at high temperature failed to rescue rhythmic behavior. A similar experiment in mammals gave rise to the opposite result, suggesting an effect of firing on circadian amplitude in that case. However, different explanations cannot be excluded, for example chronic effects of neuronal silencing or a too large phase difference between the different neuronal subgroups to reverse after a week without communication (Schlichting, 2019).

In either case, a stopped clock or an effect on clock protein oscillation amplitude, these results make another link to the mammalian literature: modeling of the clock network suggests that different neurons resynchronize more easily if the most highly-connected cells are intrinsically weak oscillators. The sLNvs are essential for DD rhythms, known to communicate with other clock neurons and are situated in the accessory medulla; this is an area of extensive neuronal interactions in many insects. These considerations rationalize weak sLNv oscillators (Schlichting, 2019).

An important role of interneuron communication in DD is in agreement with previous work showing that altering the speed of individual neuron groups can change the phase of downstream target neurons. An important signaling molecule is the neuropeptide PDF: its absence changes the phase of downstream target neurons, and silencing PDF neurons causes an essentially identical phenotype to the lack of PDF. However, the effects reported in this study are much stronger and show different levels of autonomy than PDF ablation, suggesting that other signaling molecules and/or the neuronal activity of additional clock neurons are essential to maintain proper rhythmic clock protein expression (Schlichting, 2019).

To address these possibilities, this study took two approaches. First, clock gene RNA levels were investigated after silencing. The goal was to assess whether the damping of silenced neurons is under gene expression control, likely transcriptional control. Indeed, tim mRNA profiles nicely reproduced the protein cycling profiles: robust cycling of all (assayed) clock neurons was maintained in LD even with silencing, but tim-mRNA levels in the sLNvs stopped cycling in DD; in contrast, robust cycling was maintained in the LNds. This suggests that the changes in protein cycling amplitude and also possibly phase are under transcriptional control. Importantly, the tim signal in the sLNvs disappeared upon silencing, suggesting that neuronal activity promotes clock gene transcription at least in this subset of neurons. This recapitulates for the first time in Drosophila the robust positive relationship between neuronal firing and clock gene transcription in mammals. To date, Drosophila neuronal firing had only been connected to post-transcriptional clock protein regulation, namely TIM degradation. Conceivably, these two effects are connected: TIM degradation might be required to relieve transcriptional repression and maintain cycling (Schlichting, 2019).

The second approach was a cell-specific knockout strategy, applied to the clock neuron network. Three guides were generated targeting the CDS of per and also CAS9 was expressed in a cell-specific manner. The guides caused double strand breaks in the per gene, which in turn led to cell-specific per mutations. This adult brain knockout strategy worked reliably and specifically, in glial cells as well as neurons, with high efficiency and with no apparent background effects. This strategy was successfully used to knock out most if not all Drosophila GPCRs and it is believed to be superior to RNAi for most purposes. Importantly, expression of the guides with the clk856-GAL4 driver phenocopied per⁰¹ behavior. To focus on individual clock neurons, cell-specific knockouts were generated in different clock neurons. PERKO in the PDF cells did not increase the level of arrhythmicity, only a PERKO in most lateral neurons, E cells as well as PDF cells, generated high levels of arrhythmic behavior (Schlichting, 2019).

Many animals exhibit morning and evening peaks of locomotor behavior. In Drosophila, two corresponding circadian neural oscillators-M (morning) cells and E (evening) cells-exhibit a corresponding morning or evening neural activity peak. Yet little is known of the neural circuitry by which distinct circadian oscillators produce specific outputs to precisely control behavioral episodes. This study shows that ring neurons of the ellipsoid body (EB-RNs) display spontaneous morning and evening neural activity peaks in vivo: these peaks coincide with the bouts of locomotor activity and result from independent activation by M and E pacemakers. Further, M and E cells regulate EB-RNs via identified PPM3 dopaminergic neurons, which project to the EB and are normally co-active with EB-RNs. These in vivo findings establish the fundamental elements of a circadian neuronal output pathway: distinct circadian oscillators independently drive a common pre-motor center through the agency of specific dopaminergic interneurons (Liang, 2019).

Light is one of the most important factors regulating rhythmical behavior of Drosophila melanogaster. It is received by different photoreceptors and entrains the circadian clock, which controls sleep. The retina is known to be essential for light perception, as it is composed of specialized light-sensitive cells which transmit signal to deeper parts of the brain. This study examined the role of specific photoreceptor types and peripheral oscillators located in these cells in the regulation of sleep pattern. Sleep was shown to be controlled by the visual system in a very complex way. Photoreceptors expressing Rh1, Rh3 are involved in night-time sleep regulation, while cells expressing Rh5 and Rh6 affect sleep both during the day and night. Moreover, Hofbauer-Buchner (HB) eyelets which can directly contact with s-LN (v) s and l-LN (v) s play a wake-promoting function during the day. In addition, this study showed that L2 interneurons, which receive signal from R1-6, form direct synaptic contacts with l-LN (v) s, which provides new light input to the clock network (Damulewicz, 2020).

Tetrad synapses are formed between the retina photoreceptor terminals and postsynaptic cells in the first optic neuropil (lamina) of Drosophila. They are remodelled in the course of the day and show distinct functional changes during activity and sleep. These changes result from fast degradation of the presynaptic scaffolding protein Bruchpilot (BRP) by Cryptochrome (CRY) in the morning and depend on BRP-170, one of two BRP isoforms. This process also affects the number of synaptic vesicles, both clear and dense-core, delivered to the presynaptic elements. In cry⁰¹ mutants lacking CRY and in brp^Δ170, the number of synaptic vesicles is lower in the morning peak of activity than during night-sleep while in wild-type flies the number of synaptic vesicles is similar at these two time points. CRY may also set phase of the circadian rhythm in plasticity of synapses. The process of synapse remodelling stimulates the formation of clear synaptic vesicles in the morning. They carry histamine, a neurotransmitter in tetrad synapses and seem to be formed from glial capitate projections inside the photoreceptor terminals. In turn dense-core vesicles probably carry synaptic proteins building the tetrad presynaptic element (Damulewicz, 2020).

The results confirmed earlier studies that the presynaptic element (T-bar) of tetrad synapses is remodelled during the day and night and this rhythm is regulated by light and circadian clock. In the present study, it was found additionally that cyclic changes occur in the T-bar ultrastructure and its volume. Ultrastructural changes in T-bars and synaptic vesicles were possible because of using high resolution electron microscope tomography (EMT). In the present study it was also found that the number of synaptic vesicles cycles during the day, but this rhythm is masked by light. This result indicates that in the case of tetrad synapses, which are sites of fast neurotransmission between photoreceptors and the first order interneurons, intense neurotransmission occurs during the morning peak of locomotor activity, when more synaptic vesicles are attached to the T-bar platform than during sleep (ZT16) and are transported from capitate projections located next to the T-bars. At ZT16, tetrad synapses are ready for neurotransmission, since the total number of synaptic vesicles near tetrad synapses is similar at ZT1 and ZT16, but it is in a standby mode with lower frequency of synapses and vesicles, which are not attached to the T-bar platform and are not delivered from glial cells, respectively. However, transmission can be activated and is efficient because synapses during sleep have larger volume, and synaptic vesicles can be transported to the T-bar platform, if necessary, in response to an unexpected light pulse. Cyclic remodelling of tetrad presynaptic sites depends on BRP, which must be delivered to T-bars and degraded after light exposure after binding to CRY. This fast remodelling of synapses affects the number of vesicles transported to the presynaptic element (Damulewicz, 2020).

The present study was carried out only in light/dark (LD 12:12) conditions because in earlier studies it was found that the rhythm of the changes in BRP abundance in tetrad synapses of D. melanogaster is circadian. The rhythm is maintained in constant darkness (DD) and abolished in the per⁰¹ clock null mutant. On the basis of these results, it is assumed that rhythmic changes during the 24 h cycle reported is this paper are also circadian; however, in DD, the morning peak in BRP is not present because it depends on light (Damulewicz, 2020).

Electron microscope tomography (EMT) used in this study showed that synaptic vesicles are attached to the platform of the T-bar with filamentous proteins that are reduced in brp^Δ170 and brp^Δ190 mutants, which have fewer synaptic vesicles compared with wild-type flies. Two types of vesicles, clear and dense-core were detected. Dense-core synaptic vesicles were less numerous than clear ones. It is known that synaptic vesicles of tetrad synapses contain histamine, while the content of dense-core vesicles is unknown. It is possible that they carry presynaptic proteins to the T-bar. The comparison of ultrastructure of tetrad synapses in the morning peak of activity (ZT1) and during sleep (ZT16) indicates that more vesicles are attached to the T-bar platform and to capitate projections at ZT1, but this pattern is not present in brpΔ170, brp^Δ190 and cry⁰¹ . This result confirmed an earlier study that both high motor activity in the morning and light exposure increase activity of the visual system. In the morning, there is intense transport of synaptic vesicles to T-bars and delivery of histamine in clear vesicles from the epithelial glial cells through capitate projections. The evidence for a role of capitate projections in neurotransmitter recycling has already been reported and now this study showed, using EMT, that vesicles are produced from capitate projections and directly delivered to T-bars. This intense transport is damaged in all mutants studied, suggesting that both BRP isoforms and CRY are needed for this process. In addition, in the brp^Δ190 mutant lacking BRP-190, the T-bar structure is less dense than in the other strains studied. In a previous study, it was found that there are approximately 50% fewer synapses in brp^Δ190 than in Canton S and brp^Δ170. Another study reported that BRP isoforms are important for the formation of T-bars in neuromuscular junctions, and in brp mutants T-bars are smaller than in controls. T-bar height was reduced in brp^Δ190, whereas the widths of pedestal and platform were reduced in both mutants. They also decrease transmission since the active zone was smaller in both mutants and the number of synaptic vesicles was reduced. These ultrastructural changes are correlated with cell physiology since the amplitude of evoked excitatory junctional current was decreased in both mutants with a stronger effect in brp^Δ190 (Damulewicz, 2020).

The obtained reconstructions of tetrad T-bars from TEM serial sections of the lamina showed that although there were fewer synapses during the night (ZT16), the volume of the T-bar was larger at that time than in the morning (ZT1), while the total number of synaptic vesicles was similar. In contrast, a day/night difference (ZT1 vs. ZT16) in the number of vesicles was observed in brp^Δ170 and cry⁰¹ . This suggests that the CRY protein and BRP-170 are responsible for an increase in the number of synaptic vesicles during the morning peak of activity. Since CRY co-localizes with BRP and is involved in BRP degradation in the morning, CRY is probably also involved in the degradation of other proteins of synaptic vesicle organization in the morning since the number of vesicles in cry⁰¹ was low in the morning but high at night (ZT16). It seems that in cry⁰¹, synaptic vesicles are not delivered to the photoreceptor terminals from capitate projections and tethered to the cytomatrix in the morning. It is also interesting that the daily rhythm in the number of vesicles is not maintained in BRP mutants, which confirms an earlier study, and the BRP N-terminus, which lacks brp^Δ190, is necessary to maintain daily remodelling of the T-bar structure. Although both isoforms participate in building the cytomatrix their functions seem to be different in the course of the day. It is also possible that CRY is not only responsible for degradation of synaptic proteins but also as a protein, what is known, affecting the clock. In another cell types, in clock neurons l-LNvs, CRY, except interaction with TIM, is responsible for blue light response and firing of the l-LNvs. In the lamina it was found that in cry⁰¹ mutant the daily rhythm in synapse number and their remodelling was delayed in phase and the day/night difference in Canton S increased when peaks in the number of synapses were shift to ZT4 and ZT16. In result the difference between ZT1 and ZT16 was increased in cry⁰¹ (Damulewicz, 2020).

BRP is also responsible for rapid remodelling of the presynaptic active zone (AZ), and as reported in Drosophila NMJ, presynaptic homeostatic potentiation increases the number of BRP molecules and other AZ proteins, Unc13A and RBP, within minutes (Damulewicz, 2020).

When synaptic vesicles were counted at two different distances from the T-bar, to 200 nm and above 200 nm, there were differences between clear vesicles containing histamine and dense-core ones located in both areas. More clear vesicles were located near the T-bar and fewer above 200 nm. In the case of dense-core vesicles, their number was similar in both areas. This difference was not so striking in mutants in the case of vesicles located next to the platform, but in brp^Δ170 and cry⁰¹, there were more dense-core vesicles at ZT16 in the distance above 200 nm than closer to the T-bar. This result indicates that BRP-170 and CRY are important for the distribution of clear synaptic vesicles next to the T-bar as well as dense-core vesicles located above 200 nm from the presynaptic element. It is possible that dense-core vesicles contain T-bar proteins, probably BRP. When transport along the actin cytoskeleton is disrupted, the number of tetrad synapses decreases, and rapid AZ remodelling also fails (Damulewicz, 2020).

The above mentioned ultrastructural changes depend on the level of the presynaptic scaffolding protein BRP, which changes in abundance during the day and night. These changes are controlled by the daily expression of CRY, which seems to have many functions in photoreceptors in addition to being the circadian clock photoreceptor. In an earlier study, it was found that CRY interacts with BRP but only during light exposure and leads to the degradation of BRP during the day/light phase of the 24 h cycle. This seems to be responsible for the decrease in BRP level in the middle of the day after its peak at the beginning of the day. The lack of CRY in cry⁰¹ mutants changes the pattern of the tetrad presynaptic profile frequency during the day and the size of the T-bar. However, the rhythm is not completely abolished, which indicates that other proteins are also involved in the daily remodelling of tetrad synapses. Since CRY plays several functions in photoreceptors, changes in the number of tetrad synapse and T-bar size in cry⁰¹ may result from different processes and lack of interactions of CRY with TIM and BRP. CRY is a component of the molecular clock and interacts with TIM, and this may affect daily changes in the number and size of T-bars. In addition, light-activated CRY binds BRP and targets it to degradation. Previous work showed that flies with constitutively active CRY have low BRP level. In turn, cry⁰¹ mutants show changes in the pattern of BRP expression, with higher BRP level during the day (ZT4), at the time when wild-type flies have minimum of BRP expression. The pattern of BRP expression is similar to the pattern of daily changes in tetrad synapse number, so it is possible that the number of synapses or T-bar size is directly dependent on the amount of BRP which oscillates during the day. However, CRY in the retina photoreceptors binds also actin and is involved in the organization of phototransduction cascade of proteins in rhabdomeres38. This may be also involved in the regulation of T-bar structure. The differences in T-bar size of cry⁰¹ are shown at time when in Canton S CRY is active (during the day) or its level increases (ZT16). At the beginning of the night the level of CRY is very low, so there was no effect on T-bar structure and no difference between CS and cry⁰¹ was observed (Damulewicz, 2020).

The epithelial glial cells are important for many processes during phototransduction and in recycling neurotransmitters and other compounds during the night. Glia take up histamine and metabolize it to carcinin, which is next delivered to the photoreceptor terminals, and capitate projections are involved in this process. Activity of glial cells is also controlled by the circadian clock. During the night, glial cells seem to be more active than neurons to recycle neurotransmitters, and many proteins, including proteins of ion pumps, are found at higher concentrations at that time. The high number of synaptic vesicles near the tetrad T-bar during the morning peak of activity in Drosophila seems to depend on capitate projections invaginating from the epithelial glia to the photoreceptor terminals in the lamina of Drosophila (Damulewicz, 2020).

Although the presynaptic cytomatrix can be rapidly remodelled with transmission strength, it is also affected by motor and visual system activity, external factors, such as light in the case of the visual system, and the circadian clock, showing plasticity and correlation to changes in behaviour during the day/night cycle. As was shown in the present study synaptic plasticity and synapse remodelling during the day is a complex process which involves presynaptic proteins of the T-bar as well as two types of synaptic vesicles, clear and dense-core, and glial cells. It was also found that fast degradation of proteins involved in transmission is as important as pre- and postsynaptic protein synthesis (Damulewicz, 2020).

The central complex is one of the major brain regions that control sleep in Drosophila. However, the circuitry details of sleep regulation have not been elucidated yet. This study shows a novel sleep-regulating neuronal circuit in the protocerebral bridge (PB) of the central complex. Activation of the PB interneurons labeled by the R59E08-Gal4 and the PB columnar neurons with R52B10-Gal4 promoted sleep and wakefulness, respectively. A targeted GFP reconstitution across synaptic partners (t-GRASP) analysis demonstrated synaptic contact between these two groups of sleep-regulating PB neurons. Furthermore, it was found that activation of a pair of dopaminergic (DA) neurons projecting to the PB (T1 DA neurons) decreased sleep. The wake-promoting T1 DA neurons and the sleep-promoting PB interneurons formed close associations. Dopamine 2-like receptor (Dop2R) knockdown in the sleep-promoting PB interneurons increased sleep. These results indicated that the neuronal circuit in the PB, regulated by dopamine signaling, mediates sleep-wakefulness (Tomita, 2021).

Mechanosensory neurons located across the body surface respond to tactile stimuli and elicit diverse behavioral responses, from relatively simple stimulus location-aimed movements to complex movement sequences. How mechanosensory neurons and their postsynaptic circuits influence such diverse behaviors remains unclear. Previous work has shown that Drosophila perform a body location-prioritized grooming sequence when mechanosensory neurons at different locations on the head and body are simultaneously stimulated by dust. This study identified nearly all mechanosensory neurons on the Drosophila head that individually elicit aimed grooming of specific head locations, while collectively eliciting a whole head grooming sequence. Different tracing methods were used to reconstruct the projections of these neurons from different locations on the head to their distinct arborizations in the brain. This provides the first synaptic resolution somatotopic map of a head, and defines the parallel-projecting mechanosensory pathways that elicit head grooming (Eichler, 2023).

In many behaviors such walking and swimming, animals need to coordinate their left and right limbs. In Drosophila, wing grooming can be induced by activation of sensory organs called campaniform sensilla. Flies usually clean one wing at a time, coordinating their left and right hind legs to sweep the dorsal and ventral surfaces of the wing. This study has identified a pair of interneurons located in the ventral nerve cord that was named wing projection neurons 1 (wPN1) whose optogenetic activation induces wing grooming. Inhibition of wPN1 activity reduces wing grooming. They receive synaptic input from ipsilateral wing campaniform sensilla and wing mechanosensory bristle neurons, and they extend axonal arbors to the hind leg neuropils. Although they project contralaterally, their activation induces ipsilateral wing grooming. Anatomical and behavioral data support a role for wPN1 as command neurons coordinating both hind legs to work together to clean the stimulated wing (Zhang, 2022).

This study has identified wPN1 as potential command neurons for wing grooming. Although the criteria for command neurons is debated, neurons whose activation can induce a coherent motor program and whose function is required for that behavior to occur normally, such as wPN1, provide important entry points to map the neural circuits governing that behavior. The characterization of wPN1 reveals additional complexities in the control of grooming. Wing scratch remains unaffected, indicating additional sensory-to-motor pathways between wings and legs. Other left-right coordination circuits must also exist, since inhibition of wPN1 disrupts wing grooming but leaves hind-leg rubbing largely normal. wPN1 provides an entry point to identify downstream commissural and pre-motor neurons that coordinate left and right leg movements (Zhang, 2022).

Wing cleaning normally occurs late in the hierarchical sequence of grooming subroutines, and wPN1 may provide insights into when it is selected. Perhaps sensory information from wing campaniform sensilla and mechanosensory bristles integrate or accumulate to a threshold sufficient to activate wPN1, or perhaps wPN1 is transiently inhibited by circuits driving other grooming behaviors. Identification of wPN1 makes these testable hypotheses with the potential to reveal a general control mechanism for sequential motor behaviors (Zhang, 2022).

These behavior and anatomy data provide strong evidence that wPN1 plays an essential, command-like role in coordinating left and right back legs, but higher-resolution behavioral analysis and complete neuronal circuit mapping downstream of wPN1 are necessary to confirm this function (Zhang, 2022).

Sweet and bitter compounds excite different sensory cells and drive opposing behaviors. However, it remains unclear how sweet and bitter tastes are represented by the neural circuits linking sensation to behavior. To investigate this question in Drosophila, this study devised trans-Tango(activity), a strategy for calcium imaging of second-order gustatory projection neurons based on trans-Tango, a genetic transsynaptic tracing technique. Spatial overlap was found between the projection neuron populations activated by sweet and bitter tastants. The spatial representation of bitter tastants in the projection neurons was consistent, while that of sweet tastants was heterogeneous. Furthermore, it was discovered that bitter tastants evoke responses in the gustatory receptor neurons and projection neurons upon both stimulus onset and offset and that bitter offset and sweet onset excite overlapping second-order projections. These findings demonstrate an unexpected complexity in the representation of sweet and bitter tastants by second-order neurons of the gustatory circuit (Snell, 2022).

A central question in taste coding is how sweet and bitter tastes are represented beyond the first-order sensory neurons of the circuit. This study is the first to broadly characterize the taste responses of second-order projection neurons in Drosophila, a strategy enabled by trans-Tango(activity), a novel configuration of trans-Tango. Evidence was found for a variety of GPN response profiles, including some tuned specifically to either sweet or bitter tastants, and some that respond to both. It was also discovered that bitter tastants can evoke activity in GRNs and GPNs upon both stimulus onset and offset. The spatial distribution of the GPN offset response extends beyond the onset-responsive region and overlaps with regions responding to sweet tastants. These results thus demonstrate considerable complexity in the taste responses of GPNs (Snell, 2022).

Beyond these findings on taste representation, this study demonstrates more generally that trans-Tango can fill a niche in analyzing neural circuit function. Although a plethora of genetic tools exist for probing circuit function in Drosophila, they are limited by genetic access (e.g., the expression patterns of Gal4 driver lines). There is an obvious use in having a line that labels a broad population of neurons at a defined stage of the circuit; for example, the GH146-Gal4 driver has been extensively used because it labels a broad set of second-order olfactory projection neurons with little background expression (Snell, 2022).

In many other circuits, though, no comparable driver exists. Using trans-Tango initiated from a sparse set of starter neurons, one can selectively image the calcium responses of their downstream projection neurons, as has been done in this study in the gustatory system. The increasing availability of highly specific split-Gal4 drivers will enable this strategy in a wide variety of systems (Snell, 2022).

Moreover, this study demonstrates that using a nuclear-localized reporter as an output of trans-Tango can facilitate the counting of postsynaptic neurons. This strategy can be used to assess the degree of overlap between the second-order neurons of two different circuits, as has been done in this study for the sweet and bitter circuits (Snell, 2022).

A caveat of this analysis method is that the regions of interest (ROIs) obtained through the segmentation algorithm do not necessarily have a one-to-one correspondence with underlying neurons. As a result, it was not possible to definitively conclude whether a given ROI response profile matches the tuning of an individual gustatory projection neurons (GPNs). It is thus possible that ROIs exhibiting responses to multiple tastants result from the close proximity of multiple narrowly tuned neurons. Although proximity is unlikely to fully explain these findings, determining the extent to which ROIs reflect multiple cell types will require identifying cell-type-specific markers for different subclasses of GPNs (Snell, 2022).

Furthermore, this caveat implies that the proportion of ROIs responding to a given stimulus does not necessarily correspond to the proportion of GPNs responding to that stimulus. Thus, it is not possible to draw a strong conclusion about the relative number of GPNs exhibiting each observed response profile. Although ROIs responsive to both bitter and sweet tastants are described, the underlying GPNs with this response profile may be few in number. However, even one such neuron type would be notable, as no GRNs have been reported to encode both sweet and bitter taste (Snell, 2022).

The range of tastant responses observed suggests that second-order gustatory neurons may exhibit more diverse tuning profiles than first-order neurons. For example, some GPNs appear to encode both bitter taste and either water or mechanosensory features of the stimulus. Although weak calcium responses to water have been observed in bitter GRNs, responses of similar strength to bitter and water have been observed in many GPN ROIs. This suggests that GPNs may either nonlinearly transform the bitter GRN signal or perhaps integrate sensory inputs across different taste qualities and even across different sensory modalities. Additionally, the relative magnitudes of the bitter onset and offset responses varied across regions, with some responding similarly to these features and others responding far more strongly to offset. A similar dichotomy has not been observed in GRNs. Thus, these two types of GPN bitter responses likely arise from different circuit-level mechanisms. Furthermore, contrary to observed responses in GRNs, more homogeneity has been observed in GPN responses to bitter tastants than to sweet ones. These results suggest that substantial signal transformation occurs in the transition from the first to the second order of the taste circuit. (Snell, 2022).

The discovery of a bitter offset response is supported by an additional report of this phenomenon, along with another report of an offset response of sweet GRNs to lactic acid. The fact that the fly actively encodes onset and offset of some tastants raises questions about the relevance of these responses for the animal. Distinct onset and offset responses have been observed in other sensory modalities, such as vision. In vision, for example, stimulus offset may serve as a cue for the motion of an object. Bitter offset might similarly signal a change in the fly's gustatory environment. In the wild, flies can encounter a mixture of noxious and nutritious substances at a single feeding site: Drosophila feed on microbes that grow on decaying plant matter, and some of these microbes metabolize plant-derived compounds that are normally toxic to flies, thereby making them safe for the fly's consumption (Snell, 2022).

As many plant toxins are bitter, a fly sampling such a food source would encounter a series of bitter and non-bitter patches; a bitter offset response could thus signal movement off of a toxic substrate and onto a nutritive microbial one (Snell, 2022).

The data suggests that different bitter-responsive GPNs exhibit a different balance of onset- and offset-driven activity. Separate neural channels differentially encoding bitter onset and offset may endow the fly with greater sensitivity to changes in bitter concentration. In primate vision, separate ON and OFF pathways have been proposed to enable metabolically efficient coding of intensity changes to enhance contrast sensitivity and increase the dynamic range of the system (Snell, 2022).

Likewise, the cockroach has separate olfactory sensory neurons tuned to odor onset and offset, which increase the animal's ability to detect small concentration changes. The bitter offset pathway in Drosophila might perform a similar function, enabling the fly to efficiently navigate concentration gradients of bitter compounds on heterogeneous food sources. Finally, as this parallel ON/OFF pathway circuit motif has been observed in several sensory systems, including vision, olfaction, touch, and thermosensation, it may have general utility in sensory coding (Snell, 2022).

An alternative explanation is that bitter onset and offset may have opposing hedonic valences to the fly. The cessation of punishment is known to act as a reward in conditioning experiments (Snell, 2022).

Because bitter taste is aversive, bitter offset may thus be interpreted as rewarding. Separating these components of the bitter taste response may allow for differential input to the dopaminergic neurons of the mushroom body that mediate aversive and appetitive conditioning. The intriguing observation that the bitter offset response overlaps with the sweet response in lateral regions is consistent with this idea. If there is a GPN that underlies both the bitter offset and sweet responses in this area, this neuron may functionally encode the valence of the stimulus. Yet, even if these responses to bitter offset and sweet are caused by separate GPNs, their overlapping projections in this region may indicate that these GPNs converge onto common third-order neurons in the circuit. The results suggest that some GPNs integrate sweet and bitter taste information. Such integration may indicate that a population coding model is operative in the gustatory system of the fly. However, determining the coding scheme implemented by the gustatory system will require further investigation, including a determination of how that code is read out by downstream regions. It remains to be seen how downstream circuits use the taste information encoded by GPNs to sculpt the behavior of the fly (Snell, 2022).

Neural circuit assembly features simultaneous targeting of numerous neuronal processes from constituent neuron types, yet the dynamics is poorly understood. This study used the Drosophila olfactory circuit to investigate dynamic cellular processes by which olfactory receptor neurons (ORNs) target axons precisely to specific glomeruli in the ipsi- and contralateral antennal lobes. Time-lapse imaging of individual axons from 30 ORN types revealed a rich diversity in extension speed, innervation timing, and ipsilateral branch locations and identified that ipsilateral targeting occurs via stabilization of transient interstitial branches. Fast imaging using adaptive optics-corrected lattice light-sheet microscopy showed that upon approaching target, many ORN types exhibiting 'exploring branches' consisted of parallel microtubule-based terminal branches emanating from an F-actin-rich hub. Antennal nerve ablations uncovered essential roles for bilateral axons in contralateral target selection and for ORN axons to facilitate dendritic refinement of postsynaptic partner neurons. Altogether, these observations provide cellular bases for wiring specificity establishment (Li, 2021).

The Drosophila olfactory system has served as a model for investigating the mechanisms of neural circuit assembly. Axons of 50 types of olfactory receptor neurons (ORNs) and dendrites of 50 types of projection neurons (PNs) form 1-to-1 connections in 50 discrete, stereotyped, and individually identifiable glomeruli in the antennal lobe to relay olfactory information from the periphery to the brain. During the assembly of the adult olfactory system, PNs first extend dendrites to establish a coarse map. ORN axons then choose the dorsolateral or ventromedial trajectory to circumnavigate the antennal lobe, cross the midline, and invade the ipsi- and contralateral antennal lobes to find their synaptic partners. A multitude of cellular and molecular mechanisms have been identified that direct dendrite and axon targeting of selected PN and ORN types; some mechanisms first discovered in the fly olfactory system were subsequently found to be conserved in wiring the mammalian brain. Still, these studies are far from understanding the developmental algorithms that orchestrate the precise wiring of 50 pairs of ORN and PN types. Time-lapse imaging can provide cellular context in which wiring molecules exert their functions. Moreover, the ease of identifying neuron types based on their glomerular targets can determine the degree to which different neuron types use the same wiring rules (Li, 2021).

In this study, based on a previous protocol for studying the fly visual system in explant cultures, this study developed an antennae-brain explant preparation that recapitulates the precision by which the olfactory circuit is assembled in vivo. Time-lapse imaging of 30 ORN types revealed heterogenous axon targeting behaviors that contribute to the eventual wiring specificity. High-resolution adaptive optical lattice light-sheet microscopy (AO-LLSM) enabled discovery of an axon terminal structure prior to ORN axons reaching targets. This study also found that cytoskeletal organization of ORN axon terminals differs substantially from that of the classic growth cones from neurons in primary culture. Finally, ORN axon ablation uncovered essential roles for bilateral ORN axons in contralateral target selection, and for ORN axons to facilitate dendritic refinement of PNs (Li, 2021).

Prior to this study, it was unclear what cellular mechanism is used for ipsilateral target selection. The data support the following model: ORN axons send out transient interstitial branches at multiple locations along the main axon; the branch that reaches the target region becomes stabilized, and further interstitial branches are suppressed. Stabilization of appropriately positioned branches and elimination of ectopic branches are also used for topographic retinotopic targeting, suggesting that the mechanism of transient interstitial branching followed by stabilization applies to the formation of both continuous and discrete neural maps (Li, 2021).

The exploring branches that were discovered using AO-LLSM imaging suggest a means by which a growing ORN axon may increase the chance of identifying its target. These exploring branches consist of long, microtubule-based parallel branches that extend and retract rapidly and independently, allowing them to sample a relatively large region for possible targets. The transient occurrence of exploring branches when ORN axons approach their target region suggests that they are induced by local cues near target regions to facilitate target selection. In the ipsilateral antennal lobe, exploring branches were found in ORN types that form ipsilateral branches shortly after the main axon passes by, consistent with them serving as the precursor to the eventual ipsilateral branch. Exploring branches are also found in axon terminals in the contralateral antennal lobe in ORN types, suggesting a general role in facilitating contralateral target identification. For ORN types that have a long delay in extending the ipsilateral branch, no exploring branches were observed, suggesting a distinct mechanism for consolidating the ipsilateral branch. Nevertheless, dynamic interstitial branches occur over a prolong period of time until the formation of the ipsilateral branch, suggesting that these ORN types also use stabilization of transient interstitial branches as a means to consolidate the ipsilateral branch (Li, 2021).

In summary, after the initial trajectory choice such that ORN axons navigate in the half of the antennal lobe where their eventual targets are, it is proposed that the next critical step in ORN axon development is the stabilization of transient interstitial branches by target-derived cues, aided at least in part by the exploring branches. Together, these cellular mechanisms begin to explain how each ORN chooses one of 50 glomerular targets precisely (Li, 2021).

A surprising finding is that the cytoskeletal organization of ORN terminals differs substantially from that of classic growth cones, comprising F-actin-based filopodia and lamellipodia at the periphery and a microtubule-enriched central hub. The terminal branches of ORN axons, in particular the exploring branches, are filled with microtubules, whereas F-actin is concentrated at the central hub. Similar cytoskeletal organization was also found in photoreceptor axon terminals. These differences are unlikely due to species difference, as the classic growth cone cytoskeletal organization is found in neurons (mostly dissociated in culture) from Aplysia, Drosophila, and mammals. It cannot be ruled out that F-actin is present in low amount at the terminal of each exploring or post-innervation branch but is beyond the detection limit of the utrophin-based F-actin labeling; if so, each terminal branch would have its own growth cone at its tip, resembling classic growth cones. Even if that is the case, ORN axon terminals still differ from classic growth cones by having multiple microtubule-based parallel branches emanating from an F-actin rich central hub. Indeed, EB1-GFP puncta can be found at the tip of the branch, suggesting that microtubules can fill the entire branch. Microtubule polymerization has been shown to mediate membrane extension directly in lipid vesicle (Li, 2021).

It is suspected that the deviation of cytoskeletal organization in ORN axon terminals from the classic growth cone is likely due to the more complex environments axon terminals need to explore in the brain compared with the primary culture. Indeed, a recent study showed that neurons cultured in three-dimensional environments have microtubules extending to the edge of growth cones unconstrained by F-actin. The current findings have important implications for mechanisms that convert cell-surface recognition of extracellular cues into cytoskeletal-based structural changes in axon terminals during axon targeting. Specifically, we suggest that signaling to microtubule is particularly important at initial stages of target selection (Li, 2021).

Bilaterally symmetric organization of the nervous system is a cardinal feature of all bilaterians. Unilateral antennal nerve severing indicates the requirement of bilateral axons in target selection. The simplest cellular mechanism is direct interactions between ipsilateral and contralateral ORN axons. These interactions may facilitate midline crossing by creating a critical mass of midline-penetrating axons, disruption of which may cause some axons to leave the antennal lobe instead. Later, bilateral axon-axon interactions between the same ORN types may facilitate target selection of contralateral ORNs. The data do not rule out the possibility that bilateral interactions may be indirect; for example, ipsilateral ORNs may change the properties of their partners PNs, which in turn regulate target selection of contralateral ORNs. Indeed, upon unilateral antennal nerve severing, targeting defects was mostly found in ORN types that sequentially innervate ipsilateral and contralateral glomeruli. The ease of severing antennal nerve in explant cultures provides a means to further investigate cellular and molecular mechanisms of bilateral interactions (Li, 2021).

In conclusion, time-lapse imaging has greatly enriched understanding of the cellular events that enable the stepwise assembly of the fly olfactory circuit, and highlight the precise genetic control of multiple steps during ORN axon targeting. These include the choice of a trajectory along which an ORN axon navigates the ipsilateral antennal lobe, the timing and location of stabilizing its ipsilateral branch, and the interactions with contralateral ORN axons to cross the midline and innervate its contralateral target. Finally, ORN axons also help refine dendrites of their partner PNs, which pattern the antennal lobe first. The stage is set to combine live imaging and the cellular insights it has brought with genetic manipulations of key wiring molecules identified by genetic, transcriptomic, and proteomic approaches to reach a deeper level of mechanistic understanding of the circuit assembly process (Li, 2021).

Although the targeting precision in the explant culture mimics closely in vivo, it takes ORN axons longer to reach the same developmental stage in culture than in vivo. Thus, measurements involving time in explants may be protractions of equivalent events in vivo. The small number of single axons from specific ORN types, due to limited drivers that label specific ORN types strongly in early development, did not allow assessment of the variation of targeting behavior among ORNs of the same type. Although axon targeting of a large fraction of antennal ORN types was sampled, sample axons were not examined from 6 maxillary palp ORN types because the explant did not include maxillary palp. It is unclear whether maxillary palp ORN axons follow similar rules as antennal ORN axons. However, because maxillary palp ORN axons reach the antennal lobe substantially later than antennal ORN axons, the lack of maxillary palp ORN axons in explants should not affect the early stages of antennal ORN axon targeting (Li, 2021).

Consumption of food and water is tightly regulated by the nervous system to maintain internal nutrient homeostasis. Although generally considered independently, interactions between hunger and thirst drives are important to coordinate competing needs. In Drosophila, four neurons called the Interoceptive Subesophageal zone Neurons (ISNs) respond to intrinsic hunger and thirst signals to oppositely regulate sucrose and water ingestion. This study investigated the neural circuit downstream of the ISNs to examine how ingestion is regulated based on internal needs. Utilizing the recently available fly brain connectome, this study found that the ISNs synapse with a novel cell type Bilateral T-shaped neuron (BiT) that projects to neuroendocrine centers. In vivo neural manipulations revealed that BiT oppositely regulates sugar and water ingestion. Neuroendocrine cells downstream of ISNs include several peptide-releasing and peptide-sensing neurons, including insulin producing cells (IPC), crustacean cardioactive peptide (CCAP) neurons, and CCHamide-2 receptor isoform RA (CCHa2R-RA) neurons. These neurons contribute differentially to ingestion of sugar and water, with IPCs and CCAP neurons oppositely regulating sugar and water ingestion, and CCHa2R-RA neurons modulating only water ingestion. Thus, the decision to consume sugar or water occurs via regulation of a broad peptidergic network that integrates internal signals of nutritional state to generate nutrient-specific ingestion (Lez-Segarra, 2023).

Gustatory sensory neurons detect caloric and harmful compounds in potential food and convey this information to the brain to inform feeding decisions. To examine the signals that gustatory neurons transmit and receive, this study reconstructed gustatory axons and their synaptic sites in the adult Drosophila melanogaster brain, utilizing a whole-brain electron microscopy volume. 87 gustatory projections were reconstructed from the proboscis labellum in the right hemisphere and 57 from the left, representing the majority of labellar gustatory axons. Gustatory neurons contain a nearly equal number of interspersed pre- and postsynaptic sites, with extensive synaptic connectivity among gustatory axons. Morphology- and connectivity-based clustering revealed six distinct groups, likely representing neurons recognizing different taste modalities. The vast majority of synaptic connections are between neurons of the same group. This study resolves the anatomy of labellar gustatory projections, reveals that gustatory projections are segregated based on taste modality, and uncovers synaptic connections that may alter the transmission of gustatory signals (Engert, 2002).

All animals have specialized sensory neurons dedicated to the detection of the rich variety of chemicals in the environment that indicate the presence of food sources, predators, and conspecifics. Gustatory sensory neurons have evolved to detect food-associated chemicals and report the presence of caloric or potentially harmful compounds. Examining the activation and modulation of gustatory sensory neurons is essential as it places fundamental limits on the taste information that is funneled to the brain and integrated to form feeding decisions (Engert, 2002).

The Drosophila melanogaster gustatory system is an attractive model to examine the synaptic transmission of gustatory neurons. Molecular genetic approaches coupled with physiology and behavior have established five different classes of gustatory receptor neurons (GRNs) in adult Drosophila that detect different taste modalities. One class, expressing members of the gustatory receptor (GR) family, including Gr5a and Gr64f, detects sugars and elicits acceptance behavior. A second class expressing different GRs, including Gr66a, detects bitter compounds and mediates rejection behavior. A third class contains the ion channel Ppk28 and detects water. The fourth expresses the Ir94e ionotropic receptor, whereas the fifth contains the Ppk23 ion channel. These cells have been proposed to mediate detection of low-salt and high-salt concentrations, respectively. In addition to well-characterized gustatory neurons and a peripheral strategy for taste detection akin to mammals, the reduced number of neurons in the Drosophila nervous system and the availability of electron microscopy (EM) brain volumes offer the opportunity to examine gustatory transmission with high resolution (Engert, 2002).

The cell bodies of gustatory neurons are housed in sensilla on the body surface, including the proboscis labellum, an external mouthparts organ that detects taste compounds prior to ingestion. Gustatory neurons from each labellum half send bilaterally symmetric axonal projections to the subesophageal zone (SEZ) of the fly brain via the labial nerves. Gustatory axons terminate in the medial SEZ in a region called the anterior central sensory center (ACSC). Axons from bitter gustatory neurons send branches to the midline and form an interconnected medial ring, whereas other gustatory axons remain ipsilateral and anterolateral to bitter projections. Although projections of different gustatory classes have been mapped using light-level microscopy, the synaptic connectivity of gustatory axons in adult Drosophila is largely unexamined (Engert, 2002).

To explore the connectivity of GRNs and lay the groundwork to study gustatory circuits with synaptic resolution, the recently available Full Adult Fly Brain (FAFB) EM dataset to fully reconstruct gustatory axons and their synaptic sites. 87 GRN axonal projections were reconstructed in the right hemisphere and 57 in the left, representing between 83-96% and 54-63% of the total expected, respectively. By annotating chemical synapses, it was observed that GRNs contain a nearly equal number of interspersed pre- and postsynaptic sites. Interestingly, GRNs synapse onto and receive synaptic inputs from many other GRNs. Using morphology- and connectivity-based clustering, six distinct neural groups were identified, likely representing groups of GRNs that recognize different taste modalities. This study reveals extensive anatomical connectivity between GRNs within a taste modality, arguing for presynaptic processing of taste information prior to transmission to downstream circuits (Engert, 2002).

This study has characterized different classes of gustatory projections and their interconnectivity by high-resolution EM reconstruction. Different projection patterns corresponding to gustatory neurons were identified, recognizing different taste modalities. The extensive connections between GRNs of the same taste modality provide anatomical evidence of presynaptic processing of gustatory information (Engert, 2002).

An emerging theme stemming from EM reconstructions of Drosophila sensory systems is that sensory neurons of the same subclass are synaptically connected. In general, different sensory neuron subclasses have spatially segregated axonal termini in the brain, thereby constraining the potential for connectivity. In the adult olfactory system, approximately 40% of the input onto olfactory receptor neurons (ORNs) comes from other ORNs projecting to the same olfactory glomerulus. Similarly, mechanosensory projections from Johnston's organ of the same submodality are anatomically segregated and synaptically connected. In Drosophila larvae, 25% of gustatory neuron inputs are from other GRNs, although functional classes were not resolved. In the adult Drosophila gustatory system, GRNs are interconnected, with approximately 39% of GRN input coming from other GRNs. Consistent with other classes of sensory projections, this study found that gustatory projections are largely segregated based on taste modality and form connected groups. A general function of sensory-sensory connections seen across sensory modalities may be to enhance weak signals or increase dynamic range (Engert, 2002).

By clustering neurons based on anatomy and connectivity, it was possible to resolve different GRN categories. The distinct morphologies of bitter neurons and candidate low-salt-sensing neurons, known from immunohistochemistry, are recapitulated in the projection patterns of GRN groups 1-3 of the right hemisphere, enabling high-confidence identification. The projections of high-salt-, sugar-, and water-sensing neurons are ipsilateral, with similarities in their terminal arborizations. Nevertheless, comparisons between EM and light-level projections argue that these taste categories are also resolved into different, identifiable clusters. GRN categories were identified as low salt (Ir94e) and high salt (the remaining category) based on previous studies but note that the full complement of tastes that these GRNs detect requires additional investigation. The GRN categories that were identified in this study are based on anatomical comparisons alone and remain tentative until further examination of taste response profiles of connected second-order neurons, which may now be identified by examining connectivity downstream of GRNs (Engert, 2002).

This study reconstructed 83-96% of the GRNs on the right hemisphere and 54-63% on the left, based on total GRN counts from previous studies. GRN categories may be further refined upon reconstruction of the entire GRN population or upon analysis that includes postsynaptic partners. In addition, GRNs are found at different locations on the proboscis labellum and are housed in three taste bristle types. Segregation based on labellar location or bristle type may further divide the GRN categories described in this study. Interestingly, in the clustering analysis, it was found that bitter projections cluster into two distinct groups. It is hypothesized that these different subsets are comprised of bitter GRNs from different taste bristle classes or bitter GRNs with different response properties (Engert, 2002).

Examining GRN-GRN connectivity revealed connectivity between GRNs of the same group as well as different groups. While it is tempting to speculate that interactions between different taste modalities may amplify or filter activation of feeding circuits, this study was unable to identify cross-activation between sugar and water GRNs by calcium or voltage imaging. It is possible that these interactions are dependent on a feeding state or act on a time frame not examined in this study. Alternatively, activation may fall below the detection threshold of calcium or voltage imaging. Additionally, far fewer synapses occur between anatomical classes than within classes, especially restricting analyses to neurons connected by five or more synapses, suggesting that the few synapses may not be relevant for taste processing. Finally, the anatomy and connectivity-based clustering may not categorize all individual GRNs correctly, and misclassification of GRNs would impact connectivity analyses. Regardless, these studies suggest that presynaptic connectivity between different GRN classes does not substantially contribute to taste processing (Engert, 2002).

Overall, this study resolves the majority of labellar gustatory projections and their synaptic connections, revealing that gustatory projections are segregated based on taste modality and sensory-sensory connectivity. The identification of GRNs detecting different taste modalities now provides an inroad to enable the examination of the downstream circuits that integrate taste information and guide feeding decisions (Engert, 2002).

Gustatory sensory neurons detect caloric and harmful compounds in potential food and convey this information to the brain to inform feeding decisions. To examine the signals that gustatory neurons transmit and receive, this study reconstructed gustatory axons and their synaptic sites in the adult Drosophila melanogaster brain, utilizing a whole-brain electron microscopy volume. 87 gustatory projections were reconstructed from the proboscis labellum in the right hemisphere and 57 from the left, representing the majority of labellar gustatory axons. Gustatory neurons contain a nearly equal number of interspersed pre- and postsynaptic sites, with extensive synaptic connectivity among gustatory axons. Morphology- and connectivity-based clustering revealed six distinct groups, likely representing neurons recognizing different taste modalities. The vast majority of synaptic connections are between neurons of the same group. This study resolves the anatomy of labellar gustatory projections, reveals that gustatory projections are segregated based on taste modality, and uncovers synaptic connections that may alter the transmission of gustatory signals (Engert, 2002).

All animals have specialized sensory neurons dedicated to the detection of the rich variety of chemicals in the environment that indicate the presence of food sources, predators, and conspecifics. Gustatory sensory neurons have evolved to detect food-associated chemicals and report the presence of caloric or potentially harmful compounds. Examining the activation and modulation of gustatory sensory neurons is essential as it places fundamental limits on the taste information that is funneled to the brain and integrated to form feeding decisions (Engert, 2002).

The Drosophila melanogaster gustatory system is an attractive model to examine the synaptic transmission of gustatory neurons. Molecular genetic approaches coupled with physiology and behavior have established five different classes of gustatory receptor neurons (GRNs) in adult Drosophila that detect different taste modalities. One class, expressing members of the gustatory receptor (GR) family, including Gr5a and Gr64f, detects sugars and elicits acceptance behavior. A second class expressing different GRs, including Gr66a, detects bitter compounds and mediates rejection behavior. A third class contains the ion channel Ppk28 and detects water. The fourth expresses the Ir94e ionotropic receptor, whereas the fifth contains the Ppk23 ion channel. These cells have been proposed to mediate detection of low-salt and high-salt concentrations, respectively. In addition to well-characterized gustatory neurons and a peripheral strategy for taste detection akin to mammals, the reduced number of neurons in the Drosophila nervous system and the availability of electron microscopy (EM) brain volumes offer the opportunity to examine gustatory transmission with high resolution (Engert, 2002).

The cell bodies of gustatory neurons are housed in sensilla on the body surface, including the proboscis labellum, an external mouthparts organ that detects taste compounds prior to ingestion. Gustatory neurons from each labellum half send bilaterally symmetric axonal projections to the subesophageal zone (SEZ) of the fly brain via the labial nerves. Gustatory axons terminate in the medial SEZ in a region called the anterior central sensory center (ACSC). Axons from bitter gustatory neurons send branches to the midline and form an interconnected medial ring, whereas other gustatory axons remain ipsilateral and anterolateral to bitter projections. Although projections of different gustatory classes have been mapped using light-level microscopy, the synaptic connectivity of gustatory axons in adult Drosophila is largely unexamined (Engert, 2002).

To explore the connectivity of GRNs and lay the groundwork to study gustatory circuits with synaptic resolution, the recently available Full Adult Fly Brain (FAFB) EM dataset to fully reconstruct gustatory axons and their synaptic sites. 87 GRN axonal projections were reconstructed in the right hemisphere and 57 in the left, representing between 83-96% and 54-63% of the total expected, respectively. By annotating chemical synapses, it was observed that GRNs contain a nearly equal number of interspersed pre- and postsynaptic sites. Interestingly, GRNs synapse onto and receive synaptic inputs from many other GRNs. Using morphology- and connectivity-based clustering, six distinct neural groups were identified, likely representing groups of GRNs that recognize different taste modalities. This study reveals extensive anatomical connectivity between GRNs within a taste modality, arguing for presynaptic processing of taste information prior to transmission to downstream circuits (Engert, 2002).

This study has characterized different classes of gustatory projections and their interconnectivity by high-resolution EM reconstruction. Different projection patterns corresponding to gustatory neurons were identified, recognizing different taste modalities. The extensive connections between GRNs of the same taste modality provide anatomical evidence of presynaptic processing of gustatory information (Engert, 2002).

An emerging theme stemming from EM reconstructions of Drosophila sensory systems is that sensory neurons of the same subclass are synaptically connected. In general, different sensory neuron subclasses have spatially segregated axonal termini in the brain, thereby constraining the potential for connectivity. In the adult olfactory system, approximately 40% of the input onto olfactory receptor neurons (ORNs) comes from other ORNs projecting to the same olfactory glomerulus. Similarly, mechanosensory projections from Johnston's organ of the same submodality are anatomically segregated and synaptically connected. In Drosophila larvae, 25% of gustatory neuron inputs are from other GRNs, although functional classes were not resolved. In the adult Drosophila gustatory system, GRNs are interconnected, with approximately 39% of GRN input coming from other GRNs. Consistent with other classes of sensory projections, this study found that gustatory projections are largely segregated based on taste modality and form connected groups. A general function of sensory-sensory connections seen across sensory modalities may be to enhance weak signals or increase dynamic range (Engert, 2002).

By clustering neurons based on anatomy and connectivity, it was possible to resolve different GRN categories. The distinct morphologies of bitter neurons and candidate low-salt-sensing neurons, known from immunohistochemistry, are recapitulated in the projection patterns of GRN groups 1-3 of the right hemisphere, enabling high-confidence identification. The projections of high-salt-, sugar-, and water-sensing neurons are ipsilateral, with similarities in their terminal arborizations. Nevertheless, comparisons between EM and light-level projections argue that these taste categories are also resolved into different, identifiable clusters. GRN categories were identified as low salt (Ir94e) and high salt (the remaining category) based on previous studies but note that the full complement of tastes that these GRNs detect requires additional investigation. The GRN categories that were identified in this study are based on anatomical comparisons alone and remain tentative until further examination of taste response profiles of connected second-order neurons, which may now be identified by examining connectivity downstream of GRNs (Engert, 2002).

This study reconstructed 83-96% of the GRNs on the right hemisphere and 54-63% on the left, based on total GRN counts from previous studies. GRN categories may be further refined upon reconstruction of the entire GRN population or upon analysis that includes postsynaptic partners. In addition, GRNs are found at different locations on the proboscis labellum and are housed in three taste bristle types. Segregation based on labellar location or bristle type may further divide the GRN categories described in this study. Interestingly, in the clustering analysis, it was found that bitter projections cluster into two distinct groups. It is hypothesized that these different subsets are comprised of bitter GRNs from different taste bristle classes or bitter GRNs with different response properties (Engert, 2002).

Examining GRN-GRN connectivity revealed connectivity between GRNs of the same group as well as different groups. While it is tempting to speculate that interactions between different taste modalities may amplify or filter activation of feeding circuits, this study was unable to identify cross-activation between sugar and water GRNs by calcium or voltage imaging. It is possible that these interactions are dependent on a feeding state or act on a time frame not examined in this study. Alternatively, activation may fall below the detection threshold of calcium or voltage imaging. Additionally, far fewer synapses occur between anatomical classes than within classes, especially restricting analyses to neurons connected by five or more synapses, suggesting that the few synapses may not be relevant for taste processing. Finally, the anatomy and connectivity-based clustering may not categorize all individual GRNs correctly, and misclassification of GRNs would impact connectivity analyses. Regardless, these studies suggest that presynaptic connectivity between different GRN classes does not substantially contribute to taste processing (Engert, 2002).

Overall, this study resolves the majority of labellar gustatory projections and their synaptic connections, revealing that gustatory projections are segregated based on taste modality and sensory-sensory connectivity. The identification of GRNs detecting different taste modalities now provides an inroad to enable the examination of the downstream circuits that integrate taste information and guide feeding decisions (Engert, 2002).

Sweet and bitter compounds excite different sensory cells and drive opposing behaviors. However, it remains unclear how sweet and bitter tastes are represented by the neural circuits linking sensation to behavior. To investigate this question in Drosophila, this study devised trans-Tango(activity), a strategy for calcium imaging of second-order gustatory projection neurons based on trans-Tango, a genetic transsynaptic tracing technique. Spatial overlap was found between the projection neuron populations activated by sweet and bitter tastants. The spatial representation of bitter tastants in the projection neurons was consistent, while that of sweet tastants was heterogeneous. Furthermore, it wads discovered that bitter tastants evoke responses in the gustatory receptor neurons and projection neurons upon both stimulus onset and offset and that bitter offset and sweet onset excite overlapping second-order projections. These findings demonstrate an unexpected complexity in the representation of sweet and bitter tastants by second-order neurons of the gustatory circuit (Snell, 2022).

In flies, neuronal sensors detect prandial changes in circulating fructose levels and either sustain or terminate feeding, depending on internal state. This study describes a three-part neural circuit that imparts satiety-dependent modulation of fructose sensing. Dorsal fan-shaped body neurons display oscillatory calcium activity when hemolymph glucose is high, and these oscillations require glutamatergic input from SLP-AB or 'Janus' neurons projecting from the protocerebrum to the asymmetric body. Suppression of activity in this circuit, either by starvation or by genetic silencing, promotes specific drive for fructose ingestion. This is achieved through neuropeptidergic signaling by tachykinin, which is released from the fan-shaped body when glycemia is high. Tachykinin, in turn, signals to Gr43a-positive fructose sensors to modulate their response to fructose. Together, these results demonstrate how a three-layer neural circuit links the detection of two sugars to produce precise satiety-dependent control of feeding behavior (Musso, 2021).

Regulation of energy intake is a complex process involving food search, an animal's internal state, and the sensory qualities of food. In flies, fructose, either consumed directly or rapidly metabolized from precursors, promotes feeding through activation of a brain fructose sensor called Gr43a. This study describes how a neuronal network composed of neurons in the FB and asymmetric body contributes to energy homeostasis by detecting satiety-dependent changes in hemolymph glucose and modulating fructose drive (Musso, 2021).

The central complex, which is composed of the FB, the protocerebral bridge (PB), the ellipsoid body, and the noduli, is regarded as a center for sensorimotor integration that functions in goal-directed behavior. The FB is organized in nine horizontal layers and nine vertical columns. FB large field neurons of layers 1 to 3, and inputs to these layers from the PB, encode flight direction and general sensory orientation. FB layers 6 and 7 are well known to regulate sleep and arousal, locomotor control, courtship, visual memory, and decision-making related to taste. Layer 6 also plays a role in avoiding conditioned odors, while layers 1, 2, 4, and 5 respond to electric stimuli and are required for innate odor avoidance. However, the function of the most dorsal FB layers (8 and 9), mostly innervated local tangential neurons and AB-FBl8 (or vΔA_a), remained poorly understood. The results demonstrate a role for these layers in feeding regulation (Musso, 2021).

dFB oscillations were found to be require glutamatergic input from Janus neuron projections to the asymmetric body. Described for the first time in 2004, very little is known about AB function; 92.4% of flies display asymmetry in the AB, with the body present only in the right hemisphere, while 7.6% also have a body on the left side. It is noted that oscillations in the dFB display a tendency to be faster on the right side, with clearly asynchronous activity between the two sides that may reflect their asymmetric input from Janus neurons. The small proportion of flies displaying symmetry in the AB have defects in LTM, a process that is known to require energy. It is speculated that these symmetric flies may have a dysfunctional Janus neurons-to-dFB connection, resulting in impaired Tk release. This could affect LTM either directly or through changes in feeding. A role for TK in memory has been demonstrated in honeybees and mammals, and TkR86C appears to be expressed in serotonergic paired neurons known to interact with MB-MP1 neurons required for LTM formation. Tk also acts through TkR99D to modulate activity in neurons producing insulin-like peptides, which affect LTM formation (Musso, 2021).

Modulation of dFB oscillations by Janus neurons requires glutamatergic signaling through a group of glutamate receptors including KaiR1D, NmdaR1, NmdaR2, and GluClα, but not AMPA receptors. Both KaiR1D receptors, which are homomeric, and N-methyl-D-aspartate (NMDA) receptors, which are heteromeric complexes between subunits 1 and 2, pass Ca2+ current. NMDA receptors (NMDAR) are well known for their role in mediating synaptic plasticity and can also trigger oscillatory activity. NMDAR function as molecular coincidence detectors, requiring simultaneous ligand binding and membrane depolarization for activation. It is possible that dFB neuron oscillations are triggered by the coincident detection of glutamate from Janus neurons and glucose from the hemolymph; however, because the FB are receiving many inputs from other brain region, it is suspect that dFB oscillations require additional inputs as well. The chloride channel GluClα is also required for dFB oscillations. GluClα has been previously implicated in on/off responses of the visual system of flies and memory retention in honeybees, demonstrating a role in regulating cell excitability. Perhaps, GluClα functions in repolarization of the dFB neurons between calcium bursts. Further study will be required to fully understand how the suite of glutamate receptors function together to drive oscillations, along with the source of input to Janus neurons in the protocerebrum (Musso, 2021).

Because glucose is the primary circulating energy source, one might intuitively expect that enhancing feeding in response to postingestive glucose detection would be the most efficient means of optimizing energy uptake. However, using elevation of hemolymph glucose as a signal to continue feeding is problematic because glucose levels are tightly regulated and elevated glucose serves as a signal of satiety. On the other hand, internal fructose can vary widely in response to ingestion and can therefore be a more reliable indicator of recent sugar intake. Thus, the separation of glucose as a satiety indicator and fructose as marker of sugar consumption removes the potential ambiguity of each as a signal. Moreover, fructose typically coexists with other nutritive sugars in common food sources. Therefore, it may not be the case that flies specifically benefit from fructose intake but rather that fructose serves as an effective proxy for general carbohydrate ingestion. By using fructose and the narrowly tuned Gr43a fructose receptor to survey sugar consumption, flies can effectively benefit from both a fructose-mediated positive feedback loop and glucose-mediated negative feedback to co-operatively ensure appropriate energy intake (Musso, 2021).

The finding that dFB glucose sensing modulates fructose feeding via Gr43a brain neurons fits with the established model of Gr43a brain neurons as central fructose sensors. For this mechanism to effectively sustain feeding on a rich sugar source, ingested sugars must rapidly increase fructose signaling to Gr43a brain neurons, which then must acutely promote feeding. While the precise kinetics of internal fructose elevation after sugar consumption have not been quantified, fructose levels in the head rapidly increase 10-fold after fructose feeding and then return to baseline. The role of direct fructose sensing by Gr43a brain neurons is highlighted by the observation that Gr43a knockdown in those neurons results in markedly lower relative intake of fructose compared to glucose. Unexpectedly, knockdown of Gr64a, another sugar receptor expressed in the same neurons, produced the opposite effect. This could be because Gr64a contributes to modulation of Gr43a brain neurons by other sugar cues, and the absence of this activity makes Gr43a-mediated fructose responses more pronounced. Alternatively, Gr43a may be expressed more strongly after Gr64a knockdown, leading to an increased fructose response (Musso, 2021).

Little is known about the mechanisms downstream of Gr43a brain neurons that promote feeding. All Gr43a brain neurons express the peptide Crz, but knockdown of Crz expression produced no significant effect on fructose preference over glucose. This suggests an important functional role for another neurotransmitter, although it is also possible that the RNAi knockdown was not effective. Irrespective of mechanism, two experiments support the idea that activation of Gr43a neurons acutely enhances feeding. First, silencing of dFB neurons by genetic manipulation or prolonged starvation produces Gr43a-dependent fructose preference within the first 10 min of a flyPAD assay. Second, closed-loop optogenetic activation of Gr43a brain neurons was sufficient to produce a strong positive preference within 10 min in the STROBE (Musso, 2021).

The separable functions of glucose and fructose sensing in flies bear notable resemblance to the differential effects of these two sugars in the mammalian hypothalamus. In particular, AMPK expression in the arcuate nucleus of the hypothalamus is known to link energy levels to food drive. When glycemia is low, AMPK is activated and thereby promotes feeding through orexigenic AgRP/NPY neuron activity. Glucose administration suppresses activity in these peptidergic neurons, while fructose can have the opposite effect and promote further feeding. The first description of fly Gr43a neurons noted their orexinegenic activity and suggested a potential functional homology with the hypothalamus. In the present study, a multilayered neural system centered on a brain energy sensor (dFB), was uncovered whose activation by glucose leads to anorexigenic behavior through inhibition of the brain fructose sensor Gr43a. Thus, the results are consistent with at least partial functional homology between the mammalian hypothalamus and brain Gr43a neurons of the fly (Musso, 2021).

Habituated animals retain a latent capacity for robust engagement with familiar stimuli. In most instances, the ability to override habituation is best explained by postulating that habituation arises from the potentiation of inhibitory inputs onto stimulus-encoding assemblies and that habituation override occurs through disinhibition. Previous work has shown that inhibitory plasticity contributes to specific forms of olfactory and gustatory habituation in Drosophila. This study analyzed how exposure to a novel stimulus causes override of gustatory (proboscis extension reflex; PER) habituation. While brief sucrose contact with tarsal hairs causes naive Drosophila to extend their proboscis, persistent exposure reduces PER to subsequent sucrose stimuli. This study shows that in so habituated animals, either brief exposure of the proboscis to yeast or direct thermogenetic activation of sensory neurons restores PER response to tarsal sucrose stimulation. Similar override of PER habituation can also be induced by brief thermogenetic activation of a population of tyrosine hydroxylase (TH)-positive neurons, a subset of which send projections to the subesophageal zone (SEZ). Significantly, sensory-neuron induced habituation override requires transmitter release from these TH-positive cells. Treatments that cause override specifically influence the habituated state, with no effect on the naive sucrose response across a range of concentrations. Taken together with other findings, these observations in female flies are consistent with a model in which novel taste stimuli trigger activity in dopaminergic neurons which, directly or indirectly, inhibit GABAergic cells that drive PER habituation. The implications of these findings for general mechanisms of attentional and sensory override of habituation are discussed (Trisal, 2022).

Swallowing is an essential step of eating and drinking. However, how the quality of a food bolus is sensed by pharyngeal neurons is largely unknown. This study finds that mechanical receptors along the Drosophila pharynx are required for control of meal size, especially for food of high viscosity. The mechanical force exerted by the bolus passing across the pharynx is detected by neurons expressing the mechanotransduction channel NOMPC (no mechanoreceptor potential C) and is relayed, together with gustatory information, to IN1 neurons in the subesophageal zone (SEZ) of the brain. IN1 (ingestion neurons) neurons act directly upstream of a group of peptidergic neurons that encode satiety. Prolonged activation of IN1 neurons suppresses feeding. IN1 neurons receive inhibition from DSOG1 (descending subesophageal neurons) neurons, a group of GABAergic neurons that non-selectively suppress feeding. These results reveal the function of pharyngeal mechanoreceptors and their downstream neural circuits in the control of food ingestion (Yang, 2021).

Overconsumption is harmful for animals. Although the drive to ingest can be overwhelming for a hungry animal in the initial stage of a meal, inhibition becomes more dominant with the processes of food intake. This study found that food flowing across the pharynx accumulates the satiety state in the brain, demonstrating that multiple strategies are used by the nervous systems to avoid overeating. These pharyngeal sensory neurons are sensitive to sugar and mechanical force, serving as a flow meter that monitors food quality and amount so that the brain knows how much food is ingested even before the food reaches the intestine. This circuit may coordinate with other satiety signals, such as those conveyed by mechanical feedback from the intestine, to control feeding (Yang, 2021).

Gustatory and mechanosensory neurons are well separated on the fly labellum before their axons reach the SEZ, where they interact with each other to regulate the perception of food quality. In contrast, the sensory neurons in the pharynx seem to adapt a different coding mechanism. Some of the pharyngeal neurons are polymodal because they respond to chemical and mechanical stimuli, with PM neurons being an example. A 'generalist' versus 'specialist strategy has been found in other sensory organs too. Being able to evaluate multiple properties of a bolus in the pharynx allow the animals to effectively control the feeding amount. There are sensory neurons in the pharynx that may be tuned to gustatory or mechanosensory cues. For example, the R41E11-GAL4 and nompC-QF labeled approximately 10 pairs of neurons in LSOs along the pharynx, similar to the number observed for mechanosensory neurons. Most of those neurons are likely 'generalist' and are tuned to mechanical stimuli only. It would be valuable to determine the full repertoire of these sensory neurons to understand how the swallowing maneuver is initiated, sustained, and terminated (Yang, 2021).

It has been proposed that IN1 neurons may function as a key node of the feeding control circuits to govern rapid feeding decisions. Previous studies have revealed that IN1 neurons are directly downstream of pharyngeal GRNs and that activation of IN1 neurons to sugar stimulation is correlated with a fly's motivation to feed. Because activation of IN1 neurons triggers proboscis extension to food, they are likely upstream of the motor circuit that controls feeding. IN1 neurons thus appear to function as a hub that integrates sensory information to initiate food ingestion. This study found that IN1 neurons' activity is under control of the fly's feeding states. IN1 neurons are directly downstream of DSOG1 neurons that non-selectively suppress ingestion. In fed flies, DSOG1 neurons impart inhibition on IN1 neurons, resulting in a transient and moderate response to a sugar sip that triggers a robust and sustained calcium response in fasted flies (Yang, 2021).

It has been proposed that DSOG1 neurons impart constant inhibition on the neuronal circuits that initiate food ingestion. However, the upstream circuit of DSOG1 neurons has not been identified. A cohort of neuropeptide receptor genes has been screened, but none of them seemed to function on DSOG1 neurons in feeding control. This study found that interrupting signaling of the neuropeptide MIP phenocopied overfeeding in flies with silenced DSOG1 neurons. It is tantalizing to hypothesize that MIP neurons are upstream of the DSOG1 circuit, either directly or indirectly. Because the receptors of MIP have not yet been identified, further experiments are need to differentiate between the two possibilities (Yang, 2021).

Besides PM neurons, there are many NOMPC-expressing mechanosensory neurons along the fly pharynx. Because of the lack of specific driver lines and the technique to record a single neuron's activity during feeding, their roles in feeding regulation are interesting open questions and await in-depth investigation. Moreover, the receptors of MIP peptide have not been identified, especially the ones involved in feeding regulation, making it difficult to establish a connection between MIP neurons and DSOG1 neurons (Yang, 2021).

Neural circuits carry out complex computations that allow animals to evaluate food, select mates, move toward attractive stimuli, and move away from threats. In insects, the subesophageal zone (SEZ) is a brain region that receives gustatory, pheromonal, and mechanosensory inputs and contributes to the control of diverse behaviors, including feeding, grooming, and locomotion. Despite its importance in sensorimotor transformations, the study of SEZ circuits has been hindered by limited knowledge of the underlying diversity of SEZ neurons. This study generate a collection of split-GAL4 lines that provides precise genetic targeting of 138 different SEZ cell types in adult D. melanogaster, comprising approximately one third of all SEZ neurons. The single cell anatomy of these neurons was characterized, and they were found to cluster by morphology into six supergroups that organize the SEZ into discrete anatomical domains. The majority of local SEZ interneurons are not classically polarized, suggesting rich local processing, whereas SEZ projection neurons tend to be classically polarized, conveying information to a limited number of higher brain regions. This study provides insight into the anatomical organization of the SEZ and generates resources that will facilitate further study of SEZ neurons and their contributions to sensory processing and behavior (Sterne, 2021).

This study describes the SEZ Split-GAL4 Collection, a library of 277 split-GAL4 lines covering 138 SEZ cell types, which affords unprecedented genetic access to SEZ neurons for behavioral and functional study. These studies provide insight into the diversity of SEZ cell types and their organization into discrete anatomical domains. The SEZ Split-GAL4 Collection will enable further investigation of how local SEZ circuitry and ascending SEZ paths process sensory inputs and control specific behaviors (Sterne, 2021).

Most of the SEZ Split-GAL4 lines are specific, with 149/277 lines classified as ideal or excellent. These lines will be useful to manipulate individual SEZ cell types for behavioral, functional, and imaging experiments. The remaining, less specific, lines (those belonging to the good or poor categories) will still be useful for imaging and as starting points for creating more specific reagents. Good and poor lines may be used to generate CDM masks to search for new hemidrivers to make further split-GAL4 lines. Alternatively, their expression patterns may be refined using Killer Zipper or three-way intersections with LexA or QF lines. All lines in the SEZ Split-GAL4 Collection may be used to generate further tools including complementary split-LexA and split-QF reagents. Split-LexA and split-QF lines may be used in concert with the split-GAL4 lines reported here to simultaneously manipulate two independent neuronal populations for advanced intersectional experiments, including behavioral epistasis (Sterne, 2021).

By combining insights from a single-cell transcriptome atlas with direct cell counts of SEZ neuromeres, it is estimated that the SEZ Split-GAL4 Collection labels 30% of the ~1700 neurons in the SEZ. Because of the lack of stereotyped neuronal cell body positions in D. melanogaster, it is not possible to assign cell bodies to defined neuropil regions without a genetic marker. The advantage of this method of estimating SEZ neuron number is that it is based on analysis of the four genetically defined SEZ neuromeres, the tritocerebral, the mandibular, the maxillary, and the labial neuromeres. However, previous reports demonstrate that some deutocerebral commissures cross below the esophageal foramen, and therefore an unknown number of deutocerebral cell bodies may be part of the SEZ. The limitations of this estimate of SEZ neuron number therefore include the inability to directly count cells derived from the tritocerebral neuromere, the inability to directly count neurons rather than glia, and the inability to assess deutocerebral contributions. Thus, the estimate of SEZ cell number is likely an underestimate. Once all SEZ neurons are densely reconstructed in an EM volume, direct counts of SEZ neuronal cell bodies obtained by EM will provide a more accurate assessment of SEZ neuron number. Regardless, the SEZ Split-GAL4 Collection targets 510 neuronal cell bodies, which represents a substantial improvement in the ability to precisely target SEZ cell types for functional and morphological analysis. This study did not ascertain the neuromere or neuroblast of origin of the SEZ cell types in the SEZ Split-GAL4 Collection. However, recent work has established reliable anatomical criteria that define the boundaries between the four SEZ neuromeres and has mapped all secondary lineages of the SEZ. Future efforts should focus on bridging previously identified fascicle, neuropil, and sensory domains into a common template or coordinate space to determine the neuromere and neuroblast origin of SEZ cell types (Sterne, 2021).

Discovering and genetically targeting SEZ cell types required the use of registered light-level imagery and computer-assisted searching. Four distinct strategies were used to identify 129 novel and 9 previously reported SEZ cell types in registered light-level imagery. Critically, each of these strategies allowed use of CDM mask searching to identify additional hemidrivers with which to target each cell type of interest. CDM mask searching enabled combing of large datasets and greatly increased the ease and speed of split-GAL4 generation over previous methods. The same strategies can be leveraged to gain genetic access to yet-undiscovered SEZ cell types. The recent electron microscopy (EM) volumes of the D. melanogaster brain provide an avenue for identifying SEZ cell types that are not covered by the SEZ Split-GAL4 Collection. Notably, this approach awaits comprehensive reconstruction of the SEZ, a region that is not included in the recently published dense reconstruction of the 'hemibrain' volume. Another EM volume, 'FAFB,' provides imagery of an entire adult female fly brain at synaptic resolution and includes the SEZ. Improvements in automated reconstruction of EM volumes coupled with large-scale human annotation should soon provide exhaustive reconstruction of the SEZ from which to identify additional SEZ cell types. Furthermore, available bridging registrations between EM volumes and light-level imagery should facilitate the identification of hemidrivers to target SEZ cell types discovered from EM reconstructions. Even without identifying additional SEZ cell types, the split-GAL4 reagents described will allow behavioral and functional evaluation of circuit hypotheses derived from EM imagery (Sterne, 2021).

These analyses of the SEZ Split-GAL4 Collection provide insight into the cellular architecture of the SEZ. To computationally probe the organization of the SEZ, 121 SEZ cell types were morphologically clustered using NBLAST. This approach reveals six cellular domains in the SEZ that are organized in a largely layered fashion from anterior to posterior. This layered structure is also hinted at by the recent description of SEZ neuropil domains throughout development from the larva to the adult. Based on anatomical position and the known function of a few SEZ neurons, it is tempting to speculate that different morphological clusters may participate in different behavioral functions. Group 1 contains projection neurons that innervate the region of the SMP surrounding the pars intercerebralis (PI), suggesting that group 1 neurons may impinge on neurosecretory neurons or function in energy and fluid homeostasis circuits. The proximity of group 1 interneurons to previously described interoceptive SEZ neurons (ISNs) and ingestion neurons (IN1) supports this hypothesis. Group 2 contains Fdg, a feeding-related neuron, as well as cell types (indigo, tinctoria) that are located near pumping motor neurons, suggesting that group 2 neurons have roles in feeding sequence generation. Group 3 contains G2N-1, a candidate second-order gustatory neuron, and projection neurons that innervate recently described taste-responsive SLP regions, suggesting that group 3 may, in part, be composed of taste-responsive neurons. Many interneurons in group 4 are located near proboscis motor neurons that control rostrum protraction, haustellum extension, and labellar spreading, indicating that group 4 members function in proboscis motor control. The proximity of neurons in group 5 to previously described stopping neuron MAN, and the inclusion of an antennal grooming neuron, suggests that group 5 neurons may participate in circuits that control grooming and stopping behaviors. Group 6 is located in the posterior SEZ and posterior slope, regions implicated in flight behaviors, including wing and neck control. While potential behavioral functions are hypothesized for each supergroup, it is readily acknowledged that the roles of the neurons described in this study are likely more diverse (Sterne, 2021).

These studies also shed light on information flow both within the SEZ and out of the SEZ to the higher brain. 91 local interneurons, 30 projection neurons, 16 descending neurons, and 1 sensory neuron were identified. Polarity analysis of 121/138 of the SEZ cell types covered by the SEZ Split-GAL4 Collection revealed that SEZ interneurons tend to have mixed or biased polarity while SEZ projection neurons tend to be classically polarized. Polarity analyses of the lateral horn, mushroom body, descending neurons, and protocerebral bridge identified few neurons with completely mixed polarity. Unlike these brain regions, the SEZ contains a large number of local interneurons. The mixed polarity of the SEZ interneurons argues for local and reciprocal connectivity between neurons, with information flowing in networks rather than unidirectional streams. Projection neurons, in contrast, may serve chiefly to pass information from highly interconnected SEZ circuits to other brain regions in a unidirectional manner. Notably, many SEZ projection neurons were identified that innervate the SMP-a region known to contain neurosecretory cell types. This may betray a role for acute taste detection or feeding circuit activation in the regulation of hormone secretion. In addition, the frequent innervation of the superior lateral protocerebrum and lateral horn by SEZ projection neurons may hint at the site of olfactory-gustatory synthesis. In contrast, this study did not identify projection neurons that link the SEZ directly to the central complex or mushroom body. If dense reconstruction of EM volumes corroborates the lack of direct connectivity between the SEZ and these regions, information must be conveyed through indirect pathways. As an example, taste information influences local search behaviors during foraging, a task that is expected to involve the central complex. Indirect relay of taste information to the central complex to inform foraging behavior would be consistent with previous anatomical studies suggesting that the central complex receives diverse indirect sensory inputs. Furthermore, the mushroom body is known to respond to taste, raising the possibility that taste information from gustatory sensory neuron axons in the SEZ must be relayed through yet another brain region before reaching mushroom body cell types. Thus, this analysis of SEZ neuron polarity indicates local SEZ processing and demonstrates direct pathways to a subset of higher brain regions (Sterne, 2021).

Overall, the SEZ Split-GAL4 Collection represents a valuable resource that will facilitate the study of the SEZ. This analysis of the collection reveals the cellular anatomy and polarity of individual SEZ neurons and their organization into six discrete domains in the SEZ. Coupled with emerging insights from reconstruction of EM volumes, the SEZ Split-GAL4 Collection will allow the use of genetic dissection to test circuit-level hypotheses about sensory processing and motor control in the SEZ (Sterne, 2021).

Pavlovian conditioning is a broadly used learning paradigm where defined stimuli are associated to induce behavioral switching. To define a causal relationship between activity change in a single neuron and behavioral switching, this study took advantage of a 'command neuron' that connects cellular function to behavior. To examine the cellular and molecular basis of Pavlovian conditioning, previous work identified a pair of feeding command neurons termed 'feeding neurons' in the adult Drosophila brain using genetic screening and opto- and thermo-genetic techniques. The feeding neuron is activated by sweet signals like sucrose and induces the full complement of feeding behaviors, such as proboscis extension and food pumping. Ablation or inactivation of the pair of feeding neurons abolishes feeding behavior, suggesting that this single pair of neurons is indispensable for natural feeding behaviors. This study describes a novel conditioning protocol to associate a signal-mediating rod removal from legs (conditioned stimulus [CS]) to feeding behavior induced by sucrose stimulation (unconditioned stimulus [US]). Calcium imaging of the feeding neuron demonstrated it acquires responsiveness to CS during conditioning, with inactivation of the feeding neuron during conditioning suppressing plasticity. These results suggest conditioning alters signals flowing from the CS into the feeding circuit, with the feeding neuron functioning as a key integrative hub for Hebbian plasticity (Sakurai, 2021).

This study demonstrate Pavlovian conditioning between tactile (CS) and gustatory (US) stimuli results in altered information processing by a pair of command neurons that control the Drosophila feeding circuit. This conditioning paradigm creates CS-induced excitement of the feeding neuron that commands feeding behavior in this animal, with the conditioned response requiring activity of the feeding neuron during pairing. Pioneering studies by Kandel and colleagues demonstrated the first synaptic and cellular mechanism underlying classical conditioning using the Aplysia gill withdrawal response. In Aplysia, the presynaptic terminal of a sensory neuron innervating the motor neuron was modulated by serotonin. Presynaptic modulation as a mechanism to generate Drosophila valence behaviors has been extensively studied, and recent progress indicates presynaptic terminals innervating mushroom body output neurons are modulated by dopaminergic neurons to establish Drosophila valence through appetitive and aversive olfactory association. Neither Aplysia plasticity nor Drosophila valence in these paradigms requires postsynaptic activity during learning. In contrast, Hebb proposed general principles to explain mechanisms for memory formation that better match results from commonly used mammalian experimental models, such as hippocampal long-term potentiation (LTP).Hebb postulated sequential firing of a presynaptic neuron and postsynaptic partner strengthens their connection. The requirement of feeding neuron activity for the conditioned response observed in this study fits well to a Hebbian mechanism if the underlying change is manifested in altered synaptic properties (although it cannot be excluded that inactivation of the feeding neuron and subsequent behavioral changes also alter neuromodulation, influencing memory formation). During association, CS-conveying neurons and the feeding neuron driven by sucrose stimulation would now fire together, resulting in strengthened connection between CS-conveying neurons and the feeding neuron according to a Hebbian mechanism. The response to US, however, did not change during conditioning, suggesting that connections between US-conveying neurons and the feeding neuron were not altered. Thus, one can hypothesize that the CS-feeding neuron circuit was newly established, whereas the pre-existing US-feeding neuron connection was not changed. These results suggest that Pavlovian conditioning is established through a change in information processing by the command neuron, which functions as the integrative hub of the feeding circuit (Sakurai, 2021).

This Pavlovian conditioning mechanism can also accommodate presynaptic modulation as demonstrated in Aplysia plasticity and Drosophila valence if reward signals are coupled to Hebbian plasticity through presynaptic neuromodulation. For Drosophila valance memory, reward signals consist of both sweet sensing and nutrition. Similar reward signals are likely to be relevant in vivo for Pavlovian conditioning, although the nutrition reward is eliminated in the current study due to removal of the esophagus from the preparation and application of a sucrose-wet paper strip only to the sensilla of the proboscis. Therefore, reward signals are likely constant between the groups that were tested, even for different US responses in the halorhodopsin experiments. Thus, differences in reward signal can be excluded from the altered conditioned responses observed between the groups. It is hypothesized that inactivation of the feeding neuron results in weaker memory due to postsynaptic activity in this neuron contributing to memory formation independent of changes in the reward signal. It is speculated that reward signals in the current model may also be mediated by dopamine, octopamine, or serotonin, similar to their role as reward signals in the mushroom bodies for Drosophila valence memory. In Aplysia, presynaptic adenylyl cyclase, which synthesizes cAMP, is believed to associate CS and US in this conditioning paradigm through US-driven serotonin modulation of the presynaptic terminal of the CS-conveying neuron. Adenylyl cyclase is encoded by rut, while dnc encodes a cAMP phosphodiesterase that degrades cAMP. As demonstrated in Aplysia and Drosophila, cAMP functions as a signal to modulate synaptic transmission. Given its role in LTP, cAMP is likely to play a critical role in Hebbian plasticity as well, consistent with the disruption of CS-US pairing in rut and dnc mutants. Considering the involvement of postsynaptic cells in Hebbian plasticity, retrograde signals from the postsynaptic cell can also be coupled to presynaptic cAMP signaling, as demonstrated previously at the Drosophila neuromuscular junction (Sakurai, 2021).

In the original experiment conducted by Pavlov, it is speculated there are groups of neurons that command feeding behavior in the dog. CS/US association may change responsiveness of a subset of those neurons that result in sound-induced saliva secretion, even in the absence of food signals. Electrophysiological studies have shown neural responses to CS are altered after Pavlovian conditioning in cat red nucleus and rabbit cerebellum, although how this kind of plastic change leads to alterations in command neuron function is unknown. Neurons with command function have been identified across many species. A command neuron is pivotally located within the sensorimotor watershed of a neuronal circuit and triggers a behavioral program after integrating numerous sensory inputs. Command neurons were first identified in crayfish through experiments where electrical stimulation of a certain neuron switched on or off behaviors, such as rhythmical movement of the swimmeret or escape responses. After identification of command neurons in invertebrate CNSs Mauthner cells were demonstrated to command escape behavior in fish. Recently, a group of neurons commanding feeding behavior have been identified in the mouse brain. Therefore, the scheme shown in Figure 4D may represent a common mechanism underlying Pavlovian conditioning across species, given the role of command neurons as an integrative hub within the sensorimotor watershed of neuronal circuits (Sakurai, 2021).

The feeding neuron in the Drosophila brain functions as a single command neuron pair that triggers the entire feeding program.3 This feature allowed reliable demonstration that CS-induced activation of the feeding neuron after conditioning was as robust as US-induced activation, suggesting the CS-induced activation of the feeding neuron can trigger the conditioned behavior. Thus, neurophysiological changes can be unambiguously correlated with behavioral change, making the causal relationship clear and allowing reliable manipulation. The current results are consistent with the assumption that both the CS signal and the US signal converge at a single identified neuron through a Hebbian mechanism. Taking advantage of the defined circuit with the feeding neuron at the center, it is not possible to define the cellular and molecular mechanisms for synaptic plasticity using this experimentally accessible neuron within the CNS. This approach, coupled with real-time live imaging, may allow tracking of changes in the structure or activity of identified synapses responsible for memory formation once CS-conveying neurons are defined in the experimental system. If so, it may be possible to directly observe pre- and/or postsynaptic changes mediating memory formation on the dendrite of the feeding neuron. Whether a new circuit is generated by strengthening a rudimentary pre-existing connection or a new connection forms de novo during associative conditioning will require future analysis. Molecular and cellular mechanisms underlying this plastic change can be investigated in detail as previously characterized at neuromuscular junctions. Taken together, the study of synaptic plasticity in the feeding neuron provides a model system to characterize basic principles of memory formation at the single-cell level (Sakurai, 2021).

Sucrose is an attractive feeding substance and a positive reinforcer for Drosophila. But Drosophila females have been shown to robustly reject a sucrose-containing option for egg-laying when given a choice between a plain and a sucrose-containing option in specific contexts. How the sweet taste system of Drosophila promotes context-dependent devaluation of an egg-laying option that contains sucrose, an otherwise highly appetitive tastant, is unknown. This study reports that devaluation of sweetness/sucrose for egg-laying is executed by a sensory pathway recruited specifically by the sweet neurons on the legs of Drosophila First, silencing just the leg sweet neurons caused acceptance of the sucrose option in a sucrose versus plain decision, whereas expressing the channelrhodopsin CsChrimson in them caused rejection of a plain option that was "baited" with light over another that was not. Analogous bidirectional manipulations of other sweet neurons did not produce these effects. Second, circuit tracing revealed that the leg sweet neurons receive different presynaptic neuromodulations compared to some other sweet neurons and were the only ones with postsynaptic partners that projected prominently to the superior lateral protocerebrum (SLP) in the brain. Third, silencing one specific SLP-projecting postsynaptic partner of the leg sweet neurons reduced sucrose rejection, whereas expressing CsChrimson in it promoted rejection of a light-baited option during egg-laying. These results uncover that the Drosophila sweet taste system exhibits a functional division that is value-based and task-specific, challenging the conventional view that the system adheres to a simple labeled-line coding scheme (Chen, 2022).

Swallowing is an essential step of eating and drinking. However, how the quality of a food bolus is sensed by pharyngeal neurons is largely unknown. This study finds that mechanical receptors along the Drosophila pharynx are required for control of meal size, especially for food of high viscosity. The mechanical force exerted by the bolus passing across the pharynx is detected by neurons expressing the mechanotransduction channel NOMPC (no mechanoreceptor potential C) and is relayed, together with gustatory information, to IN1 neurons in the subesophageal zone (SEZ) of the brain. IN1 (ingestion neurons) neurons act directly upstream of a group of peptidergic neurons that encode satiety. Prolonged activation of IN1 neurons suppresses feeding. IN1 neurons receive inhibition from DSOG1 (descending subesophageal neurons) neurons, a group of GABAergic neurons that non-selectively suppress feeding. These results reveal the function of pharyngeal mechanoreceptors and their downstream neural circuits in the control of food ingestion (Yang, 221).

Feeding is an essential part of animal life that is greatly impacted by the sense of taste. Although the characterization of taste-detection at the periphery has been extensive, higher order taste and feeding circuits are still being elucidated. This study used an automated closed-loop optogenetic activation screen to detect novel taste and feeding neurons in Drosophila melanogaster. Out of 122 Janelia FlyLight Project GAL4 lines preselected based on expression pattern, this study identified six lines that acutely promote feeding and 35 lines that inhibit it. As proof of principle, R70C07-GAL4, which labels neurons that strongly inhibit feeding, was analyzed. Using split-GAL4 lines to isolate subsets of the R70C07-GAL4 population, both appetitive and aversive neurons were found. Furthermore, this study shows that R70C07-GAL4 labels putative second-order taste interneurons in the subesophageal zone that contact both sweet and bitter sensory neurons. These results serve as a resource for further functional dissection of fly feeding circuits (Lau, 2021).

This study reconstructed, from a whole CNS EM volume, the synaptic map of input and output neurons that underlie food intake behavior of Drosophila larvae. Input neurons originate from enteric, pharyngeal and external sensory organs and converge onto seven distinct sensory synaptic compartments within the CNS. Output neurons consist of feeding motor, serotonergic modulatory and neuroendocrine neurons. Monosynaptic connections from a set of sensory synaptic compartments cover the motor, modulatory and neuroendocrine targets in overlapping domains. Polysynaptic routes are superimposed on top of monosynaptic connections, resulting in divergent sensory paths that converge on common outputs. A completely different set of sensory compartments is connected to the mushroom body calyx. The mushroom body output neurons are connected to interneurons that directly target the feeding output neurons. These results illustrate a circuit architecture in which monosynaptic and multisynaptic connections from sensory inputs traverse onto output neurons via a series of converging paths (Miroschnikow, 2018).

Motor outputs of a nervous system can be broadly defined into those carried out by the muscles to produce movements and by the glands for secretion. Both of these behavioral and physiological events are regulated by a network of output neurons, interneurons and sensory neurons, and a major open question is how one neural path is selected from multiple possible paths to produce a desired output. Nervous system complexity and tool availability have strongly dictated the type of experimental system and analysis that can be used to address this issue, such as a focus on a particular organism, behavior or type of neuron. In this context, the detailed illustrations of different parts of nervous systems at neuronal level as pioneered by Cajal, to the first complete description of a nervous system wiring diagram at synaptic level for C. elegans, demonstrate the power of systematic neuroanatomical analysis in providing a foundation and guide for studying nervous system function. However, the technical challenges posed by such analysis have limited the type of organisms for which synaptic resolution mapping can be performed at the scale of an entire nervous system (Miroschnikow, 2018).

Analysis of the neural circuits that mediate food intake in the Drosophila larvae offers numerous advantages in meeting the challenge of neuroanatomical mapping at a whole brain level, and combining it with the ability to perform behavioral and physiological experiments. The muscle system that generates the different movements necessary for transporting food from the pharynx to the esophagus, as well as the endocrine system responsible for secreting various hormones for metabolism and growth, have both been well described. These are also complemented by the analysis of feeding behavior in adult flies. Although there is broad knowledge at the morphological level on the organs underlying larval feeding behavior and physiology, as well as on the nerves innervating them in the periphery, the central connectivity of the afferent and efferent neurons within these nerves are largely unknown. At the same time, advances in the EM reconstruction of an entire CNS of a first instar larva offers an opportunity to elucidate an animals' feeding system on a brain-wide scale and at synaptic resolution. As part of this community effort, an integrated analysis of fast synaptic and neuropeptide receptor connections for an identified cluster of 20 interneurons that express the neuropeptide hugin, a homolog of the mammalian neuropeptide neuromedin U, and which regulates food intake behavior. This analysis showed that the class of hugin neurons modulating food intake receives direct synaptic inputs from a specific group of sensory neurons, and in turn, makes mono-synaptic contacts to output neuroendocrine cells. The study not only provided a starting point for a combined approach to studying synaptic and neuropeptidergic circuits, but a basis for a comprehensive mapping of the sensory and output neurons that innervate the major feeding and endocrine organs (Miroschnikow, 2018).

Feeding is one of the most universal and important activities that animals engage in. Despite large differences in the morphology of the external feeding organs, the internal gut structures are quite similar across different animals; indeed, even within closely related species, there can be large differences in the external organs that detect and gather food, whereas the internal organs that transport food through the alimentary canal are much more similar. Recent studies have also pointed out the functional similarities between the subesophageal zone in insects and the brainstem in vertebrates for regulating feeding behavior. In mammals, the different cranial nerves from the medulla innervate distinct muscles and glands of the foregut (Figure 1A). For example, the VIIth cranial nerve (facial nerve) carries taste sensory information from anterior 2/3 of the tongue, and innervates the salivary glands, and lip and facial muscles. The IXth cranial nerve (glossopharyngeal nerve) receives taste inputs from the posterior 1/3 of the tongue, and innervates the salivary glands and pharynx muscles. The Xth cranial nerve (vagus nerve) receives majority of the sensory inputs from the enteric nervous system of the gut, and innervates pharynx and esophagus muscles. The XIth cranial nerve (spinal accessory nerve) and the XIIth cranial nerve (hypoglossal nerve) are thought to carry strictly motor information which innervate the pharynx and neck muscles, and the tongue muscles. The distinct cranial nerves project onto topographically distinct areas in the medulla of the brainstem. It is also noted that olfactory information is carried by cranial nerve I, a strictly sensory nerve that projects to the olfactory bulb (OB), an area topographically distinct from the brainstem. In addition, there are direct neuronal connections between the brainstem and the hypothalamus, the key neuroendocrine center of vertebrates (Miroschnikow, 2018).

Analogously, distinct pharyngeal nerves of the Drosophila larva are connected to the subesophageal zone (SEZ), and also carry sensory and motor information that regulate different parts of the body. The AN (antennal nerve) carries sensory information from the olfactory, pharyngeal and internal organs, and innervates the pharyngeal muscles for pumping in food. The serotonergic neurons that innervate the major endocrine center and the enteric nervous system also project through the AN. Note also that the olfactory sensory organs project to the antennal lobe (AL), which abuts the SEZ yet is topographically separate. The MxN (maxillary nerve) carries external and pharyngeal sensory information, and innervates the mouth hooks, whose movements are involved in both feeding and locomotion. The PaN (prothoracic accessory nerve) carries external sensory information from the upper head region, and innervates the muscles involved in head tilting (see EM reconstruction of the pharyngeal nerves of Drosophila larva). Furthermore, the SEZ has direct connections to median neurosecretory cells (mNSCs) and the ring gland. In sum, although a large body of knowledge exists on the gross anatomy of the nerves that target the feeding organs in vertebrates and invertebrates, the synaptic pathways within the brain that interconnect the sensory inputs and output neurons of the individual nerves remain to be elucidated (Miroschnikow, 2018).

This study has reconstructed all sensory, serotonergic modulatory (Se0) and motor neurons of the three pharyngeal nerves that underlie the feeding motor program of Drosophila larvae. The activity of these nerves has previously been shown to be sufficient for generating the feeding motor pattern in isolated nervous system preparations, and that the central pattern generators (CPGs) for food intake lie in the SEZ. All monosynaptic connections were identified between the sensory inputs and the motor, Se0 and previously described median neurosecretory ouput neurons, thus providing a full monosynaptic reflex circuit for food intake. Polysynaptic pathways that are integrated onto the monosynaptic reflex circuits. In addition, the multisynaptic non-olfactory neuron connections from the sensory neurons to the mushroom body memory circuit were also mapped, and these were shown to be different from those involved in monosynaptic reflex circuits. Finally, a set of mushroom body output neurons were traced onto the neurosecretory and other feeding output neurons. Reflex circuits can be seen to represent the simplest synaptic architecture in the nervous system, as formulated by Charles Sherrington. Anatomical reconstructions of monosynaptic and polysynaptic reflex circuits can also be seen in the works of Cajal. A model is proposed of how different mono- and polysynaptic pathways can be traversed from a set of sensory neurons to specific output neurons, which has relevance for understanding the mechanisms of action selection (Miroschnikow, 2018).

It was asked how the hugin neuropeptide circuit, which relays gustatory information to the protocerebrum, would be positioned with respect to the monosynaptic reflex and multisynaptic MB memory circuits. Based on earlier studies on mapping sensory inputs onto hugin protocerebrum neurons (huginPC), it was expected that most inputs from the ACp, which is the primary gustatory sensory compartment. However, most of the huginPC neurons receive inputs from the sensory compartments ACa and AVa, which are the two major monosynaptic compartments that originate from enteric regions. HuginPC neurons do receive inputs from the external and pharyngeal organs (i.e., through sensory compartment ACp), but to a much smaller degree. Thus, unlike the MB circuit that utilizes a completely new set of sensory inputs, the huginPC circuit is associated with a feeding related monosynaptic circuit (Miroschnikow, 2018).

Based on these observations from the hugin neuropeptide circuit in interconnecting sensory and neuroendocrine outputs, a broader question concerning input-output connections was asked: for any given pair of neurons comprising the monosynaptic reflex circuit, how many additional polysynaptic paths exist and what could be the functional significance of such parallel pathways? To illustrate, a target neuron was selected from a cluster of neurosecretory and serotonergic modulatory output cells (Dilps and Se0ens), and all sensory neurons that make monosynaptic connections with at least two synapses were listed listed. It was then asked, using the same threshold, how many different di-synaptic paths (2-hop) exist and how often a particular interneuron is used for the different possible converging paths ('degree' of convergence). The relative synaptic strengths of the connection among the various paths ('ranking index' of 1.0 represents highest synaptic strength from all possible inputs to the output neuron). Several properties are revealed: (1) different sensory neurons make monosynaptic contacts to a common output target (2) each output neuron can be reached from a given sensory neuron by multiple routes through the use of different interneurons (3) a given interneuron can receive inputs from different sensory neurons to target the same output neuron; this would fit the definition of the ‘common path’ that Sherrington described. These observations hold true for the majority of monosynaptic sensory-output pairs examined. However, no correlation was seen between the relative synaptic strength and the commonness of the respective paths (i.e., how often a path is used) (see Integration of polysynaptic connections onto monosynaptic circuits) (Miroschnikow, 2018).

This study provides a comprehensive synaptic map of the sensory and output neurons that underlie food intake and metabolic homeostasis in Drosophila larva. Seven topographically distinct sensory compartments, based on modality and peripheral origin, subdivide the SEZ, a region with functional similarities to the vertebrate brainstem. Sensory neurons that form monosynaptic connections are mostly of enteric origin, and are distinct from those that form multisynaptic connections to the mushroom body (MB) memory circuit. Different polysynaptic connections are superimposed on the monosynaptic input-ouput pairs that comprise the reflex arc. Such circuit architecture may be used for controlling feeding reflexes and other instinctive behaviors (Miroschnikow, 2018).

Reflex circuits represent a basic circuit architecture of the nervous system, whose anatomical and physiological foundations were laid down by Cajal and Sherrington. The Drosophila larval feeding reflex circuit comprises the motor neurons that innervate the muscles involved in pharyngeal pumping, as well as the neurosecretory neurons that target the endocrine organs. They also include a cluster of serotonergic neurons that innervate the entire enteric nervous system, and which may have neuromodulatory effects on the feeding system in a global manner. The vast majority of output neurons are targeted monosynaptically from a set of topographically distinct sensory synaptic compartments in the CNS. These compartments target the output neurons in overlapping domains: the first, ACa, targets all neuroendocrine cells as well as the serotonergic neurons; the second, AVa, targets a subset of neuroendocrine cells, the serotonergic neurons and most of the pharyngeal motor neurons, while the third, AVp, targets the serotonergic neurons and a different set of pharyngeal motor neurons. With these outputs, one can in principle fulfill the most basic physiological and behavioral needs for feeding: neurosecretory cells for metabolic regulation and pharyngeal motor neurons for food intake. This set of monosynaptic connections can thus be seen to represent an elemental circuit for feeding, since the connections between the input and output neurons cannot be broken down any further (Miroschnikow, 2018).

Vast majority of the sensory inputs comprising this 'elemental feeding circuit' derive from the enteric nervous system to target the pharyngeal muscles involved in food intake and neuroendocrine output organs. However, there is a small number of monosynaptic reflex connections that originate from the somatosensory compartment. The output neurons targeted by these somatosensory neurons are motor neurons that control mouth hook movements and head tilting, movements which are involved in both feeding and locomotion. In this context, it is noteworthy that monosynaptic reflex connections are found to a much lesser degree in the larval ventral nerve cord, which generates locomotion. An analogous situation exists in C. elegans, where majority of the monosynaptic reflex circuits are found in the head motor neurons and not in the body. One reason could be due to the relative complexity in the response necessary for food intake as compared to locomotion. For example, a decision to finally not to swallow a harmful substance, once in the mouth, may require a more local response, for example muscles limited to a very specific region of the pharynx and esophagus, where monosynaptic arc might suffice. By contrast, initiating escape behaviors requires a more global response with respect to the range and coordination of body movements involved, although it also employs multimodal sensory integration via a multilayered circuit (Miroschnikow, 2018).

The inter-sensory connections show a combination of hierarchical and reciprocal connections, which may increase the regulatory capability and could be especially important for monosynaptic circuits. By contrast, very few monosynaptic connections exist between the larval olfactory, chordotonal or nociceptive class IV sensory neurons in the body. Interestingly, there is also a much higher percentage of intersensory connections between olfactory receptor neurons in the adult as compared to the larva, which could function in gain modulation at low signal intensities. This might be attributable to adults requiring faster processing of olfactory information during flight navigation (or mating), and/or to minimize metabolic cost. Whether such explanation also applies to the differences in intersensory connection between the different types of sensory neurons in the larvae remains to be determined (Miroschnikow, 2018).

This study found very few cases where a monosynaptic path between any sensory-output pair is not additionally connected via a polysynaptic path. An interesting question in the context of action selection mechanism is which path a sensory signal uses to reach a specific target neuron. For example, a very strong sensory signal may result in a monosynaptic reflex path being used. However, a weaker sensory signal may result in using a different path, such as one with less threshold for activation. This would also enable the integration of different types of sensory signals through the usage of multiple interneurons, since the interneurons may receive sensory inputs that are not present in monosynaptic connections. For example, sensory neurons can target the neuroendocrine cells directly (monosynaptically), or through a hugin interneuron (di-synaptically). The sensory compartments that directly target the neuroendocrine cells are of enteric origin; however, when hugin neurons are utilized as interneurons, not only is the number of sensory neurons from the same sensory compartment increased, but sensory neurons are added from a completely new peripheral origin. Thus, the hugin interneurons enable sensory inputs from different peripheral origins, for example to integrate enteric inputs with pharyngeal gustatory inputs, to influence an output response, which, in this case, is to stop feeding (Schoofs et al., 2014a) (Miroschnikow, 2018).

The coexistence of polysynaptic and monosynaptic paths could also be relevant for circuit variability and compensation: destruction of any given path would still enable the circuit to function, but with more restrictions on the precise types of sensory information it can respond to. In certain cases, this may even lead to strengthening of alternate paths as a form of synaptic plasticit (Miroschnikow, 2018).

An open issue is how the sensory synaptic compartments might be connected to the feeding central pattern generators (CPGs) which have been demonstrated to exist in the SEZ, especially since CPGs are defined as neural circuits that can generate rhythmic motor patterns in the absence of sensory input. However, the modulation of CPG rhythmic activity can be brought about by sensory and neuromodulatory inputs. A complete circuit reconstruction of the larval SEZ circuit may shed some light on the circuit structure of feeding CPGs (Miroschnikow, 2018).

A more complex circuit architecture is represented by the MB, the site of associative learning and memory in insects: a completely different set of sensory synaptic compartments is used to connect the various projection neurons to the MB calyx. Thus, the MB module is not superimposed onto the monosynaptic reflex circuits but rather forms a separate unit. The classical studies by Pavlov demonstrated conditioned reflex based on an external signal and an autonomic secretory response in response to food. Although a comparable autonomic response has not been analyzed in the larvae, analogous associative behavior based on odor choice response has been well studied. It is also noteworthy that in the Aplysia, classical conditioning of the gill withdrawal reflex involves monosynaptic connections between a sensory neuron (mechanosensory) and a motor neuron, and neuromodulation by serotonin. This constellation has similarities with the elemental feeding circuit consisting of sensory, motor and serotonergic modulatory neurons. For more complex circuits of feeding behavior in the mouse, a memory device for physiological state, such as hunger, has been reported involving synaptic and neuropeptide hormone circuits. Functional studies on MB output neurons such as the MBON-f1, which may be part of a ‘psychomotor’ pathway and which targets a number of interneurons that connect to the neurosecretory, serotonergic and pharyngeal motor neurons, may help address how memory circuits interact with feeding circuits (Miroschnikow, 2018).

Feeding behavior manifests itself from the most primitive instincts of lower animals, to deep psychological and social aspects in humans. It encompasses cogitating on the finest aspects of food taste and the memories evoked by the experience, to sudden reflex reactions upon unexpectedly biting down on a hard seed or shell. Both of these extremes are mediated, to a large degree, by a common set of feeding organs, but the way these organs become utilized can vary greatly. The architecture of the feeding circuit described in this study allows the various types of sensory inputs to converge on a limited number of output responses. The monosynaptic pathways would be used when fastest response is needed. The presence of polysynaptic paths would enable slower and finer control of these reflex events by allowing different sensory inputs, strengths or modalities to act on the monosynaptic circuit. This can be placed in the context in the control of emotions and survival circuits, or by cortex regulation of basic physiological or autonomic processes. In a striking example, pupil dilation, a reflex response, has been used as an indicator of cognitive activity. Here, a major function of having more complex circuit modules on top of monosynaptic circuits may be to allow a finer regulation of feeding reflexes, and perhaps of other reflexes or instinctive behaviors (Miroschnikow, 2018).

As an outlook, this analysis provides an architectural framework of how a feeding circuit is organized in the CNS. The circuit is divided into two main axes that connect the input to the output systems: the sensory-neurosecretory cell axis and the sensory-motor neuron axis. The sensory system targets overlapping domains of the output neurons; for example, a set of sensory neurons targets exclusively the neuroendocrine cells, other targets both neuroendocrine and pharyngeal motor neurons, and another just the pharyngeal motor neurons. The inputs derive mostly from the internal organs. These connections form the monosynaptic reflex circuits. With these circuits, one can perform the major requirements of feeding regulation, from food intake and ingestion to metabolic homeostasis. Additional multisynaptic circuits, such as the CPGs, those involving sensory signaling from the somatosensory system (external inputs), or those comprising the memory circuits, are integrated or added to expand the behavioral repertoire of the animal see Input-output synaptic organization of the larval feeding system and its connectivity architecture in the brain. Although circuit construction may proceed from internal to the external, the sequence is reversed in a feeding animal: the first sensory cues are external (olfactory), resulting in locomotion (somatic muscles) that can be influenced by memory of previous experience; this is followed by external taste cues, resulting in food intake into the mouth; the final action is the swallowing of food, involving pharyngeal and enteric signals and reflex circuits. However, regardless of the types of sensory inputs, and whether these are transmitted through a reflex arc, a memory circuit or some other multisynaptic circuits in the brain, they will likely converge onto a certain set of output neurons, what Sherrington referred to as the ‘final common path’. The current work is a first step towards finding the common paths (Miroschnikow, 2018).

The decision to engage in one behavior often precludes the selection of others, suggesting cross-inhibition between incompatible behaviors. For example, the likelihood to initiate feeding might be influenced by an animal's commitment to other behaviors. This study examined the modulation of feeding behavior in the fruit fly, Drosophila melanogaster, and identified a pair of interneurons in the ventral nerve cord that is activated by stimulation of mechanosensory neurons and inhibits feeding initiation, suggesting that these neurons suppress feeding while the fly is walking. Conversely, inhibiting activity in these neurons promotes feeding initiation and inhibits locomotion. These studies demonstrate the mutual exclusivity between locomotion and feeding initiation in the fly, isolate interneurons that influence this behavioral choice, and provide a framework for studying the neural basis for behavioral exclusivity in Drosophila (Mann, 2013).

The neurons that inhibit proboscis extension (which are named PER_in) have cell bodies and processes in the first leg neuromeres of the VNC and projections to the SOG, the brain region that contains gustatory sensory axons and proboscis motor neuron dendrites. Labeling with the presynaptic synaptotagmin- GFP marker and the postsynaptic DenMark marker indicated that the dendrites of PER_in neurons are restricted to the first leg neuromeres, whereas axons are found in both the SOG and the first leg neuromeres. The anatomy of these neurons suggests that they convey information from the leg neuromeres to a region of the fly brain involved in gustatory processing and proboscis extension. Anatomical studies examining the proximity of PER_in fibers to gustatory sensory dendrites or proboscis motor axons revealed that PER_in neurons do not come into close contact with known neurons that regulate proboscis extension (Mann, 2013).

Many behaviors are mutually exclusive, with the decision to commit to one behavior excluding the selection of others. This study shows that feeding initiation and locomotion are mutually exclusive behaviors and that activity in a single pair of interneurons influences this behavioral choice. PER_in neurons are activated by stimulation of mechanosensory neurons and activation of PER_in inhibits proboscis extension, suggesting that they inhibit feeding while the animal is walking. Consistent with this, leg removal or immobilization enhances proboscis extension probability and this is inhibited by increased PER_in activity. The opposite behavior is elicited upon inhibiting activity in PER_in neurons: animals show constitutive proboscis extension at the expense of locomotion. This work shows that activity in a single pair of interneurons dramatically influences the choice between feeding initiation and movement (Mann, 2013).

The precise mechanism of activation of PER_in neurons remains to be determined. PER_in dendrites reside in the first leg neuromere, suggesting that they process information from the legs. Stimulation of leg chemosensory bristles with sucrose or quinine or activation of sugar, bitter, or water neurons using optogenetic approaches did not activate PER_in neurons, nor did satiety state change tonic activity. Stimulation of sensory nerves into the ventral nerve cord and stimulation of mechanosensory neurons, using a nompC driver, activated PER_in. In addition, by monitoring activity of PER_in while flies moved their legs, it was demonstrated that activity was coincident with movement (Mann, 2013).

These studies argue that PER_in is activated by nongustatory cues in response to movement, likely upon detection of mechanosensory cues. Additional cues may also activate PER_in. Studies of behavioral exclusivity in other invertebrate species suggest two mechanisms by which one behavior suppresses others. One strategy is by competition between command neurons that activate dedicated circuits for different behaviors. More common is a strategy in which decision- making occurs by distributed activity changes across neural populations. Although this studies are a starting point to begin to examine these models in Drosophila, the circuits for proboscis extension and locomotion drive different motor neurons, muscles, and behaviors, suggesting that they may be connected by a few links rather than largely overlapping circuitry. PER_in is likely to inhibit feeding initiation while the animal is moving and is one critical link. The observation that simply gluing the proboscis in an extended state, but not in a retracted state, inhibits locomotion suggests that motor activity or proprioceptive feedback from the proboscis acts as a reciprocal link to locomotor circuits (Mann, 2013).

Neurons act over different timescales and in response to different sensory cues to influence behavior. The powerful molecular genetic approaches available in Drosophila enable the precise manipulation of individual neurons and allow for the examination of their function in awake, behaving animals. Modulatory neurons such as PER_in are difficult to identify by calcium imaging or electrophysiological approaches because they influence gustatory-driven behavior but are not activated by gustatory stimulation. The ability to probe the function of neurons in unbiased behavioral screens facilitates the identification of neurons that act as critical nodes to influence behavior. The identification and characterization of PER_in as a significant modulator of feeding initiation provides a foundation for future studies determining how PER_in influences proboscis extension circuits to alter behavioral probability and how mechanosensory inputs activate PER_in. In addition, examining how proboscis extension suppresses locomotion will provide important insight into the links between different behaviors (Mann, 2013).

Neural circuits for a given behavior do not work in isolation. Information from multiple sensory cues, physiological state, and experience must be integrated to guide behavioral decisions. This work uncovers a pair of interneurons that influences the choice between feeding initiation and locomotion. The discovery of the PER_in neurons will aid in examining the neural basis of innate behaviors and the decision-making processes that produce them (Mann, 2013).

Odor-based learning and innate odor-driven behavior have been hypothesized to require separate neuronal circuitry. Contrary to this notion, innate behavior and olfactory learning were recently shown to share circuitry that includes the Drosophila mushroom body (MB). But how a single circuit drives two discrete behaviors remains unknown. This study defines an MB circuit responsible for both olfactory learning and innate odor avoidance and the distinct dDA1 dopamine receptor-dependent signaling pathways that mediate these behaviors. Associative learning and learning-induced MB plasticity require rutabaga-encoded adenylyl cyclase activity in the MB. In contrast, innate odor preferences driven by naive MB neurotransmission are rutabaga independent, requiring the adenylyl cyclase ACXD. Both learning and innate odor preferences converge on PKA and the downstream MBON-γ2α'1. Importantly, the utilization of this shared circuitry for innate behavior only becomes apparent with hunger, indicating that hardwired innate behavior becomes more flexible during states of stress (Noyes, 2023).

The data reveal the shared use of a discrete circuit for both state-dependent odor-driven behavior and experience-dependent odor learning. The shared components include the upstream DA neurons, the MBn-expressed DA receptor dDA1, and the downstream MBON. Odor response processing for state-dependent behavior and odor learning diverge at the level of the dDA1 receptor-activated adenylyl cyclase, with ACXD employed for innate state-dependent odor driven behavior and rut employed for olfactory learning. The unique activation of rut for olfactory learning is explained by the fact that this adenylyl cyclase functions as a coincidence detector, synergistically responding to both DA receptor activation from the unconditioned stimulus and Ca2+ increases due to the conditioned stimulus. ACXD is a transmembrane AC that is expressed in a number of tissues including the brain and is orthologous to the mammalian AC2. Mammalian AC2 activity is Ca2+ independent. If ACXD is also Ca2+ independent, it would provide a mechanism for the engagement of distinct cAMP pathways by dDA1 for state-dependent versus experience-dependent olfactory behavior. Thus, common neural circuitry is employed for both state-dependent and conditioned behaviors with the unique changes of MBn output influenced by the intracellular signaling pathways that are mobilized (Noyes, 2023).

Dopaminergic input to the dDA1 DA receptor expressed in the γ2 compartment of mushroom body neurons (MBn) activates an intracellular signaling pathway that includes the ACXD adenylyl cyclase, PKA activity, and the release of ACh. The downstream MBON-γ2α'1 responds to the MB ACh release through the α2 nACh receptor, with the activity of the MBON-γ2α'1 ultimately dictating the balance in state-dependent odor approach/avoidance. The simplest model to account for the state-dependent MBON activity would have the internal state modulating DA input into the MBn to increase or decrease ACh release onto the MBON. However, the data failed to detect a significant change in ACh release between the fed and starved conditions. Nevertheless, the activity of the PPL1-γ2α'1 (PPL referring to protocerebral posterior lateral region of the brain) that influences the MB γ2 compartment is required for state-dependent behavioral responses to odor. A proposed model for reconciling these observations envisions that the basal activity of this circuit is required for behavioral-state odor choice but that starvation mobilizes a qualitative or quantitative signal independent of the magnitude ACh release by the MBn to increase MBON activity. An unidentified signal representing hunger could directly enhance MBON excitability. For example, octopamine has been proposed as a feeding signal that acts directly on MBONs (Noyes, 2023).

Similarly, a hunger signal could act on neurons elsewhere in the brain that ultimately connect to MBONs through intermediary neurons. The hunger-responsive neuropeptide leucokinin acts on DAns that connect to MBONs (Noyes, 2023).

Alternatively, there may be a co-neurotransmitter released by the MBn due to starvation that works to increase MBON activity. Finally, the possibility that starvation does modulate MBn ACh output is left open, but the reporters employed lack the sensitivity to detect this change. Future investigations into state-dependent changes in MBON-γ2α'1 physiology will need to grapple with numerous competing hypotheses (Noyes, 2023).

Changes in odor responses in MBONs have suggested that learning induces a change in connectivity in the MBn-MBON synapse. In addition, compartment specific plasticity in MB ACh release was discovered that fits with the idea that plasticity observed in MBONs occurs from the input of MB compartments. However, there has been a lack of data connecting the MBn ACh release plasticity with MBON plasticity and particularly to the central role for the rut adenylyl cyclase. The data offer this important connection. The MB γ2 compartment undergoes a rapid depression in response to odor/shock pairings during aversive learning and that rut is required for the acquisition of this depression. Downstream of the MB γ2 compartment, MBON-γ2α'1 drives approach and also undergoes a learning-induced depression (Noyes, 2023).

These results put prior speculation about how the genetic regulation of cAMP signaling through the rut adenylyl cyclase drives Drosophila memory on concrete ground. This work does not conclusively delineate a role for dDA1 in the MB plasticity. Loss of MB dDA1 dramatically reduced naive odor responses in MB γ2. This precluded attempts to measure dDA1 effects on MB γ2 depression because the naive responses were already low. Interestingly, both learning- and starvation-dependent odor avoidance require PKA. A likely explanation is that rut and ACXD are spatially segregated, creating distinct cAMP microdomains or signaling platforms (Noyes, 2023).

Thus, PKA activity would result in the phosphorylation of unique substrates within those microdomains. The characterization of the MB as a brain region for learned, but not innate, olfactory behavior was motivated by experiments eliminating MBn or blocking MB output. Disrupting the MB eliminates odor-associated memory but has no effect on innate avoidance of those same odors. Recent work has overturned this simple categorization demonstrating some DANs, MBONs, and MBns do contribute to innate olfactory behavior in certain circumstances. Interestingly, the majority of reports define a role for MBns in innate behavioral responses to food-related odors. The current results, using more general, non-food odors, puts the hypothesis that MBn regulates innate behavior on more solid ground. Importantly, MB dDA1 was found to be required for state-dependent behavior to general odors. This is in contrast to a report finding that dDA1 is not involved but that DAMB is required for state-dependent behavior to food odors. This difference will be a key element to understand state-dependent behavior moving forward. The Drosophila and mammalian olfactory systems are remarkably similar in terms of anatomical organization and function (Noyes, 2023).

In both, odorant molecules activate olfactory sensory neurons (OSNs), with each OSN only expressing one type of odorant receptor (OR). Each OSN expressing the same type of OR project to the same glomeruli. Within the glomeruli, the OSNs synapse onto projection neuron (PN) dendrites, and PN activity is modified in the glomeruli by local inhibitory interneurons before being sent on to multiple higher-order brain regions. PN neurons connect to downstream neurons in the mammalian piriform cortex and in the Drosophila MB in a seemingly random manner. Like the Drosophila MB, the piriform cortex is critical for olfactory memory. It is not clear how the piriform cortex is involved in state-dependent olfactory behavior. However, in humans, odor coding changes in the piriform cortex with hunger and sleep deprivation, and piriform cortex neuron activity levels are inversely correlated with sexual satiety in rats (Noyes, 2023).

It is concluded from the results demonstrating dDA1-dependent MBn Ach release and a dDA1-dependent MBON-γ2&alpha'1 Ca2+ in response to odor that dDA1 directly modulates the MBn/MBON connection. However, due to a limitation in the sensitivity of the ACh sensor employed, it was not possible to directly record ACh input to MBON-γ2&alpha'1. Based on the established direct cholinergic connection between these MBn and MBONs and the lack of any known non-MB cholinergic innervation to this brain region, it is believed that the conclusions are merited. However, it is necessary to leave open as a formal possibility that other intermediary neurons mediate this relationship (Noyes, 2023).

The sensitivity of animals to sensory input must be regulated to ensure that signals are detected and also discriminable. However, how circuits regulate the dynamic range of sensitivity to sensory stimuli is not well understood. A given odor is represented in the insect mushroom bodies (MBs) by sparse combinatorial coding by Kenyon cells (KCs), forming an odor quality representation. To address how intensity of sensory stimuli is processed at the level of the MB input region, the calyx, this study characterized a set of novel mushroom body output neurons that respond preferentially to high odor concentrations. A pair of MB calyx output neurons, MBON-a1/2, were shown to be postsynaptic in the MB calyx, where they receive extensive synaptic inputs from KC dendrites, the inhibitory feedback neuron APL, and octopaminergic sVUM1 neurons, but relatively few inputs from projection neurons. This pattern is broadly consistent in the third-instar larva as well as in the first instar connectome. MBON-a1/a2 presynaptic terminals innervate a region immediately surrounding the MB medial lobe output region in the ipsilateral and contralateral brain hemispheres. By monitoring calcium activity using jRCamP1b, it was found that MBON-a1/a2 responses are odor-concentration dependent, responding only to ethyl acetate (EA) concentrations higher than a 200-fold dilution, in contrast to MB neurons which are more concentration-invariant and respond to EA dilutions as low as 10^-4. Optogenetic activation of the calyx-innervating sVUM1 modulatory neurons originating in the SEZ (Subesophageal zone), did not show a detectable effect on MBON-a1/a2 odor responses. Optogenetic activation of MBON-a1/a2 using CsChrimson impaired odor discrimination learning compared to controls. It is proposed that MBON-a1/a2 form an output channel of the calyx, summing convergent sensory and modulatory input, firing preferentially to high odor concentration, and might affect the activity of downstream MB targets (Mohamed, 2023).

Animals retain some but not all experiences in long-term memory (LTM). Sleep supports LTM retention across animal species. It is well established that learning experiences enhance post-learning sleep. However, the underlying mechanisms of how learning mediates sleep for memory retention are not clear. Drosophila males display increased amounts of sleep after courtship learning. Courtship learning depends on Mushroom Body (MB) neurons, and post-learning sleep is mediated by the sleep-promoting ventral Fan-Shaped Body neurons (vFBs). This study shows that post-learning sleep is regulated by two opposing output neurons (MBONs) from the MB, which encode a measure of learning. Excitatory MBONs-γ2α'1 becomes increasingly active upon increasing time of learning, whereas inhibitory MBONs-β'2mp is activated only by a short learning experience. These MB outputs are integrated by SFS neurons, which excite vFBs to promote sleep after prolonged but not short training. This circuit may ensure that only longer or more intense learning experiences induce sleep and are thereby consolidated into LTM (Lei, 2022).

This study has identified a neural circuit that regulates learning-induced sleep for LTM consolidation. This circuit links neurons essential for learning and memory in Drosophila, the MB neurons, with those critical for post-learning sleep, the vFBs8. It is proposed that only a longer learning experience is sufficient to induce sleep, and thereby be consolidated into LTM. Given that the increasing duration of a learning experience correlates with the total amount of time males spend on futile courtship towards mated females during training, selective activation of vFBs likely depends on the amount or intensity of a learning experience, rather than just its duration. Post-learning sleep induction requires integration of two MB outputs, previously implicated in courtship memory in SFSs. Post-learning activity of the excitatory MBONs-γ2α'1 increases linearly with the duration of the prolonged learning experience. In contrast, activity of the inhibitory MBONs-β'2mp peaks after a short experience sufficient to induce STM. As a result, only when the males court mated females sufficiently long or intensely, the activity of MBONs-γ2α'1 reaches the threshold required to activate SFSs. This in turn leads to activation of vFBs to promote post-learning sleep and the reactivation of those dopaminergic neurons (DANs) that were involved in memory encoding. Consequently, biochemical processes essential for LTM consolidation become engaged (Lei, 2022).

How might MBONs-γ2α'1 and MBONs-β'2mp measure the learning experience to control post-learning sleep? In homeostatic sleep regulation, the potentiation of R2 neurons reflects a measure of sleep loss that is sensed by dFBs, likely in response to the accumulation of byproducts of oxidative stress during sleep loss. In the case of learning-induced sleep, it is envisioned that learning results in lasting changes in the molecular pathways essential for memory formation in the MB. For example, the cAMP pathway along with the dopamine receptor are activated during sleep in a discrete 3-h time window after learning in rodents and Drosophila males lacking a dopamine receptor, and hence unable to learn, do not display increased post-learning sleep. Thus, the accumulation of changes in the cAMP signaling pathway upon increasing learning experience with mated females might lead to the increasing potentiation of MBONs-γ2α'1 and MBONs-β'2mp after learning. Interestingly, MBONs-γ2α'1 and MBONs-β'2mp display distinct temporal activity patterns upon learning which likely reflects their distinct neuronal properties (Lei, 2022).

This study reveals a circuit mechanism that ensures that only persistent, and thus likely significant, learning experiences generate post-learning sleep to consolidate LTM. Recent findings suggest that dFBs, involved in sleep homeostasis, might mediate a paradoxical type of sleep, in humans also called Rapid Eye Movement (REM) sleep. This in conjunction with the current data, provide an opportunity to investigate whether the post-learning sleep, mediated by vFBs, might represent another type of sleep implicated in mammals in memory consolidation (Lei, 2022).

Dynamic functional connectivity within brain circuits requires coordination of intercellular signaling and intracellular signal transduction. Critical roles for cAMP-dependent protein kinase A (PKA) signaling are well established in the Drosophila mushroom body (MB) learning and memory circuitry, but local PKA activity within this well-mapped neuronal network is uncharacterized. This study use an in vivo PKA activity sensor (PKA-SPARK) to test spatiotemporal regulatory requirements in the MB axon lobes. Immature animals had little detectable PKA activity, whereas postcritical period adults showed high field-selective activation primarily in just 3/16 defined output regions. In addition to the age-dependent PKA activity in distinct α'/β' lobe nodes, females show sex-dependent elevation compared with males in these same restricted regions. Loss of neural cell body Fragile X mental retardation protein (FMRP) and Rugose [human Neurobeachin (NBEA)] suppresses localized PKA activity, whereas overexpression (OE) of MB lobe PKA-synergist Meng-Po (human SBK1) promotes PKA activity. Elevated Meng-Po subverts the PKA age-dependence, with elevated activity in immature animals, and spatial-restriction, with striking γ lobe activity. Testing circuit signaling requirements with temperature-sensitive shibire (human Dynamin) blockade, broadly expanded PKA activity was found within the MB lobes. Using transgenic tetanus toxin to block MB synaptic output, greatly heightened PKA activity was found in virtually all MB lobe fields, although the age-dependence is maintained. It is concluded spatiotemporally restricted PKA activity signaling within this well-mapped learning/memory circuit is age-dependent and sex-dependent, driven by FMRP-Rugose pathway activation, temporally promoted by Meng-Po kinase function, and restricted by output neurotransmission providing network feedback (Sears, 2022).

This study explored the spatiotemporal regulation of PKA activity within the MB lobes. PKA signaling initiates in early adulthood, with heightened activity in just 3/16 MB lobe output neuron fields (α'1, β'1, and β'2ap). In addition to age-dependence, this regional PKA signaling displays sex-dependence, with elevation in females over males. These findings were made possible with the PKA-SPARK biosensor; this fluorescent reporter uses motifs found in earlier PKA sensors, with pharmacological and genetic approaches promoting and preventing PKA activity verifying this new tool in cell culture and in vivo. While it is possible that the PKA-SPARK reporter is revealing only the strongest PKA activity, at the least α'1, β'1, and β'2ap connectivity regions have much higher PKA activity levels compared with the rest of the MB neuropil. In two disease models of intellectual disability and ASDs, from loss of either FMRP or Rugose/NBEA, PKA activity remains spatiotemporally restricted, but is dramatically reduced. There is surprisingly little change from overactivation of the FMRP→Rg control pathway, but PKA activity is profoundly altered by the PKA pathway Meng-Po kinase, with OE enhancing spatiotemporal PKA signaling and loss suppressing PKA activity. Network feedback downstream of KC neurotransmission strongly suppresses PKA activity, since blocking KC synaptic output with conditional shibire^ts or transgenic TNT induces widespread PKA activity signaling. Thus, localized PKA activity is highly regulated at the circuit level (Sears, 2022).

At a macro level, the α'1 and β'1 regions with the highest localized PKA activity levels have been linked to valence, i.e., whether local activation causes animals to approach or avoid a stimulus. The α'1 and β'2 fields exhibit opposing output valence (positive and negative, respectively), while the β'1 role appears less clear. However, higher PKA signaling can be activated in the γ3 region (via Meng-Po), which in combination with the β'1 region drives positive valence. Moreover, dopaminergic and serotonergic biosensor signals function through the β' region from a variety of external sense stimuli, to which the α'1/β'1 fields display high sensitivity, suggesting broad responsiveness. Very recent work shows high spontaneous activity in α'/β' restricted to young animals , suggesting α'/β' PKA signaling may be controlled by selective developmental activity. Moreover, inhibiting miR-92a in α/β and γ, but not α'/β', was shown to enhance memory, indicating another layer of lobe-selective circuit regulation. Note that the output neurons from these MB lobe regions are different (α'1 cholinergic, β'1 GABAergic, and β'2 glutamatergic), suggesting more complex integrative circuit functions. PKA signaling activation after the early-use critical period is consistent with sensory experience-dependent regulation. Importantly, MB sensory integration functions differ markedly between females and males, correlating with the report in this study of sex-dependent PKA signaling differences in females compared with males (Sears, 2022).

Two learning and memory proteins, FMRP and Rugose, are needed for full PKA activity in the α'1 and β'1 MB lobe output neuron fields. RNA-binding FMRP is a translational regulator known to facilitate PKA signaling, which is lost in the FXS, the commonest heritable cause of intellectual disability and ASD. Rugose/NBEA is a PKA-anchor that facilitates learning and memory, and is also associated with ASDs. Previous work has shown FMRP binds to rugose mRNA to drive KC expression. As predicted, disruption of this FMRP→Rg regulative pathway strongly impairs PKA activity in the MB lobes. In contrast, localized PKA signaling is dramatically strengthened by MB OE of the Meng-Po kinase, which induces early-onset PKA activity before adult sensory experience, spatially expands high PKA activity to the γ3 MBON field, and profoundly elevates PKA activity within all the normal MB lobe regions of heightened PKA signaling. Consistently, Meng-Po kinase OE also greatly improves learning and memory, via PKA phosphorylation, but additionally via signaling feedback synergy. Moreover, meng-po RNAi causes the opposite result of reducing localized PKA-SPARK puncta. Based on both loss and gain of function, it is suggested that Meng-Po enhances localized PKA activity, reflecting circuit level kinase regulation. Determining how Meng-Po-regulated PKA activity determines circuit excitability and regional balance will be a major subject of future research (Sears, 2022).

Two different KC synaptic output blocking methods dramatically expand PKA activity signaling in the MB. Both conditional shibire^ts and transgenic tetanus toxin tools block KC neurotransmission, but through quite different mechanisms. At 33°C, KC-targeted shibire^ts drives PKA activity expansion in the MB γ lobe. This change could indicate cross-compartment network interactions between the γ lobe and other MB regions. Increasing γ1 PKA activity is especially interesting, as γ1 toggles inhibition of other MB regions. The tetanus toxin protease blocks neurotransmission through eliminating SV exocytosis, and therefore provides a stronger and more selective means to silence KC synaptic output. Consistently, TNT animals show a more profound expansion of PKA activity throughout the MB lobes, albeit again affecting only spatial and not temporal patterning. Neither shibire^ts shibire^ts nor TNT blockade alters early PKA activity, suggesting induction of PKA signaling is determined primarily by later experience-driven activity. Localized PKA activity changes with KC output block implies active circuit balance; for example, weighing aversive versus attractive responses to sensory stimuli. The widespread PKA activity upregulation with KC output block leads to a hypothesis that enhancing KC neuron activity should result in elevated PKA signaling (Sears, 2022).

At the MB circuit level, multiple candidate synaptic pathways need to be explored for roles in local PKA activity regulation in different MB lobe output neuron fields. GABAergic inputs to dopaminergic neurons are one likely candidate, since GABA treatment has been shown to correct dfmr1 mutant circuit defects exacerbated by glutamate exposure. Moreover, GABAergic anterior paired lateral (APL) neurons broadly control MB activity through feedback to the KCs, and are most strongly activated by the α'/β' lobes. Recent work shows that treatment with a dopamine transport inhibitor also ameliorates rugose mutant social interaction and memory deficits. Another candidate is the Amnesiac neuropeptide from the serotonergic dorsal paired medial neurons, required for their normal development in the broad innervation of the MB lobes. In the context of these studies, de facto depression may feedback onto KCs to promote PKA activity signaling. The impact of upstream input onto the MB lobes is an important consideration, including how this circuitry combines with spontaneous MB activity and internal lobe circuitry to determine PKA signaling. Future research directions should attempt to dissect how these different layers of neuromodulation control localized PKA activity signaling within the MB lobe circuit, and between females and males, by manipulating input-specific neuronal activity in targeted transgenic studies (Sears, 2022).

In conclusion, this study reports that PKA activity signaling in the Drosophila brain MB learning and memory center is highly induced during early experiential adulthood, with selective upregulation in the α'1, β'1, and β'2ap MB lobe output neuron regions. Age-specific and sex-specific PKA signaling controlled within KCs and downstream of KC output shows that spatiotemporally restricted MB lobe PKA activity is regulated through a combination of both intracellular control and intercellular network-level mechanisms. Importantly, PKA signaling can be precociously promoted and spatial expanded though the activity of the PKA-synergist Meng-Po kinase. Moreover, KC neurotransmission inhibits localized PKA signaling within the MB circuit. Future studies will be aimed toward generating new genetic responder tools to test KC signaling with both neurotransmission output blockade and activity promotion of upstream and downstream MB circuit components, simultaneously and independently of KCs. PKA activity signaling will be tested with the manipulation of specific KC partners, by altering neurotransmission signaling in combination with postsynaptic neurotransmitter receptor mutants to determine network communication cues. Taken together with the current work, these ongoing studies will continue to expand understanding of circuit-level PKA signaling regulation in normal function and in neurologic disease model contexts (Sears, 2022).

Associative brain centers, such as the insect mushroom body, need to represent sensory information in an efficient manner. In Drosophila melanogaster, the Kenyon cells of the mushroom body integrate inputs from a random set of olfactory projection neurons, but some projection neurons-namely those activated by a few ethologically meaningful odors-connect to Kenyon cells more frequently than others. This biased and random connectivity pattern is conceivably advantageous, as it enables the mushroom body to represent a large number of odors as unique activity patterns while prioritizing the representation of a few specific odors. How this connectivity pattern is established remains largely unknown. This study tested whether the mechanisms patterning the connections between Kenyon cells and projection neurons depend on sensory activity or whether they are hardwired. A large number of mushroom body input connections were mapped in partially anosmic flies-flies lacking the obligate odorant co-receptor Orco-and in wild-type flies. Statistical analyses of these datasets reveal that the random and biased connectivity pattern observed between Kenyon cells and projection neurons forms normally in the absence of most olfactory sensory activity. This finding supports the idea that even comparatively subtle, population-level patterns of neuronal connectivity can be encoded by fixed genetic programs and are likely to be the result of evolved prioritization of ecologically and ethologically salient stimuli (Hayashi, 2022).

Dopaminergic neurons with distinct projection patterns and physiological properties compose memory subsystems in a brain. However, it is poorly understood whether or how they interact during complex learning. This study identified a feedforward circuit formed between dopamine subsystems and showed that it is essential for second-order conditioning, an ethologically important form of higher-order associative learning. The Drosophila mushroom body comprises a series of dopaminergic compartments, each of which exhibits distinct memory dynamics. A slow and stable memory compartment can serve as an effective 'teacher' by instructing other faster and transient memory compartments via a single key interneuron, which was identified by connectome analysis and neurotransmitter prediction. This excitatory interneuron acquires enhanced response to reward-predicting odor after first-order conditioning and, upon activation, evokes dopamine release in the 'student' compartments. These hierarchical connections between dopamine subsystems explain distinct properties of first- and second-order memory long known by behavioral psychologists (Yamada, 2023).

To identify and memorize discrete but similar environmental inputs, the brain needs to distinguish between subtle differences of activity patterns in defined neuronal populations. The Kenyon cells (KCs) of the Drosophila adult mushroom body (MB) respond sparsely to complex olfactory input, a property that is thought to support stimuli discrimination in the MB. To understand how this property emerges, the role was explored of the inhibitory anterior paired lateral (APL) neuron in the input circuit of the MB, the calyx. Within the calyx, presynaptic boutons of projection neurons (PNs) form large synaptic microglomeruli (MGs) with dendrites of postsynaptic KCs. Combining electron microscopy (EM) data analysis and in vivo calcium imaging, it was shown that APL, via inhibitory and reciprocal synapses targeting both PN boutons and KC dendrites, normalizes odour-evoked representations in MGs of the calyx. APL response scales with the PN input strength and is regionalized around PN input distribution. These data indicate that the formation of a sparse code by the KCs requires APL-driven normalization of their MG postsynaptic responses. This work provides experimental insights on how inhibition shapes sensory information representation in a higher brain centre, thereby supporting stimuli discrimination and allowing for efficient associative memory formation (Prisco, 2021).

Sleep is a fundamental behavioral state important for survival and is universal in animals with sufficiently complex nervous systems. As a highly conserved neurobehavioral state, sleep has been described in species ranging from jellyfish to humans. Biogenic amines like dopamine, serotonin and norepinephrine have been shown to be critical for sleep regulation across species but the precise circuit mechanisms underlying how amines control persistence of sleep, arousal and wakefulness remain unclear. The fruit fly, Drosophila melanogaster, provides a powerful model system for the study of sleep and circuit mechanisms underlying state transitions and persistence of states to meet the organisms motivational and cognitive needs. In Drosophila, two neuropils in the central brain, the mushroom body (MB) and the central complex (CX) have been shown to influence sleep homeostasis and receive aminergic neuromodulator input critical to sleep-wake switch. Dopamine neurons (DANs) are prevalent neuromodulator inputs to the MB but the mechanisms by which they interact with and regulate sleep- and wake-promoting neurons within MB are unknown. This study investigated the role of subsets of PAM-DANs that signal wakefulness and project to wake-promoting compartments of the MB. PAM-DANs were found to be GABA responsive and required GABA(A)-Rdl receptor in regulating sleep. In mapping the pathways downstream of PAM (protocerebral anterior medial) neurons innervating γ5 and β'2 MB compartments it was found that wakefulness is regulated by both DopR1 and DopR2 receptors in downstream Kenyon cells (KCs) and mushroom body output neurons (MBONs). Taken together, this study has identified and characterized a dopamine modulated sleep microcircuit within the mushroom body that has previously been shown to convey information about positive and negative valence critical for memory formation. These studies will pave way for understanding how flies balance sleep, wakefulness and arousal (Driscoll, 2021).

The mushroom body lobes are tiled by discrete anatomic compartments defined by the axons of a specific subset of DANs and the dendrites of one or two mushroom body output neurons (MBONs). This anatomical arrangement positions DANs to strategically convey positive and negative reinforced information by changing the synaptic weight of KC-MBONs in producing aversive and appetitive responses (Driscoll, 2021).

While, the most in-depth analysis of these synapses and distinct DAN-KC-MBON connectivity and behavioral output comes from studies of olfactory conditioning, there is evidence that these synapses play a critical role in innate behaviors like feeding and sleep. Although, role of DA on sleep has been extensively investigated in Drosophila, the commonly used TH-Gal4 driver line labels most dopamine neuron clusters, but is absent from the several PAM clusters that projects to MB (Driscoll, 2021).

This study specifically probed PAM subsets that project to γ5, γ4, and β'2 MB compartments. This study focused on this subset because KCs and MBONs downstream of these PAM neurons can be neuroanatomically resolved and have been shown to be required for wakefulness. Further, KCs and MBONs that form the γ5, γ4, and β'2 synaptic compartments alter their spontaneous neural activity in response to sleep need (induced by mechanical sleep-deprivation). The ability to use cell-specific split-GAL4 tools provides opportunity to resolve the precise circuit mechanisms by which PAM neurons regulate wakefulness (Driscoll, 2021).

GABA signaling also modulates sleep and wake microcircuits within MB. The key source of GABA in the MB is anterior paired lateral neurons, APL and dorsal paired medial neurons (DPM), which are electrically coupled and increase sleep by GABAergic inhibition of wake-promoting KCs. In the context of associative learning, there is strong evidence for interactions between KCs, APL, DPM and DANs but it is not clear if GABA and dopamine signaling represent opposing inputs to the KCs and MBONs in the regulation of sleep. This study found that the excitability of PAM DANs involved in wakefulness is blocked by sleep-promoting GABA signaling and mediated by ionotropic receptor subtype GABAA-Rdl (Driscoll, 2021).

A recent study showed that GABA inhibitory input to the presynaptic terminals of the PAM neurons regulates appetitive memory and that this interaction is mediated by GABA-B3 receptors that are clustered in PAM boutons localized to PAM-γ5 and -α1 compartments. These data are consistent with the findings that PAM-γ5 are GABA responsive and that multiple receptors are critical to this interaction. Since, no role was found for GABA-B3 in PAM mediated sleep regulation, it is likely that PAM γ5, γ4, and β'2 express multiple GABA receptors which are differentially recruited in sleep and learning. How and what regulates the expression of these receptors in PAM subsets presents a potential mechanism of presynaptic gating to MB core circuits. Transcriptomic analysis of PAM neurons reveals extremely high levels of Rdl expression followed by GABA-B3. Among the PAM subsets mean TPM or transcripts per million of Rdl receptor in PAM γ5, γ4, and β'2 are much higher as compared to other PAM subsets (Driscoll, 2021).

Simple connection query search of the recently released hemibrain data85 reveals there is significant bidirectional connectivity between APL, DPM, and PAM neurons. Further, a recent study showed that APL neurons express the inhibitory D2R receptor55. APL mediated GABAergic inhibition of the PAM neurons was recently shown to control the intensity and specificity of olfactory appetitive memory but previous results show that blocking GABA release from APL neurons only modestly affects sleep phenotypes (Driscoll, 2021).

While, the role of APL in GABA signaling to PAM γ5, γ4, and β'2 cannot be completely ruled out, other inputs to wake-regulating PAM DANs could also be GABAergic and critical for promoting sleep. A recent study using EM dataset of a Full Adult Female Fly Brain (FAFB) mapped the inputs and outputs of the PAMγ5 DANs and identified that this cell type is highly heterogenous and in addition to recurrent feedback from MBON01 γ5β'2a, it receives extensive input from other MBONs, sub-esophageal output neurons (SEZONs) and lateral horn output neurons86. The EM data also reveals that octopaminergic neurons synapse onto PAM γ5, γ4, and β'2 DANs. Whether, these inputs play a role in wakefulness is unknown but suggests that the PAMγ5 could serve as a key link between sensory inputs, wake-promoting octopamine signal and core sleep regulating circuitry within the MB. Each of these inputs could modulate PAM-DAN activity and dopamine release in regulating wakefulness via the MB (Driscoll, 2021).

In addition to probing the release and activity of these PAM-DANs the dopamine receptors and their location within the MB in signaling wakefulness were also explored. To this end validated RNAi lines were expressed in subsets of KCs and MBONs; DopR1 and DopR2 were found to be critical in mediating the wakefulness signal via KCs and γ5β'2 MBONs. Knocking down the receptor consistently increased total sleep and bout length. Furthermore, specific manipulations of DopR receptors within the MB did not directly alter locomotor activity as observed by manipulation of these receptors in CX. Although, loss-of-function mutations of D1 dopamine receptor DopR are shown to enhance repetitive air puff startle-induced arousal and increase sleep. Expression and restoration of DopR in the mutant background specifically in the central complex rescues the startle response, while, the sleep phenotype is rescued via a broad MB driver. The current data extends these findings by showing that the DopR receptors regulate sleep via the MB γ5 and β'2 compartment. Although, targeted RNAi experiments show that DopR's are required for sleep regulation by KCs and MBONs, the lack of a sleep phenotype in DopR2 mutant could be a result of global loss of receptor in the mutant as opposed to targeted loss of receptor function within MB. Dopamine signals wakefulness by activation of wake-promoting neurons of MB via DopR1 and DopR2 and within. the central complex, neurons of dFB are inhibited by dopamine via DopR2. Hence, DopR2 has opposing effects within MB and CX (Driscoll, 2021).

In vitro characterization indicates that DopR's signal through distinct G-proteins, with DopR1 via Gαs to stimulate cAMP production and DopR2 coupling to Gαq via increased calcium. These receptors are thought to have differential sensitivity to dopamine and could be potentially recruited by varying DA release or DAN activity. In the context of sleep regulation, this work reveals that both DopR1 and DopR2 induce wakefulness via the γ5 β'2 MB compartment but not γ4 compartment. Although, chronic activation of PAM γ4 induces wakefulness, the glutamatergic MBON γ4 < γ1,2 projects to multiple compartments and could potentially activate or inhibit MBONs and PAMs projecting to γ1 and γ2 compartment. The interaction between compartments is not well understood in the context of sleep and wake regulation and requires further investigation to better understand the role of DopR2 in regulating the γ4 compartment. The neuroanatomical specificity obtained from split-Gal4 lines combined with EM data has paved way for more detailed analysis of the role of dopamine signaling to MB in the context of sleep and other behaviors (Driscoll, 2021).

The sleep-regulating PAM DANs and associated KCs and MBONs identified in this study are also involved in mediating satiety, novelty, caffeine induced arousal, punishment and reward associated experiences suggesting that the activity of these neurons is tuned to several wake and arousal associated behaviors. This is further supported by the EM connectome data showing that MB receives extensive gustatory, auditory and visual input in addition to olfactory input (Driscoll, 2021).

Current models of sleep regulation rely on two main processes, the circadian clock and the sleep homeostat and don't completely account for multiple external and internal factors that influence wakefulness. The ability to sleep, however, is influenced by motivational or cognitive stimuli. It is therefore envisioned that sleep, wakefulness and arousal within MB are not located in distinct circuits, but rather mediated by distinct processes within a common circuit (Driscoll, 2021).

Animal behavior is shaped both by evolution and by individual experience. Parallel brain pathways encode innate and learned valences of cues, but the way in which they are integrated during action-selection is not well understood. This study used electron microscopy to comprehensively map with synaptic resolution all neurons downstream of all Mushroom Body output neurons (encoding learned valences) and characterized their patterns of interaction with Lateral Horn neurons (encoding innate valences) in Drosophila larva. The connectome revealed multiple convergence neuron types that receive convergent Mushroom Body and Lateral Horn inputs. A subset of these receives excitatory input from positive-valence MB and LH pathways and inhibitory input from negative-valence MB pathways. Functional connectivity from LH and MB pathways and behavioral roles of two of these neurons was confirmed. These neurons encode integrated odor value and bidirectionally regulate turning. Based on this it is speculated that learning could potentially skew the balance of excitation and inhibition onto these neurons and thereby modulate turning. Together, this study provides insights into the circuits that integrate learned and innate valences to modify behavior (Eschbach, 2021).

Selecting whether to approach or avoid specific cues in the environment is essential for survival across the animal kingdom. Many cues have both innate valences acquired through evolution and learned valences acquired through experience that can guide action selection. Innate and learned valences are thought to be computed in distinct brain areas, but the circuit mechanisms by which they are integrated and by which learned valences can override innate ones are poorly understood. Using the tractable Drosophila larva as a model system, this study describes with synaptic resolution the patterns of convergence between the output neurons of a learning center (the MB) and an innately attractive pathway in the lateral horn (the LH). 62 neurons per brain hemisphere were identified that represent direct points of convergence between the MB and the LH, that fall into a number of different subtypes based on their patterns of MB and LH inputs and potentially encode a number of distinct features. One subtype of 10 convergence neurons (CNs) receives excitatory input from positive-valence MB and LH pathways and inhibitory input from negative-valence MB pathways. Functional connectivity from LH and MB pathways and behavioral roles of two of these neurons were confirmed. These neurons encode integrated odor value (coding for positive value with an increase in their activity) and regulate turning. They are activated by an attractive odor, and when activated they repress turning. Conversely, when inactivated, they increase turning. Based on this, it is speculated that learning could potentially skew the balance of excitation and inhibition onto these neurons and thereby modulate turning. For one of these neurons, this study indeed verified that aversive learning skews inputs towards inhibition. Together, this study provides insights into the circuit mechanisms by which learned valences could interact with innate ones to modify behavior (Eschbach, 2021).

The brain areas that compute innate and learned valences of stimuli interact with each other, but despite recent progress, their patterns of interaction are not fully understood. In principle, MBONs could synapse directly onto LHNs thereby directly modifying innate valences. Alternatively, LHNs could directly synapse onto MBONs. Finally, learned and innate valences could initially be kept separate, and MBONs and LHNs could converge on downstream neurons. This study found that in Drosophila larva (1) some MBONs synapse directly onto some LHNs; (2) some MBONs received direct synaptic input from LHNs and (3) many MBONs and LHNs converge onto downstream CNs, similar to findings in the adult. One MBON (m1) was also a CN, receiving significant direct input from other MBONs and from LHNs. Overall, the architecture suggests some early mixing of representations of innate and learned valences, but to some extent these representations are also kept separate in the LH and MB, and then integrated by the downstream CNs. Maintaining some initial separation of representations of innate and learned valences prior to integration could offer more flexibility and independent regulation, for example by context or internal state. Convergence neurons could compute learned valence by comparing odor-drive to positive- and negative-valence MBONs (Eschbach, 2021).

The prevailing model of MB function in adult Drosophila proposes that in naive animals the odor drive to positive- and negative-valence MBONs is equal such that their outputs cancel each other out and the LH circuits guide olfactory behavior. Learning alters the overall output towards positive- or negative-valence MBONs by modifying specific KC-to-MBON connections. This model raises several important questions. First, how is the output from MBONs of opposite valence integrated to compute a learned valence? Second, how does it interact with the output of the LH. The current findings provide further support for this model and shed insight into these questions (Eschbach, 2021).

EM reconstruction combined with neurotransmitter information has revealed a class of 10 CNs that receive excitatory input from positive-valence MBONs and inhibitory input from negative-valence MBONs. These CNs are well poised to compute the learned odor valence by comparing the odor drive to positive- and negative-valence MBONs (Eschbach, 2021).

For two members of this class (MBON-m1 and CN-33), their MB drive was tested in untrained larvae, in vivo or in explants. On average, across the population of untrained individuals, KC activation did not induce a significant change in ∂F/F0 in either CN. However, it was found that in some individuals the MB drive onto the CNs was excitatory, and in others inhibitory, indicating that the MB can provide both excitatory and inhibitory drive to the CNs. Consistently, when inhibitory neurotransmission was blocked using PTX, MBON-m1 and CN-33 showed strong excitatory responses to KC activation (Eschbach, 2021).

High variability was observed in the responses of both CN-33 and MBON-m1 both to activation of the whole MB pathway, as well as to odor presentation. This variability may in part be due to technical reasons. Indeed, the variable responses to pan-KC optogenetic activation could be due to differences in CsChrimson expression in KCs across individuals (the expression of CsChrimson was verified for each individual recording but it was not quantified). However, such a difference is unlikely to be a reason for a range of responses that spans inhibition to excitation following the stimulation of the same neuronal population. Likewise, the exposure of the sensory organ to an odor may vary from one animal to the other depending on the location of the larva's head in the chip's channel. Together with other technical aspects, this may account for some of the variance observed in in vivo. Indeed, a decomposition of the variance within the datasets by ANOVA revealed a significant effect of the identity of the individual on the variance. Interestingly though, the inter-individual variability contributed to a higher fraction of the overall variance in odor response in larvae with functional MBs than in larvae with silenced MBs. This analysis suggests that the MB pathway is a significant source of variability in odor-evoked responses in MBON-m1 and CN-33 and is consistent with the highly variable responses of these CNs to direct KC activation. The variability in response of CNs to MB inputs across untrained individuals could be due to different experiences prior to these experiments, as suggested also by the fact that MBON-m1 responses to trained odors are modified with training. High variability in MBON responses to odor across individuals has previously also been observed in the adult flies and has been related to different individual experiences. The MBs can also compute other kinds of information, such as internal state which may modulate an individual's disposition toward sensory stimuli. Therefore, the average response across untrained individuals might be similar to the response of a single individual in a naive neutral state (with any interindividual differences averaged out across a population), and the variability may represent the degree of freedom for the MB to tune this response to a stimulus depending on previous experience or state (Eschbach, 2021).

How do the learned valences modify the innate ones? How is conflict between opposite innate and learned valences resolved during action selection? One possibility could involve the integration of conflicting signals into a unified representation of value, a notion similar to common currency valuation of options, which could then be used to promote or suppress specific actions. The findings suggest that the 10 CNs that can read out the learned valence by comparing MBON inputs of opposite valence also integrate the learned valence with the innate one. Thus, the 10 CNs that receive inhibitory input from negative-valence MBONs and excitatory input from positive-valence MBONs, also receive inputs from a positive-valence LHN pathway. These neurons are therefore well poised to compute an integrated odor value and code for a positive value with an increase in their activity. For two members of this class (MBON-m1 and CN-33) it was possible to confirm these predictions. This study has shown they are activated by an innately attractive odor via the LH pathway in untrained animals. Furthermore, when the innate attractiveness of an odor was reduced through aversive training, the activity of MBON-m1 was also reduced. Finally, it was shown that activating MBON-m1 and CN-33 represses turning, further supporting the idea that their activation encodes positive value. Interestingly, it was also found that a decrease in their activity promotes turning, raising the possibility that they could bi-directionally encode value, coding for negative value with a decrease in their activity (Eschbach, 2021).

In principle, a single CN of this type could potentially be sufficient to compute integrated odor value and regulate turning, but a population of 10 CNs with similar patterns of input from positive- and negative-valence pathways was found that likely operate partially redundantly with each other. Each CN also had unique aspects of connectivity, raising the possibility that they may also encode partially complementary features and that the integrated value could be distributed across the CN population (Eschbach, 2021).

Based on these findings the following model is proposed that could explain the way in which learning could modulate innate responses to odors through a population of integrative CNs: (1) In naive animals, the CNs are activated by innately attractive odors, mainly via the LH pathway (3) when activated these neurons repress turning, which enables crawling towards an attractive odor source; (3) in naive animals the net MB drive onto these CNs is close to 0; (4) aversive learning can skew the net MB output onto CNs towards inhibition, so that aversively-conditioned odor fails to activate these neurons; (5) if CNs are not activated, turning rate remains high when crawling towards the odor; (6) without sufficient suppression of turning by an odor, the animals' ability to approach the odor source is impaired. This proposed model could explain how aversive learning suppresses approach of innately attractive odors. To fully suppress approach, multiple CNs of this class would likely need to be silenced (Eschbach, 2021).

The model presented above could readily be extended to explain how appetitive learning could enhance odor approach and how strong aversive learning could switch innate odor approach to learned avoidance. Thus, appetitive learning could skew the net MB drive towards excitation thereby repressing turning even more. In contrast, strong aversive learning could skew the conditioned odor drive towards inhibitory negative-valence MBONs so much that the inhibition would become stronger than the excitatory LH drive. The CN would then be inhibited by the aversively conditioned odor. Since inhibition of CNs promotes turning, the aversively conditioned odor could promote odor avoidance by inducing turning. Consistent with this idea, inhibitory responses to the innately attractive odor were observed in MBON-m1 and in CN-33, in some untrained individuals, and strong excitatory responses in others, proving that odor drive onto CNs can range all the way from strong excitation to inhibition. Inhibition of MBON-m1 following aversive learning in the microfluidic device was not observed, but this could be because the animals did not form a very strong aversive memory under these conditions. Testing these proposed extensions of the model will require the future development of automated single-animal training assays and calcium imaging tracking microscopes to correlate the strength of the learned behavior with the conditioned-odor response of the CNs (Eschbach, 2021).

Finally, EM reconstruction did reveal a potentially opposite population of CNs that are inhibited by MBONs of positive value. It will be interesting to test in the future whether these neurons potentially encode negative value with an increase in their activity and positive value with a decrease in their activity and regulate turning in the opposite way: promoting turning when they are activated and repressing it when they are inhibited. Having two different populations of neurons that encode value in opposite ways could further increase the dynamic range of a distributed value code (Eschbach, 2021).

This study found that many of the CNs that receive input from both MBONs and LHNs also provide direct (including CN-33) or indirect (including MBON-m1) feedback to MB modulatory neurons that provide teaching signals for learning (n = 18). In a former study, it was shown that at least some of these feedback connections are functional and can influence memory formation. For example, CN-33/FAN-7 is capable of generating an olfactory memory when it is paired with an odor. This type of connectivity is consistent with learning theories that propose that future learning is influenced by predicted value computed based on prior learning. A major role of the CNs discovered in this study may therefore be not only to organize current actions, but also to regulate learning (Eschbach, 2021).

In summary, the comprehensive synaptic-resolution architecture of the circuits downstream of the learning center output neurons presented in this study is a valuable resource for constraining future modelling and function studies of value computation, action selection, and learning (Eschbach, 2021).

The mushroom bodies of Drosophila contain circuitry compatible with race models of perceptual choice. When flies discriminate odor intensity differences, opponent pools of αβ core Kenyon cells (on and off αβ(c) KCs) accumulate evidence for increases or decreases in odor concentration. These sensory neurons and "antineurons" connect to a layer of mushroom body output neurons (MBONs) which bias behavioral intent in opposite ways. All-to-all connectivity between the competing integrators and their MBON partners allows for correct and erroneous decisions; dopaminergic reinforcement sets choice probabilities via reciprocal changes to the efficacies of on and off KC synapses; and pooled inhibition between αβ_c KCs can establish equivalence with the drift-diffusion formalism known to describe behavioral performance. The response competition network gives tangible form to many features envisioned in theoretical models of mammalian decision making, but it differs from these models in one respect: the principal variables-the fill levels of the integrators and the strength of inhibition between them-are represented by graded potentials rather than spikes. In pursuit of similar computational goals, a small brain may thus prioritize the large information capacity of analog signals over the robustness and temporal processing span of pulsatile codes (Vrontou, 2021).

Two-alternative forced-choice tasks, in which a subject must commit to one of two alternatives, sometimes under time pressure and nearly always with uncertain information, are a commonly studied laboratory simplification of real-world decision making. The neural processes that culminate in a binary choice have been compared to the deliberations of a jury before a verdict: neurons, like jurors, gather evidence from witnesses over the course of a trial and then reconcile their divergent views in a majority vote (Vrontou, 2021).

The problem of how neural circuits implement this form of trial by jury has been approached in a range of species, from primates and rodents to fish and flies. A pioneering and influential body of work is built on a two-alternative forced-choice task in which monkeys distinguish directions of motion in a noisy random dot display. Recordings of correlated neuronal activity suggest that motion-sensitive neurons in the middle temporal visual area (MT or V5) provide momentary evidence that is temporally integrated in lateral intraparietal cortex (LIP) before passing an unspecified thresholding mechanism. Although the precise role attributed to LIP is a matter of debate, the principle that ephemeral sensory signals flow into integrators whose fill levels rise to a response threshold appears general; similar arrangements have been inferred to support visual motion discrimination in zebrafish and odor intensity discrimination in the fly (Vrontou, 2021).

In Drosophila, a rate-limiting integration step takes place in a particular group of third-order olfactory neurons. When flies decide on the direction of an odor concentration change, the membrane potentials of Kenyon cells (KCs) in the αβ core (αβ_c) division of the mushroom bodies drift noisily toward action potential threshold, just as accumulating evidence would drift toward a response bound. Consistent with the proposed correspondence of membrane voltage and integrated sensory information, and of action potential and decision thresholds, neurometric functions based on the average timing of the first odor-evoked spikes in the αβ_c KC population can account for the speed and accuracy of the decision-making animal; psychophysical estimates of noise in the decision process match the measured membrane potential noise of αβ_c KCs; and genetically targeted manipulations that alter the latencies of αβ_c KC spikes have the expected impact on reaction times (Vrontou, 2021).

Two functionally separate groups of αβ_c KCs, termed up and down or on and off cells, respond to increases or decreases in odor concentration and can therefore represent the strength of evidence for either of the two alternatives in the choice. This explicit representation of support for each of the competing hypotheses (as opposed to an aggregate representation of the extent to which one hypothesis is favored over the other) suggests that a decision involves a race between two integrators-one built from neurons that accumulate evidence for an increase in odor concentration and another composed of 'antineurons' that do the opposite. Changes in odorant receptor occupancy at the periphery alter the baseline activity of olfactory receptor neurons and the second-order projection neurons (PNs) with which they form receptor-specific glomerular channels. Large odor concentration changes in a channel's preferred direction drive high-frequency transmission from PNs to αβ_c KCs that promotes steep depolarizations to spike threshold and fast, accurate decisions, whereas small concentration changes in the preferred direction, or any change in the null direction, cause only a trickle of synaptic release; shallow, undulating membrane potential rises; and long spike delays that lead to slow, error-prone choices (Vrontou, 2021).

This study examined whether the circuitry downstream of αβ_c KCs is compatible with a model of two competing integrators. Three predictions of such a model were tested. First, to adjudicate the rival hypotheses advocated by on and off αβ_c KCs, mushroom body output neurons (MBONs) sampling the cores of the αβ lobes must listen to both. It is therefore expected that each core-innervating MBON is excited by increases as well as decreases in odor concentration. Second, as an animal learns the rules of the two-alternative forced-choice task-that an increase in odor intensity predicts imminent electric shock, whereas a decrease signals protection-the influence of αβ_c KCs championing the correct choice should be enhanced while that of proponents of the incorrect choice should be diminished. In other words, antagonistic changes are expected in the strengths of connections of on and off αβ_c KCs with the same action selection neurons if evidence for the competing alternatives is accumulated separately. Third, race models become equivalent to a drift-diffusion process-the formalism shown accurately to describe the psychophysics of the decision-making animal-only if they include an element of mutual or pooled inhibition to establish response competition between the integrators. Inhibition is needed to ensure that the integrators are anti-correlated so that evidence for one choice simultaneously counts as evidence against the other. This study therefore predicts the existence of inhibitory interactions between αβ_c KCs (Vrontou, 2021).

The idea that decisions are based on the accumulated spikes of oppositely tuned sensors was born in early attempts to unite psychophysical and neurophysiological measurements under the umbrella of signal detection theory. The recorded spike count distributions of direction-selective units in the monkey's area MT to motion in the preferred or null directions were taken to represent the responses of two neurons-the recorded neuron and its imagined antineuron conjugate-to movement in the neuron's preferred direction. The likely direction of motion can then be inferred as the probability that a draw from the neuron's response distribution yields a larger spike count than a draw from the antineuron's. At minimal motion strengths, when the two distributions are congruent, these odds are even and choices are random, but as the neuron responds ever more vigorously to increasingly coherent motion while the antineuron's response stays flat, the distributions unmix and the probability of a correct choice rises toward one. Comparing the spike counts of two sensors rather than thresholding the output of one removes shared sources of variation and with them the need of adjusting the discrimination threshold to achieve the best separation of the changing response distributions: a neuron-antineuron pair always returns a quantity proportional to likelihood ratio, the optimal hypothesis test. Although opponent sensory channels in one or another guise feature prominently in many decision-making models, their involvement in the brain is unproven: neurons and antineurons owe their status to each other, as inputs to comparator circuits, but these circuits remain uncharacterized (Vrontou, 2021).

This study draws back the curtain on one such circuit in the fly. Changes in odor intensity are registered by pools of on and off αβ_c KCs, which represent the strengths of the accumulated evidence for an increase or decrease in odor concentration. These pools of sensory neurons and antineurons couple to a second layer of neurons and antineurons, the core-innervating MBONs, which bias behavioral intent in opposite ways. Members of both neuronal pools in the sensory layer connect to both types of MBON in the action selection layer via plastic synapses. With two sets of neurons and antineurons and all-to-all feedforward connectivity between them, the comparator circuit allows for approach or avoidance following judgments of upward or downward changes in odor intensity-that is, it comprises neural pathways representing the possible contingencies seen behaviorally. The perceptual decision is won-correctly or incorrectly-by the αβ_c KC pool that reaches spike threshold first, and it is expressed in the behavior instructed by that pool's favored MBON partners (Vrontou, 2021).

Unlike neurons comprising the ON and OFF pathways of motion vision, on and off αβ_c KCs cannot be distinguished and manipulated genetically. This study has therefore exploited the sensitivity of KC-to-MBON synapses to the timing of reinforcement to reveal the convergence of separate on and off channels onto the same MBONs. KC-to-MBON synapses in their ground state exert finely balanced drive on the MBON ensemble, so that votes cast by its members cancel one another as in a hung jury, but experience can shift the synaptic weight distribution and the resulting pattern of MBON activation away from net zero. This study has documented such shifts for the approach-advocating MBON-γ1pedc>αβ: pairing odor on- or offset with electric shock weakens transmission from the on αβ_c KC pool and strengthens transmission from the off αβ_c KC pool (or vice versa), synergistically changing odor preference. The underlying mechanism is a switch from synaptic depression to synaptic potentiation when the order of odor-evoked KC activation and dopaminergic reinforcement is reversed. This mechanism operates at KC connections with all core-innervating MBONs but is likely engaged at different timescales that may reflect sequential memory phases; to demonstrate the mechanism's ubiquity, this temporal sequence was artificially collapsed by photostimulating DANs directly. Within the short time frame of these behavioral experiments, only PPL1-γ1pedc, but not PPL1-α'2α2, shows significant pain responses that modulate its sensitivity to a predictive odor, consistent with the view that PPL1-γ1pedc and its cognate MBON-γ1pedc>αβ represent the core circuit for the storage and expression of short-term aversive memories (Vrontou, 2021).

A crucial element of many neural network models of decision making is inhibitory feedback from a common interneuron pool driven by the competing integrators, which helps to amplify small differences in conflicting sensory evidence until, eventually, one integrator prevails. The response competition circuit this study has delineated contains such an inhibitory element but with the intriguing twist that the key variables are represented by membrane voltages rather than spikes. Analog processing may be a consequence of numerical constraints: if the mushroom bodies lack the neuron numbers needed to approximate continuous quantities with discrete-time action potentials, there may be little choice but to swap the advantages regenerative spikes could provide (such as long time windows for adding and retaining sensory evidence) for the greater information capacity of graded potentials. Perhaps more is different (Vrontou, 2021).

The capacity to use past experience to guide future action is a fundamental and conserved function of the nervous system. Associative memory formation, initiated by the coincident detection of a conditioned stimulus (CS; e.g., odor) and an unconditioned stimulus (US; e.g., sugar reward), leads to a short-lived memory (STM) trace within distinct circuits. Memories can be consolidated into long-term memories (LTMs) through processes that depend on de novo protein synthesis, require structural modifications within the involved neuronal circuits, and might lead to the recruitment of additional ones. Compared with modulation of existing connections, the reorganization of circuits affords the unique possibility of sampling for potential new partners. Nonetheless, only few examples of rewiring associated with learning have been established thus far (Baltruschat, 2021).

The formation and retrieval of olfactory-associative memories in Drosophila require the mushroom body (MB). Within the main MB input compartment, the calyx (MBC), second-order projection neurons (PNs), delivers olfactory information through cholinergic synapses to the intrinsic MB neurons, the Kenyon cells. In the MBC, large, olfactory PN boutons are enwrapped by the claw-like dendrite termini of ∼11 KCs on average, thereby forming characteristic synaptic complexes, the microglomeruli (MGs), which display functional and structural plasticity in adaptation and upon silencing. To start systematically addressing the mechanisms that support memory consolidation, this study sought to investigate the properties of identifiable synaptic MGs in the MB of the adult brain of Drosophila after the establishment of LTMs (Baltruschat, 2021).

Combining behavioral experiments with high-resolution microscopy and functional imaging, this study demonstrates that the consolidation of appetitive olfactory memories closely correlates with an increase in the number of MGs formed by the PNs that deliver the conditioned stimulus and their postsynaptic KC partners. These structural changes result in additional, functional synaptic connections. Thus, the circuit in the calyx of the fly MB reorganizes accompanying the consolidation of associative memories (Baltruschat, 2021).

This study reports input-specific reorganization of the adult MBC circuit associated with the formation of long-term, appetitive memory. By visualizing presynaptic markers in PNs and the KC postsynaptic densities, this study uncovered an increase in the number of PN boutons and, at the same time, reveal that these boutons are enveloped by KC postsynaptic profiles, suggesting that new MGs are formed during memory consolidation. These findings are particularly remarkable, given the high degree of complexity of the MG microcircuits revealed by EM reconstruction and including the dendrite claws of multiple KCs of distinct subtypes. The cellular mechanisms leading to the increased number of odor-specific complex MGs remain to be clarified, but they will require a tight coordination between pre- and postsynaptic partners. In this context, mutations in synaptic proteins or in proteins mediating cell-cell interactions, which specifically block LTM, will be of great interest (Baltruschat, 2021).

It is suggested that remodeling could be driven by intrinsic reactivation of KCs during the consolidation phase or by modulatory inputs into the calyx. In either case, a complex pattern of activation is expected, that might be difficult to reproduce in artificial settings. Although the present observations are limited for technical reasons to the specific case of cVA, the overall density of PN boutons in the MBC increases after appetitive long-term conditioning in honeybees, as well as in leaf-cutting ants after avoidance learning. Based on that and given that the olfactory pathway of cVA is not distinguishable from that of other odors, it is thus suggested that the findings might be generalizable. In comparison with those systems, however, genetic and functional identification of PN subsets were used to reveal that the structural modifications are specific and limited to the PNs conveying the conditioned odor. Importantly, in vivo functional imaging data support the view that the circuit reorganization leads to additional functional MGs responding to the conditioned odor. In addition, they demonstrate a specific change in functional response in the KC dendrites toward the trained odor because the calcium levels drop faster toward baseline after appetitive associative conditioning. The faster decay kinetics and more skewed response toward the onset of the stimulus could contribute to a more-efficient temporal summation of responses or refine the KC response and might be related to inhibitory modifications. An important open question is the effect of the increased number of responding MGs on the pattern of KC activation. KCs respond sparsely to odor input and require the coincident activation of multiples of their claws to produce an action potential. The data might underlie the addition of connections between the active PNs and a set of already-responding KCs, leading to facilitated response to the conditioned odor without changing the set of responding KCs. A recent publication, however, suggests an exciting alternative view. After aversive LTM establishment, the number of KCs responding to the conditioned odor is increased (Delestro, 2020). If it is hypothesized that appetitive conditioning leads to a similar outcome, the data could provide anatomical and functional support to these findings. The pattern of KC response could, thus, be modulated by experience in adulthood and might represent a rich signifier of sensory stimulus and context. Reconstruction of an MG from EM serial sections derived from FAFB dataset (Baltruschat, 2021).

Spatial orientation memory plays a crucial role in animal navigation. Recent studies of tethered Drosophila melanogaster (fruit fly) in a virtual reality setting showed that the head direction is encoded in the form of an activity bump, i.e., localized neural activity, in the torus-shaped ellipsoid body (EB). However, how this system is involved in orientation working memory is not well understood. This study investigated this question using free moving flies (D. melanogaster) in a spatial orientation memory task by manipulating two EB subsystems, C and P circuits, which are hypothesized for stabilizing and updating the activity bump, respectively. To this end, two types of inhibitory ring neurons (EIP and P) which innervate EB were suppressed or activated, and it was discovered that manipulating the two inhibitory neuron types produced distinct behavioral deficits, suggesting specific roles of the inhibitory neurons in coordinating the stabilization and updating functions of the EB circuits. The neural mechanisms underlying such control circuits were further elucidated using a connectome-constrained spiking neural network model (Han, 2021).

Maintaining spatial orientation is a crucial cognitive capability required for animal navigation, and understanding the detailed neural mechanisms of spatial orientation is of great interest to researchers in the fields of neurobiology or neuromorphic engineering. In recent years, significant progress has been made in identifying the neural circuits that support spatial orientation in the central complex of Drosophila melanogaster. The central complex has long been associated with short-term spatial memory, visual pattern memory, and motor control. The recent discoveries of head-direction selectivity and localized neural activity in two central complex neuropils, the ellipsoid body (EB) and the protocerebral bridge (PB), have also linked the central complex to the function of spatial orientation. These studies suggested that the head orientation is encoded by localized neural activity, called activity bump, and the bump location in EB shifts in accordance with changes of heading during movement. The function of the EB neurons resemble that of a compass and is, therefore, termed 'neural compass' (Han, 2021).

In light of these empirical observations, several neural circuit models of the central complex have been proposed to elucidate the neural circuit mechanisms of head-direction selectivity or other functions associated with the central complex. Some models focused on the stability of the activity bump or on the differences in the circuit dynamics between locus and fruit fly. Other models studied the plasticity involved in the flexible retinotopic mapping but used simpler firing rate models or schematic models. A large-scale firing-rate neural network model that covered the entire central complex was able to reproduce the steering and homing behavior of bees, but the EB circuits were rather simple with minimal details (Han, 2021).

Recently, a spiking-neuron model of the EB-PB circuits was proposed. The model used a more realistic spiking-neuron model and synaptic dynamics to elucidate how the circuits can maintain a stable activity bump when fruit flies switch between forward movement and rotation states in the absence of landmarks. The model suggested the involvement of two subcircuits: one forms an attractor network and maintains (or stabilizes) an activity bump; the other forms a shifter network and shifts (or updates) the bump position in accordance with changes in body orientation. The model successfully demonstrated the angular errors when a fly moved in darkness and predicted the asymmetric activity in the PB during rotation (Han, 2021).

The model made an important and unique prediction: the function of spatial orientation working memory requires coordinated activation of the bump-maintaining (or stabilizing) and bump-shifting (or updating) circuits that are controlled by the upstream ring neurons (Han, 2021).

However, most of the experimental studies used tethered flies in a virtual reality setting and focused on how manipulation of neurons affects the bump activity. It is not clear how these neurons, in particular those involved in stabilizing and updating the activity bump, play roles in cognition-relevant behavior such as spatial orientation memory in free-moving flies with a more realistic behavioral setting. This study aimed to address these questions and designed a behavioral task of spatial orientation working memory based on the classic Buridan's paradigm. Specifically, two types of GABAergic ring neurons were manipulated that are hypothesized to control these neurons. Ring neurons project their axons into EB and inhibit neurons including those display the activity bump. Previous studies have reported the roles of the ring neurons in visually-guided behavior, ethanol sensitivity, sleep regulation, olfactory memory and mating behavior. However, their roles in the working memory of spatial orientation in the presence or absence of visual cues remain unclear. In addition to the neural functional experiments, computer simulations were performed using the EB-PB model, which produced neural activities that were consistent with the behavioral changes observed in the fruit flies with different experimental conditions. The present study provides a detailed picture on how coordinated activation between the neural processes of stabilization and update plays a crucial role in spatial orientation working memory (Han, 2021).

It was hypothesized that the C circuit and P circuit in the EB circuits stabilize and update the orientation-encoding activity bump and they are regulated by corresponding GABAergic ring neurons. We tested this hypothesis by manipulating two types of GABAergic ring neurons in a spatial orientation working memory task with free moving fruit flies, and discovered manipulating each ring neuron type led to different behavioral abnormality. By performing computer simulations on a previously proposed EB-PB neural circuit model, it was possible to explain the results of the experiments and provided a picture of the neural circuit mechanism underlying spatial orientation working memory: the orientation-encoding bump is maintained through two alternately activated neural processes: one that stabilizes the position of the activity bump and one that updates the position of the bump. The former is activated when a fruit fly maintains a steady head direction and the latter is activated when the fly rotates its body. The control of this process is performed through specific GABAergic ring neurons. Therefore, overactivating or suppressing the ring neurons disrupts the alternation of the two processes and leads to incorrect or even loss of orientation memory (Han, 2021).

There are a few more interesting discoveries worth discussing. Performing fixation toward previous landmark directions requires two things to be remembered: the earlier event of the landmark presentation (what) and the directions of the landmarks (where). Flies that fail to remember the former would not exhibit the fixation behavior at all, while flies that forget the latter would still perform the fixation but toward incorrect directions. Discoveries of strong fixation but with large deviation from the true directions of the landmarks for flies with EIP-ring neuron suppression during the third stage may imply the segregation of the neural mechanisms of orientation memory regarding the 'where' and 'what' of a landmark (Han, 2021).

One interesting finding of the present study is a long duration of spatial orientation working memory during the poststimulus stage. Previous studies reported the occurrence of poststimulus fixation behavior that lasted only for a few seconds immediately following the offset of the landmarks. Indeed, this study observed that the flies tended to stop their movement a few seconds after the sudden disappearance of the landmarks in the third stage. But they usually resumed the movement in a few seconds. This might be the reason why earlier studies only claimed a few seconds of fixation if their analyses did not include the resumed movement. Further studies are needed to investigate this issue (Han, 2021).

A couple issues regarding the choice of molecular tools should be discussed. In the present study tub-GAL80ts was used in combination with UAS-Kir2.1 or UAS-TNT to suppress targeted ring neurons. The method involves raising the temperature 1 d before the behavioral experiments and therefore taking effects on a much longer time scale than using optogenetic tools. This long-term suppression may induce other effects at the cellular or circuit levels, which are beyond what the model can simulate. Further study may be required to carefully examine the long-term effects. An ideal solution is to transiently suppress the ring neurons using optogenetic tools such as UAS-NpHR (peak sensitivity wavelength ~589 nm) or GtACRs (peak sensitivity wavelength ~ 527 nm for GtACR1 and ~457 nm for GtACR2). However, the wavelength of the required activation light is within the visible range of the fruit flies. Preliminary tests on UAS-NpHR showed that the onset of the activation light seriously disrupted the fixation pattern of wild-type flies. A new optogenetic tool or a carefully re-designed optical system is required to transiently suppress targeted ring neurons while not interfering the visual experiments. The second issue is related to the GAL4 lines. In the present study only two most specific lines, c105-GAL4 and VT5404-GAL4, were used to target the EIP-ring and P-ring neurons, respectively. There are several other less specific GAL4 lines available for the two types of neurons. It is necessary to conduct the same experiments using these overlapping lines to further confirm the results presented in this study (Han, 2021).

Several other important questions remain to be addressed. Previous studies showed that EB does not maintain a fixed retinotopic map and a bump can start from a random location in the beginning of a trial. For the sake of modeling simplicity, this study did not model the random starting point feature in the model. But this feature is easy to implement and does not affect the conclusion of this study. A global excitation needs to be applied to the entire EB to reset the system. The excitation will induce strong competition between the EIP neurons and a new bump will start at a random location through the winner-take-all dynamics. Following this issue, random regeneration of an activity bump also needs to be discussed. The model showed that photoactivation of either EIP-ring or P-ring neurons during the poststimulus stage permanently abolished the activity bump. However, based on the observation of spontaneous generation of activity bump in other studies, the bump is likely to be regenerated at a random location after the offset of photoactivation. The regeneration can be easily implemented in the current model using the same mechanism described above. Since the regenerated bump starts from a random location, the fruit flies lose the reference to the landmark locations. Thus, adding a spontaneous bump or not both lead to the same conclusion: the files fail to fixate on the previous landmark locations. Although not affecting the conclusion of the present study, the spontaneous bump feature may be crucial in future studies that involve modeling of the steering mechanism (Han, 2021).

Another issue is that a couple experimental and modeling studies suggested that ring neurons provides the mechanisms underlying flexible retinotopic mapping in EPG (or EIP) neurons rather than the simple suppression/activation mechanism as hypothesized in the present study. However, the ring neurons (R2 and R4d) tested in one study are of different types from what was tested (R1 and R6) in this study. Additional experiments that measure the activities of R2 and R4d using the setups described in previous studies are required to clarify this issue (Han, 2021).

It is important to compare and discuss differences between computational models of the central complex in terms of the functions investigated in the present study. However, most models focused on different aspects of the compass circuit functions. The major difference from previous models is that they proposed that the PB intrinsic neurons as the main source of inhibition that regulates the attractor dynamics, while in the current model this function is conducted by the EIP ring neurons with two additional ring neuron types (C-ring and P-ring neurons) modulating different subcircuits of the system. An in-depth model comparison and experimental manipulation of PB intrinsic neurons and ring neurons under the present behavioral task may be able to clarify this issue (Han, 2021).

A final issue is related to the function of the activity bump which is commonly thought to represent the fly's sense of orientation in a manner similar to that of the head-direction system found in rodents. However, as aforementioned 'what' and 'where' mechanisms, performing the fixation behavior as an indication of orientation working memory may require several serial or parallel neural components beyond EB and PB. For example, how is this innate fixation behavior initiated (motivation)? When a fly stops fixating, it is not clear whether the fly forgets the landmark directions or simply enters a different behavior state (but still remembers the landmark directions). It also remains unclear whether the memory is stored in another neural circuit and the EB merely provides a reference frame for orientation, or whether the activity bump in the EB represents the actual memory of the landmarks. The current experimental setup is not able to address this issue. A novel task that can disassociate these two components is required for further investigation (Han, 2021).

The present study concludes the following. First, the experiment indicated that long-term suppression of EIP-ring neurons reduced the accuracy of orientation working memory (fixation with an increased deviation angle), whereas long-term suppression of P-ring neurons abolished the memory completely (no fixation). Similarly, transiently activating either ring neuron types in the absence of landmark immediately abolished the memory. Second, the experimental observation can be explained by the EB-PB neural circuit model in which the EIP-ring neurons are responsible for controlling the width of the bump and the P-ring neurons are responsible for shifting (updating) the position of the bump. Third, put the experiment and the theory together, the present study suggests that coordinated activation of the two ring neuron types which control the downstream EB-PB subcircuits is crucial for spatial orientation working memory (Han, 2021).

Olfactory learning and conditioning in the fruit fly is typically modelled by correlation-based associative synaptic plasticity. It was shown that the conditioning of an odor-evoked response by a shock depends on the connections from Kenyon cells (KC) to mushroom body output neurons (MBONs). Although on the behavioral level conditioning is recognized to be predictive, it remains unclear how MBONs form predictions of aversive or appetitive values (valences) of odors on the circuit level. This study presents behavioral experiments that are not well explained by associative plasticity between conditioned and unconditioned stimuli, and two alternative models are suggested for how predictions can be formed. In error-driven predictive plasticity, dopaminergic neurons (DANs) represent the error between the predictive odor value and the shock strength. In target-driven predictive plasticity, the DANs represent the target for the predictive MBON activity. Predictive plasticity in KC-to-MBON synapses can also explain trace-conditioning, the valence-dependent sign switch in plasticity, and the observed novelty-familiarity representation. The model offers a framework to dissect MBON circuits and interpret DAN activity during olfactory learning (Zhao, 2021).

Predicting the future from sensory input is fundamental for survival. Co-appearing stimuli can be used for improving a prediction, or for predicting important events themselves, as observed in classical conditioning. In fruit fly odor conditioning, an odor that will become the conditioned stimulus (CS), is paired with the unconditioned stimulus (US), in this study an electroshock, that triggers an avoidance behavior and in internal representation of a negative value (valence). After conditioning, and the negative value representation (although not the full unconditioned response) previously elicited by the electroshock will be reproduced by the odor itself. Classical conditioning theories posit that throughout learning the odor becomes predictive for the electroshock. During learning, the prediction error decreases, and learning stops when the predictive odor value matches the strength of the electroshock (Zhao, 2021).

Predictive olfactory learning in fruit flies is a widely recognised concept in the experimental literature, and dopaminergic neurons (DANs) in the mushroom body (MB) have been suggested to predict punishment or reward. Yet, despite the acknowledgment of its predictive nature, computational models on fruit fly conditioning are mostly guided by the formation of associations, a notion that relates more to memories rather than predictions. Similarly, the concept of predictive learning is well recognized for olfactory conditioning in insects in general, but synaptic plasticity models are not formulated in terms of explicit predictions, but rather in terms associations and correlations, with plasticity being driven by two or three factors, each representing a temporal nonlinear function of the pre- or postsynaptic activities or of a modulatory signal, sometimes combined with homeostatic plasticity. This type of associative models exist for fruit flies, locusts or honey bees. They differ from target learning, where the unconditioned stimulus sets a target that is learned to be reproduced by the conditioned stimulus. Target learning becomes predictive learning when including a temporal component. It involves a difference operation, and learning stops when the target is reached. The stop-learning feature is difficult to be reproduced by purely correlation-based associative learning, while a purely predictive model also intrinsically captures associative properties (Zhao, 2021).

Associative learning was suggested to be implemented through spike- or stimulus-timing dependent plasticity (STDP) that would underlie conditioning. STDP strengthens or weakens a synapse based on the temporal correlation between the US (electroshock) and the CS (odor), both on the neuronal time scale of 10s of milliseconds and on the behavioral time scale of 10s of seconds. Whether an association is strengthened by just repeating the pairing until the behavioral saturation is reached, or the association saturates due to a faithful prediction, however, has not been investigated in the fruit fly so far. This study shows that olfactory conditioning in Drosophila is better captured as predictive plasticity that stops when a US-imposed target is reached, rather than by correlation-based plasticity, such as STDP, that does not operate with an explicit error or a target. According to this scheme, it is only the aversive/appetitive value of the US that is predicted by the CS after faithful learning, not the US itself. Based on the common value representation in the mushroom body output neurons (MBONs), the corresponding avoidance/approach reaction as one aspect of the unconditioned response is elicited by the CS alone (Zhao, 2021).

The Drosophila olfactory system represents a unique case for studying associative /predictive learning, and the MB is known to be essential in olfactory learning. The Kenyon cells (KCs) receive olfactory input from olfactory projection neurons and form a sparse representation of an odor. The parallel axons of the Kenyon cells (KCs) project to the MB lobes, along which the compartmentalized dendritic arbors of the MB output neurons (MBONs) collect the input from a large number of KCs. Reward or punishment activates specific clusters of DANs PAM and PPL1, respectively which project to corresponding compartments of the MB lobes, modulating the activity of the MBONs and the behavioral response. Recently, a detailed mapping of the MB connectome has been accomplished for larvae and of the vertical lobe for the adult Drosophila. Several studies show that not only the feedforward modulation from DANs to MBONs, but also the feedback from MBONs to DANs play an important role in olfactory learning (Zhao, 2021).

Previous studies have given insights into the possible cellular and subcellular mechanisms of olfactory conditioning. Yet, the suggested learning rules remain correlation-based and miss the explicit predictive element postulated by the classical conditioning theories. This study presents distinctive conditioning experiments showing that olfactory learning is best explained by predictive plasticity. These experiments, in contrast, could not be reproduced by various types of correlation-based associative learning rules. A mathematical model captures the new and previous data on olfactory conditioning, including trace conditioning. The model encompasses the odor/shock encoding and the learning of the aversive odor value with the stochastic response. It is further suggested how the predictive plasticity could be implemented in the MB circuit, with MBONs encoding the value ('valence') of the odor stimulus, and DANs calculating either the error or the target that drives the KC-to-MBON plasticity. The predictive plasticity rule for the KC-to-MBON synapses is shown to be consistent with the experimental results showing the involvement of these synapses in the novelty-familiarity representation (Zhao, 2021).

This study has reconsidered classical odor conditioning in the fruit fly and presents experimental and modeling evidence showing that olfactory learning, also on the synaptic level, is better described as predictive rather than associative. The key observation is that repetitive and time-continuous odor-shock pairing stops strengthening the conditioned response after roughly 1 minute of pairing, even if the shock intensity is below the behavioral saturation level. During conditioning, the odor is learned to predict the co-applied shock stimulus. As a consequence, the odor-evoked avoidance reaction stops strengthening at a level that depends on the shock strength, irrespective of the pairing time beyond 1 min. Associative synaptic plasticity, defined by a possibly nonlinear function of the CS-US correlation strength, as suggested by STDP models, fails to reproduce the early saturation of learning (Zhao, 2021).

A simple phenomenological model for predictive plasticity is suggested according to which synapses change their strength proportionally to the prediction error. This error is expressed as a difference between the internal shock representation and the value representation of the odor. The model encompasses a description of the shock and value representation, the stochastic response behavior of individual flies, and the synaptic dynamics (using a total of 5 parameters). It faithfully reproduces the conditioning experiments (with a total of 28 data points from 3 different types of experiments) as well as previously studied trace conditioning experiments (without need for further fitting). As compared to the associative rules (Hebbian, linear and nonlinear STDP, covariance rule), the predictive plasticity rule obtained the best fits with the least number of parameters. The model was further compared by the Akaike information criterion that considers the number of parameters beside the fitting quality. This criterion yields a likelihood for the predictive plasticity rule to be the best one that is at least 7 orders of magnitude larger as compared to the other four associative rules considered (Zhao, 2021).

The same phenomenological model of predictive learning may be implemented in two versions by the recurrent MB circuitry. In both versions the MBONs code for the odor value ('valence') that drives the conditioned response. For the error-driven predictive plasticity, the DANs directly represent the shock-prediction error by comparing the shock strength with its MBON estimate, and this prediction error modulates the KC-to-MBON plasticity (see Suggested implementation of error- and target-driven predictive plasticity). For the target-driven predictive plasticity, the DANs represent the shock stimulus itself that is then provided as a target for the KC-to-MBON plasticity. In this target-driven predictive learning, the DANs may also learn to predict the shock stimulus based on the MBON feedback, preventing a fast extinction of the KC-to-MBON memory (Zhao, 2021).

Predictive plasticity for both types of implementation has its experimental support. In general, MBON activity is well recognized to encode the aversive or appetitive value of odors and to evoke the corresponding avoidance or approach behavior, while KC-to-MBON synapses were mostly shown to undergo long-term depression, but also potentiation. DAN responses are shown to be involved in both the representation of punishment and reward that drive the aversive or appetitive olfactory conditioning. This conditioning further involves the recurrent feedback from MBONs to DANs that may be negative or positive. Moreover, the connectome from the larvae and adult fruit fly MBON circuit reveals feedback projections from DANs to the presynaptic side on the KC and the postsynaptic side on the MBONs at the KC-to-MBON synaptic connection giving different handles to modulate synaptic plasticity (Zhao, 2021).

With regard to the specific implementations, the error-driven predictive plasticity is consistent with the observation that DAN activity decreases during the conditioning. The two models have opposite predictions for learning while blocking MBON activity. The error-driven predictive plasticity would yield a higher learning index (LI) while the target-driven predictive plasticity would yield a lower LI. It was also shown that some DANs increased their activity with learning while other DANs, in the same PPL1 cluster that is supposed to represent aversive valences, decreased their activity. In fact, error- and target-driven predictive plasticity may both act in concert to enrich and stabilize the representations. DAN activity would decrease in those DANs involved in error-driven predictive plasticityand increase in those involved in target-driven predictive plasticity (Zhao, 2021).

While error-driven predictive plasticity offers access to an explicit error representation in DANs, target-driven predictive plasticity has its own merits. If DANs and MBONs code for similar information, they can support a positive feedback-loop to represent a short-term memory beyond the presence of an odor or a shock, as it was observed for aversive valences in PPL1 DANs and for appetitive PAM DANs. A positive feedback-loop between MBONs and DANs is further supported by the persistent firing between these cells after a rejected courtship that may consolidate memory of the rejection, linked to a specific pheromone (Zhao, 2021).

Target-driven plasticity has further functional advantages in terms of memory retention time. Any odor-related input to the DANs, arising either through a forward hierarchy from KC48 or a recurrence via MBONs to the DANs, will extend the memory life-time in a 2-stage prediction process: the unconditioned stimulus (s) that drives the DAN activity (d) to serve as a target for the value learning in the MBONs via KC-to-MBON synapses, will itself be predicted in the DANs. Extending the memory life-time through circuit plasticity might be attractive under the light of energy efficiency, showing that long-term memory in a synapse involving de novo protein synthesis can be costly, while cheaper forms of individual synaptic memories likely have limited retention times. Moreover, distributed memory that includes the learning of an external target representation offers more flexibility, including the regulation of the speed of forgetting (Zhao, 2021).

Target-driven predictive plasticity may also explain the novelty-familiarity representation observed in the recurrent triple of KCs, DANs and MBONs. The distributed representation of valences allows for expressing temporal components of the memories. Spontaneous activity in the KCs and their downstream cells injures a minimal strength of the KC-to-MBON synapses through predictive plasticity. A novel odor that drives KCs will then also drive MBONs and, to a smaller extent (as is assumed in this study), also DANs. If the DANs that represent the target for the KC-to-MBON plasticity are only weakly activated by the odor, the KC-to-MBON synapses learn to predict this weaker activity and depress. The depression results in a repetition suppression of MBONs and the corresponding familiarization of the fly to the ongoing odor. However, when the odor is cleared away, the MBON activity induced by spontaneously active KCs via depressed synapses now becomes lower than the spontaneous DAN activity, and predictive plasticity recovers the original synaptic strength. Eventually the spontaneous MBON and DAN activites match again and the response to the originally novel odor is also recovered, as seen in the experiment (Zhao, 2021).

Olfactory learning is likely distributed across several classes of synapses in the MB. The acquisition of olfactory memories was shown to be independent of transmitter release in KC-to-MBON synapses, although the behavioral recall of these memories required the intact transmission. In fact, learning may also be supported by plasticity upstream of the MBONs such that the effect of blocking KC-to-MBON transmission during learning is behaviorally compensated. Predictive plasticity at the KC-to-MBON synapses requires the summed synaptic transmissions across all synapses to be compared with the target d, also during the memory acquisition. This type of plasticity would therefore be impaired by blocking the release (Zhao, 2021).

Distributed learning also offers flexibility in acquiring predictions from new cues. While the original Rescorla-Wagner rule would predict blocking, this has not been observed in the fruit fly. Blocking refers to the phenomenon that, if the first odor of a compound-CS is pre-conditioned, the second odor of the compound will not be learned to become predictive for the shock. Because the predictive plasticity rules are expressed at the neuronal but not at the phenomenological level, predictions about blocking will depend on the neuronal odor representation. If the two odors activate the same MBONs, blocking would be observed since the MBONs are already driven to the correct value representation by the first odor. If they activate different MBONs, however, blocking would not be observed since the MBONs of the second odor did not yet have the chance to learn the correct value during the first conditioning. Hence, since blocking has not been observed in the fruit fly, it is postulated that the odors of the compound-CS in these experiments were represented by different groups of MBONs (Zhao, 2021).

How does this model relate to the concentration-specificity and the timing-specifity of odor conditioning? First, olfactory learning was found to be specific to the odor concentration, with different concentrations changing the subjective odor identity. The response behavior was described to be non-monotonic in the odor intensity, with the strongest response for the specific concentration the flies were conditioned with. It was suggested that this may arise from a non-monotonic odor representation in the KC population as a function of odor intensity. Given such a presynaptic encoding of odor concentrations, the predictive olfactory learning in the KC-to-MBON connectivity would also inherit the concentration specificity from the odor representation in the KCs. This predictive plasticity, and also the Rescorla-Wagner model, further predicts that learning with a higher odor concentration (but the same electroshock strength) only speeds up learning, but would not change the asymptotic performance (Zhao, 2021).

Second, olfactory conditioning was also shown to depend on the timing of the shock application before or after the conditioning odor. While a shock application 30s after an odor assigns this odor an aversive valence, an appetitive valence is assigned if the shock application arises 30s before the odor presentation. Modeling the approaching behavior in the context of predictive plasticity would require duplicating this model to also represent appetitive valences, and the action selection would depend on the difference between aversive and appetitive valences. Inverting the timing of CS and US may explain 'relief learning' if a stopping electroshock would cause a decrease of the target for aversive MBONs and an increase of the target for appetitive MBONs. An odor presented after the shock would then predict the increased appetitive target and explain the relief from pain behavior, similar to the model of relief learning in humans (Zhao, 2021).

Overall, these behavioral experiments and the plasticity model for the KC-to-MBON synapses support the notion of predictive learning in olfactory conditioning, with the DANs representing either the CS-US prediction error or the prediction itself. While predictive coding is recognized as a hierarchical organization principle in the mammalian cortex that explains animal and human behavior it may also offer a framework to investigate the logic of the MB and the multi-layer MBON readout network as studied by various experimental work (Zhao, 2021).

The mushroom body (MB) is a well-characterized associative memory structure within the Drosophila brain. Analyzing MB connectivity using multiple approaches is critical for understanding the functional implications of this structure. Using the genetic anterograde transsynaptic tracing tool, trans-Tango, this study identified divergent projections across the brain and convergent downstream targets of the MB output neurons (MBONs). This analysis revealed at least three separate targets that receive convergent input from MBONs: other MBONs, the fan-shaped body (FSB), and the lateral accessory lobe (LAL). A multilayer circuit is described, both anatomically and functionally, in which inhibitory and excitatory MBONs converge on the same genetic subset of FSB and LAL neurons. This circuit architecture enables the brain to update and integrate information with previous experience before executing appropriate behavioral responses. This use of trans-Tango provides a genetically accessible anatomical framework for investigating the functional relevance of components within these complex and interconnected circuits (Scaplen, 2021).

The MB is a high-level integration center in the Drosophila brain with an established role in learning and memory. The iterative nature of converging and diverging MB neural circuits provides an excellent example of the anatomical framework necessary for complex information processing. For instance, on a rapid timescale, interactions between MB compartments could generate different output patterns to drive behavior, whereas on a slower timescale, interactions between MB compartments could reevaluate memories of a context (Scaplen, 2021).

This study sought to map the projections from the MB using the genetic anterograde transsynaptic technique, trans-Tango. The connectivity of MBONs is reported across multiple subjects in both males and females and the variability in connectivity is highlighted that potentially exists across animals. This study complements the ongoing efforts of EM reconstruction of a whole brain of a single female fruit fly and confirms previous anatomical predictions. Although the complete EM dataset of an adult fly brain has been an invaluable resource that significantly accelerated the mapping of the neural circuits underlying innate and learned behaviors, the massive undertaking of acquiring a full EM dataset renders it impractical to perform for multiple individuals. Thus, trans-Tango, expands the value of the EM reconstruction data by examining circuit connectivity across multiple individuals. Further, trans-Tango can be readily adapted to functional studies in which the activity of the postsynaptic neurons is altered by expressing optogenetic/thermogenetic effectors or monitored by expressing genetically encoded sensors. Tracing studies reported in this study serve as the foundation for these future experiments (Scaplen, 2021).

These studies reveal that the MB circuits are highly interconnected with multiple regions of converging projections both within and downstream of the MB. These experiments also show diverging projections in the downstream postsynaptic targets. A multilayer circuit is identified that includes GABAergic and cholinergic MBONs that converge on the same subset of FSB and LAL neurons. This circuit architecture allows for rapid updating of the online processing of sensory information before executing behavior. Further, this circuit organization is likely a conserved motif among insects (Scaplen, 2021).

Successive levels of convergence and divergence across the brain permit functional flexibility. Like the mushroom body, cerebellar circuits in mammals exhibit large divergence in connectivity, and this can support diverse types of synaptic plasticity. Previous neuroanatomical work in insects described divergent afferent and efferent MB neurons, although the extent of this divergence was unknown. The data revealed varying levels of divergence of postsynaptic connections of MBONs across the brain. Every one of the analyzed MBONs had postsynaptic partners projecting to multiple brain regions. Further, nearly the entire superior protocerebrum as well as portions of the inferior protocerebrum received input from at least one MBON, providing opportunities for comprehensive integration of signals from the MBON network (Scaplen, 2021).

Multiple feedforward and feedback circuits exist within the MB. The current data revealed at least two MBONs that receive convergent input from multiple MBONs and are also reciprocally connected. The convergent MBON input to β'2mp is especially interesting as cholinergic (MBON γ2α'1), GABAergic (MBON γ3β'1), and glutamatergic (MBON γ5β'2a) MBONs drive opposing behaviors. For instance, activation of the cholinergic or GABAergic MBON results in naive odor preference, whereas activation of the glutamatergic MBON results in robust naive avoidance. Similarly, the cholinergic MBON activity mediates aversive associations, whereas glutamatergic MBON activity mediates appetitive associations and extinction of aversive (Scaplen, 2021).

Considering that MBON β'2mp receives convergent input from these parallel and opposing pathways, it likely serves as a decision hub by integrating activity to modulate cue-induced approach and avoidance behavior. How MBON β'2mp integrates information across MBONs and drives behavioral responses remains to be determined. Naive activation of MBON β'2mp does not appear to influence behavioral choice, it instead acts as a sleep suppressor. Inhibition of MBON β'2mp during sleep enhances long-term memory. Separately, local protein synthesis within MBON β'2mp, has been implicated in the consolidation of long-term memory. This makes MBON β'2mp an ideal model for understanding how sleep and memory signals might be integrated at a molecular level. It should be mentioned that MBON γ3β'1 reportedly acts as a sleep activator and local protein synthesis within this MBON is also important for the consolidation of long-term memory. Thus, MBON γ3β'one likely also plays a role in integrating sleep and memory signals through its reciprocal connections MBON β'2mp (Scaplen, 2021).

This provides a well-characterized anatomical framework to understand how opposing memories are acquired, consolidated, expressed and updated. Since the roles of these converging MBONs in naive and learned behaviors are state dependent, it is hypothesized that MBON γ3β'1 and MBON β'2mp, both receiving convergent input from other MBONs, providing opportunities for feedforward networks to update information processing depending on the state of the animal (Scaplen, 2021).

Some of the feedback connections originally hypothesized to exist in the MB were between MBONs and DANs. The current analysis revealed neurons postsynaptic to MBONs that are TH positive. Recent studies that combined EM annotation and calcium imaging to identify specific MBON-DAN connections suggest extensive recurrent connectivity between MBONs and DANs, validating these findings. For example, previous studies using both GFP Reconstitution Across Synaptic Partners (GRASP) and EM annotation revealed that MBON α1 and DAN α1 are synaptically connected. This study similarly identified a few DAN neurons that innervate the horizontal MB lobes within the MBON α1 postsynaptic signal. A recent study showed that the 20 DANs that innervate the γ5 MB compartment are clustered into five different subtypes that innervate distinct anatomical regions within the γ5 compartment. According to this study, only one of the γ5 DANs receives direct recurrent feedback from γ5β'2a MBONs. Based on these recent anatomical characterizations, it is believed that the TH+ neurons within the postsynaptic signal of γ5β'2a are the γ5 DANs (Scaplen, 2021).

The FSB is the largest substructure of the central complex, and it serves as a sensory-motor integration center. The FSB comprises nine horizontal layers that are innervated by large-field neurons. Previous work in blow flies and, later work in Drosophila, predicted that the FSB was postsynaptic to output neurons of the MB. The current data confirm that the large-field, tangential neurons of the dorsal FSB are postsynaptic to the majority of MBONs. Although there exists some variation across brains, glutamatergic and GABAergic MBONs predominately project to FSB layers 4 and 5, whereas cholinergic MBONs mainly project to FSB layer 6. Connections between MBONs and FSB were consistent across different split-GAL4 lines that have overlapping expression patterns. Similar extensive direct connectivity between these MBONs and the dorsal FSB, especially layers 4 and 5, were found in the recently annotated EM hemibrain dataset. Together, these observations suggest that the connectivity between the MB and FSB are structurally, and perhaps in some cases functionally, conserved across insect species (Scaplen, 2021).

How are FSB layers 4/5 and 6 functionally distinct? The dorsal FSB has a well-established role in modulating sleep and arousal, locomotor control, courtship, and visual memory. FSB layer 5 has been specifically implicated in processing information regarding elevation in a foraging- and rutabaga-dependent manner. More recent studies have implicated the dorsal FSB in processing nociceptive information. FSB layer 6 plays a specific role in avoidance of a conditioned odor, whereas layers 4 and 5 respond to aversive stimuli and are responsible for innate, but not conditioned, avoidance. Moreover, recent connectome data suggest that differences exist in the postsynaptic connections of layers 4/5 and 6 as well. Overall, there is high degree of interconnectivity within the FSB. The predominate output of FSB layer 6 neurons are other FSB neurons. In fact, many FSB layer 6 neurons project exclusively to other FSB neurons. In contrast, FSB layer 4 neurons send direct projections to other brain structures -- CRE, SMP, and LAL -- in addition to projecting to other FSB neurons. The connections with the LAL position the FSB layer 4 to directly influence downstream motor output signals prior to executing behavior. Recent EM analysis also suggests that some FSB layer 6 neurons synapse back onto PAM DAN neurons. This connectivity is in line with the associative role in conditioned nociception avoidance described for FSB layer 6 (Scaplen, 2021).

Interestingly, this study found that the pattern of FSB postsynaptic targets of the MBONα1 is dissimilar to other glutamatergic MBONs. FSB layers 4/5 and 6 are not present in the MBON α1 postsynaptic signal. Instead, MBON α1 project to neurons that innervate the ventral and most dorsal aspect of the FSB. The ventral FSB is implicated in innate avoidance of electric shock, and recent data suggest that its activity is tuned to airflow cues for orientation during flight. Artificial activation of MBON α1 does not result in significant avoidance behavior. However, it has been implicated in the acquisition, consolidation, and expression of 24 hr long-term sucrose memory. It is possible that MBON α1 provides appetitive valence signals to the ventral FSB to guide goal-directed flight. Functionally validating the role of MBON α1 and its relationship with its putative downstream neurons is key to appreciating how learning signals can drive behavioral decisions (Scaplen, 2021).

More research is necessary to further understand the functional role of different FSB layers and how information is integrated across these layers. Based on the anatomical data, it is clear that although the MB and FSB can function in parallel during memory formation, they act as parts of a dynamic system to integrate information and adjust behavioral responses (Scaplen, 2021).

The LAL is an important premotor waystation for information traveling from the central complex to descending neurons innervating thoracic motor centers across insects. Accordingly, the LAL has been implicated in orientation to pheromones in the moth, flight in the locust and dragonfly, locomotion in Drosophila stimulus-directed steering in Drosophila, the cockroach, cricket, and moth and in response to mechanosensory stimuli in the locust. In the moth, recordings from neurons innervating the LAL have a characteristic 'flip-flop' firing property, which is thought to mediate walking command. More recent work has suggested a functional organization whereby the neurons in the upper division of the LAL receive convergent input from the protocerebrum and neurons in the lower division generate locomotor command (Scaplen, 2021).

The current data show that the MB network converges with the protocerebrum input, thereby providing an opportunity for MBONs to indirectly influence descending motor outputs. It was also demonstrated that two MBONs (γ3β'1 and γ2α'1) synapse on the same subset of LAL and FSB cells, revealing a convergent circuit that connects both structures. Further, in support of anatomical observations, optogenetic activation of MBON γ2α'1 resulted in activation of both LAL and FSB layer four neurons. Given that MBON γ3β'1 is GABAergic, the equivalent experiment was not performed for this neuron. Thus, understanding the functional consequences of these inhibitory connections will require further investigation. Interestingly, despite the fact that MBON γ3β'1 and γ2α'1 express different neurotransmitters and innervate different MB compartments, their manipulation has similar behavioral phenotypes: both promote sleep, and artificial activation of either results in naive preference. Further, activation of both MBON γ3β'1 and γ2α'1 together has an additive effect, which results in a significant increase in preference (Aso et al., 2014b) (Scaplen, 2021).

The FSB and LAL have a well-established structural and functional connectivity. The LAL integrates information from the central complex, including the FSB, and provides a premotor signal to motor centers. However, the behavioral significance of MBON γ3β'1 and γ2α'1 projections to both the FSB and LAL is less clear. Previous work demonstrated that activation of these MBONs while the flies explored an open arena did not significantly affect average speed or angular speed of individual flies. By contrast, this study found that inactivation of the putative downstream LAL neurons significantly increased overall activity of behaving flies in a social context and locomotor assay. Thus, the γ3β'1 and γ2α'1 MBONs may play a modulatory rather than required role in influencing behavioral response to an associated cue (Scaplen, 2021).

Recent work in Drosophila has demonstrated that the DANs that innervate MBON γ2α'1 regulate flight bout durations, and may provide a motivation signal via MBONs to the FSB and LAL to regulate motor activities. The LAL neurons receive multisensory input, and some LAL neurons make direct connections to descending neurons that control movement. Thus, this circuit organization enables integration of sensory signals with punishment or reward to direct the motion of the animal. In contrast, MBON connections with the FSB might play a role in providing context for flexible navigation, goal-directed actions, and memory-based navigation (Scaplen, 2021).

If homology can be defined by shared expression of transcription factors and similar functional roles, the MB-FSB connection may be an appropriate model for understanding functional connections between the hippocampus and striatum and serve as an accessible model for understanding connectivity between more complex brain structures associated with memory. Further, given that the integrative relay role of the LAL is somewhat reminiscent of the vertebrate thalamus, the complex connectivity between the MBONs, FSB, and LAL may also serve as an effective model for predicting and understanding functional connections between the hippocampus, striatum, and thalamus in the context of memory formation and action selection (Scaplen, 2021).

Insects exhibit a great variety of complex behaviors, and significant effort has been devoted to understand the neural circuits that underlie these behaviors. The genetically accessible Drosophila is a great model for studying the interplay between circuit architecture and behavior owing to their complex yet tractable brains. The MB circuits and their role in learning and memory are among the most studied circuits in Drosophila. Although, the majority of these studies have focused on olfactory memory, it is clear that the MB plays a much broader role in insect behavior. In Drosophila, the MB is important for courtship memory, taste aversive memory as well as visual memory. In cockroaches, the MB has a role in place memory and recent data in two different species of ants implicate the MB in spatial navigation to learned locations using visual cues. In mammals, the hippocampus is similarly required for multiple forms of associative memory, including spatial navigation using visual cues. Thus, cross-species similarity in circuit organization and function may exist between the mushroom body and the hippocampus. However, such anatomical and functional cross-species comparisons can also be made between the mushroom body and the cerebellum, suggesting that similar convergent-divergent architecture may be a general principle of structures that encode and update memories (Scaplen, 2021).

In this context, the implementation of trans-Tango to study the MB has high potential in the era of EM reconstruction of the Drosophila brain. Through examination of the circuit connectivity in several individuals, easily afforded by trans-Tango, the value of the EM reconstruction data could be augmented by overlaying on it potential nuanced differences between individuals. In addition, trans-Tango-mediated discoveries in the fly could help illuminate principles of circuit organization in other species. Further, due to the modular design of trans-Tango, it could be readily reconfigured for other types of studies beyond circuit tracing. For example, only minimal modifications are required for implementing a configuration of trans-Tango for identifying the molecular composition of the postsynaptic partners. This strategy could be used to examine the evidence that MBONs stratify the FSB through different classes of peptidergic neurons. Confirmation of these observations would suggest that the MB plays a critical role in regulating modulatory systems of a midbrain region that shares structural and functional commonalities with the vertebrate basal ganglia. Finally, through combining it with new genome editing strategies, trans-Tango could become a useful tool for comparative anatomy in other insects. This would enable the study of synaptic connections in non-model organisms and lead to deeper understanding of biological diversity (Scaplen, 2021).

Understanding how memories are formed, stored, and retrieved necessitates knowledge of the underlying neural circuits. This characterization of the architecture of the neural circuits connecting the MB with downstream central complex structures lays the anatomical foundation for understanding the function of this circuitry.These studies may also provide insight into general circuitry principles for how information is processed to form memories and update them in more complex brains (Scaplen, 2021).

Active forgetting is an essential component of the memory management system of the brain. Forgetting can be permanent, in which prior memory is lost completely, or transient, in which memory exists in a temporary state of impaired retrieval. Temporary blocks on memory seem to be universal, and can disrupt an individual's plans, social interactions and ability to make rapid, flexible and appropriate choices. However, the neurobiological mechanisms that cause transient forgetting are unknown. This study identified a single dopamine neuron in Drosophila that mediates the memory suppression that results in transient forgetting. Artificially activating this neuron did not abolish the expression of long-term memory. Instead, it briefly suppressed memory retrieval, with the memory becoming accessible again over time. The dopamine neuron modulates memory retrieval by stimulating a unique dopamine receptor that is expressed in a restricted physical compartment of the axons of mushroom body neurons. This mechanism for transient forgetting is triggered by the presentation of interfering stimuli immediately before retrieval (Sadandal, 2021).

Memory formation, consolidation and retrieval are well-known functions that support memory expression; however, the processes that limit these functions -- including forgetting -- are less understood. Forgetting has been characterized as either passive or active, and is crucial for memory removal, flexibility and updating. Memory may be removed completely, resulting in permanent forgetting; or temporarily irretrievable, resulting in transient forgetting (Sadandal, 2021).

One form of active forgetting-known as intrinsic forgetting-involves one dopamine neuron (DAN) that innervates the γ2α'1 compartment of the axons of mushroom body neurons (MBNs) and the dendrites of the downstream, compartment-specific mushroom-body output neurons (MBONs). This DAN resides in a cluster of 12 DANs in each brain hemisphere that is known as the protocerebral posterior lateral 1 (PPL1) cluster. Current evidence indicates that the ongoing activity of these DANs after aversive olfactory conditioning slowly and chronically erodes labile and nonconsolidated behavioural memory, as well as a corresponding cellular memory trace that forms in the MBONs. This intrinsic forgetting mechanism is shaped by external sensory stimulation and sleep or rest, and is mediated by a signalling cascade in the MBNs that is initiated by the activation of the dopamine receptor DAMB, which leads to the downstream activation of the actin-binding protein Cofilin and the postulated reorganization of the synaptic cytoskeleton (Sadandal, 2021).

By contrast, there is little understanding of the mechanisms that arbitrate transient forgetting. Neuropsychological studies of failures or delays in retrieval in humans have primarily focused on lexical access. Phonological blockers or interfering stimuli produce a tip-of-the-tongue state-the failure to recall the appropriate word or phrase. Tip-of-the-tongue states are resolved when the distracting signals dissipate. Several brain regions have been implicated in tip-of-the-tongue states from functional magnetic resonance imaging studies, but the neurobiological mechanisms that produce a temporary state of impaired retrieval are unknown. This study offers an entry point into this area of brain function (Sadandal, 2021).

Memory retrieval has been proposed to consist of an interplay between internal or external cues and memory engrams, with cue-induced reactivation of engrams across multiple regions of the brain facilitating memory expression. But a central question about this process is how interfering stimuli temporarily block memory retrieval, resulting in transient forgetting. This study offers insights into one such mechanism. Behavioural and functional imaging data reveal that PPL1-α2α'2, working through the DAMB receptor expressed in the α2α'2 MBN axonal compartment, mediates the transient forgetting of PSD-LTM. This effect occurs without altering a cellular memory trace in the postsynaptic MBON-α2sc. This process can be triggered by distracting stimuli, illustrating a neural-genetic-environmental interplay that modifies memory expression (Sadandal, 2021).

This study considered why the cellular memory trace remains unaffected by DAN stimulation despite the occurrence of behavioural forgetting. Because blocking synaptic output from MBON-α2sc reduces PSD-LTM expression, the simplest hypothesis posits that cellular memory traces form with conditioning in the MBON in addition to the cytoplasmic Ca2+-based memory trace that was detected in this study. This is expected: neurons undergo broad changes in physiology as they adopt new states, so it is plausible that such plastic mechanisms-especially ones that gate synaptic release-are inactivated by DAN activity while leaving the Ca2+-based memory trace intact (Sadandal, 2021).

The discovery that loss of function of DAMB leads to enhanced PSD-LTM was surprising, because of a previous study reporting that this insult attenuates PSD-LTM. The experiments argue strongly that DAMB functions normally to suppress expression of PSD-LTM. However, this leads to the question of why a receptor involved in transient forgetting would lead to enhanced PSD-LTM when inactivated. Previous experiments have shown that PPL1-α2α'2-similar to PPL1-γ2α'1-exhibits ongoing activity, leading to a slow release of dopamine onto MBNs. This activity should slowly degrade or suppress existing memory so that when the receptor is inactivated memory expression is enhanced (Sadandal, 2021).

PPL1-α2α'2 has no important role in the forgetting of labile nonconsolidated memory. Instead, previous studies have identified a different DAN (PPL1-γ2α'1) as having a role in this process and the apparent erasure of the downstream cellular memory trace-perhaps an indication of 'permanent forgetting'. This process is modulated by internal and external factors, and is mediated by key molecules expressed in the MBN that receive PPL1-γ2α'1 input. No robust decrement was found in expression of PSD-LTM after PPL1-γ2α'1 stimulation, which points to the existence of two separate dopamine-based circuits for permanent and transient forgetting. This functional separation may indicate a fundamental principle in the organization of circuits that mediate several forms of forgetting (Sadandal, 2021).

However, the DAMB receptor is used for both permanent and transient forgetting. DAMB is widely expressed across the MBN axons but alters synaptic plasticity differently across MBN compartments. It is possible that DAMB signalling may be distinct for the two forms of forgetting. DAMB preferentially couples with Gq, the knockdown of which inhibits the potent erasure of memory, but its potential role in transient forgetting is unknown. The scaffolding protein Scribble orchestrates the activities of Rac, Pak and Cofilin, all of which are important for the permanent forgetting pathway. However, Scribble knockdown or inhibition of Rac1 does not enhance the PSD-LTM as is the case in DAMB-knockdown flies, which suggests that this scaffolding signalosome does not have a large role in transient forgetting. In summary, the two distinct forms of forgetting-transient and permanent-share a dopaminergic mechanism and a common dopamine receptor, but differ in upstream and downstream neural circuits and in downstream signalling pathways within MBNs (Sadandal, 2021).

Making inferences about the computations performed by neuronal circuits from synapse-level connectivity maps is an emerging opportunity in neuroscience. The mushroom body (MB) is well positioned for developing and testing such an approach due to its conserved neuronal architecture, recently completed dense connectome, and extensive prior experimental studies of its roles in learning, memory and activity regulation. This study identified new components of the MB circuit in Drosophila, including extensive visual input and MB output neurons (MBONs) with direct connections to descending neurons. Unexpected structure was found in sensory inputs, in the transfer of information about different sensory modalities to MBONs, and in the modulation of that transfer by dopaminergic neurons (DANs). This study provides insights into the circuitry used to integrate MB outputs, connectivity between the MB and the central complex and inputs to DANs, including feedback from MBONs. The results provide a foundation for further theoretical and experimental work (Li, 2020).

Understanding how memories of past events are formed and then used to influence ongoing behavior are key challenges in neuroscience. It is generally accepted that parallel changes in connection strength across multiple circuits underlie the formation of a memory and that these changes are integrated to produce net changes in behavior. Animals learn to predict the value of sensory cues based on temporal correlations with reward or punishment. Such associative learning entails lasting changes in connections between neurons. It is now clear that different parts of the brain process and store different aspects of the information learned in a single event. In both flies and mammals, dopaminergic neurons play a key role in conveying information about whether an event has a positive or negative valence, and there are compelling parallels between the molecular diversity of dopaminergic cell types across these evolutionarily distant animals. However, there is only a limited understanding of how information about the outside world or internal brain state reaches different dopaminergic populations. Nor is the nature of the information that is stored in each parallel memory system understood, nor how these parallel memories interact to guide coherent behavior. It is believed such processes are governed by general and evolutionarily-conserved principles. In particular, it is believed the circuit logic that allows a brain to balance the competing demands of rapid learning with long-term stability of memory are likely to be the same in flies and mammals. Developing a comprehensive understanding of these circuits at the resolution of individual neurons and synapses will require the synergistic application of a variety of experimental methods together with theory and modeling. Many of the required methods are well developed in Drosophila, where the circuits underlying learning and memory are less complex than in mammals, and where detailed anatomical knowledge of the relevant circuits, which are believed will be essential, has just now become available. This study provide analysis of the complete connectome of a circuit involved in parallel processing of associative memories in adult fruit flies. The core architecture of this circuit is strikingly similar to that of the vertebrate cerebellum (see The shared circuit architecture of the mushroom body and the cerebellum) (Li, 2020).

The MB is the major site of associative learning in insects, and species that perform more complex behavioral tasks tend to have larger MBs. In the MB of each brain hemisphere, sensory stimuli are represented by the sparse activity of ~2000 Kenyon cells (KCs) whose dendrites form a structure called the MB calyx and whose parallel axonal fibers form the lobes of the MB (see Anatomy of the adult Drosophila MB) (Li, 2020).

The major sensory inputs to the Drosophila MB are olfactory, delivered by ~150 projection neurons (PNs) from the antennal lobe to the dendrites of the KCs in the MB calyx. KCs each receive input from an average of six PNs. For a KC to fire a spike, several of its PN inputs need to be simultaneously activated. This requirement, together with global feedback inhibition, ensures a sparse representation where only a small percentage of KCs are activated by an odor. The MB has a three layered divergent-convergent architecture in which the coherent information represented by olfactory PNs is expanded and decorrelated when delivered to the KCs. But the degree to which the structure of the sensory input representation is maintained by the KCs has been debated. This issue is explored, taking advantage of a nearly comprehensive dataset of KC inputs and outputs (Li, 2020).

While best studied for its role in olfactory associative learning, the MB also receives inputs from several other sensory modalities. A subset of projection neurons from the antennal lobe delivers information about temperature and humidity in both the larva. Taste conditioning also requires the MB and is believed to depend on specific KC populations, although the relevant inputs to these KCs have not yet been reported. This study identified one likely path for gustatory input to the MB (Li, 2020).

Drosophila MBs are also known to be able to form memories based on visual cues. Until a few years ago, it was thought that visual input reached the Drosophila MBs using only indirect, multisynaptic pathways as direct visual input from the optic lobes to the MBs, well known in Hymenoptera, had not been observed in any dipteran insect. A previous study identified two types of visual projection neurons (VPNs) connecting the optic lobes and the MB and additional connections have been observed recently by light microscopy. This study found that visual input was much more extensive than previously appreciated, with about 8% of KCs receiving predominantly visual input, and present in this study is a detailed description of neuronal pathways connecting the optic lobe and the MB. Visual sensory input appears to be segregated into distinct KC populations in both the larva and the adult, as is the case in honeybees. This study found two classes of KCs that receive predominantly visual sensory input, as well as MBONs that get the majority of their input from these segregated KC populations (Li, 2020).

MBONs provide the convergence element of the MB's three layer divergent-convergent circuit architecture. Previous work has identified 22 types of MBONs whose dendrites receive input from specific axonal segments of the KCs. The outputs of the MBONs drive learned behaviors. Approximately 20 types of dopaminergic neurons (DANs) innervate corresponding regions along the KC axons and are required for associative olfactory conditioning. Specifically, the presynaptic arbors of the DANs and postsynaptic dendrites of the MBONs overlap in distinct zones along the KC axons, defining the 15 compartmental units of the MB lobes. A large body of evidence indicates that these anatomically defined compartments of the MB are the units of associative learning (Li, 2020).

The DANs innervating different MBON compartments appear to play distinct roles in signaling reward vs. punishment, novelty vs. familiarity, the presence of olfactory cues and the activity state of the fly. These differences between DAN cell types presumably reflect in large part the nature of the inputs that each DAN receives, but knowledge of these inputs is just emerging and is far from comprehensive. DANs adjust synaptic weights between KCs and MBONs with cell type-specific rules and, in at least some cases, these differences arise from the effects of co-transmitters. In general, a causal association of KC responses with the activation of a DAN in a compartment results in depression of the synapses from the active KCs onto MBONs innervating that compartment. Different MB compartments are known to store and update non-redundant information as an animal experiences a series of learning events. In rodent and primate brains, recent studies have revealed that dopaminergic neurons are also molecularly diverse and encode prediction errors and other information based on cell type-specific rules (Li, 2020).

MBONs convey information about learned associations to the rest of the brain. Activation of individual MBONs can cause behavioral attraction or repulsion, according to the compartment in which their dendrites arborize. The combined output of multiple MBONs is likely to be integrated in downstream networks, but it is not understood how memories stored in multiple MB compartments alter these integrated signals to guide coherent and appropriate behaviors. Prior anatomical studies implied the existence of multiple layers of interneurons between MBONs and descending motor pathways. What is the nature of information processing in those layers? Anatomical studies using light microscopy provided the first hints. MBONs from different compartments send their outputs to the same brain regions, suggesting that they might converge on shared downstream targets. DANs often project to these same brain areas, raising the possibility of direct interaction between MBONs and DANs. The functional significance of such interactions has just begun to be investigated, and studies of the Drosophila larva, where a connectome of a numerically less complex MB is available, are providing valuable insights (Li, 2020).

The recently determined connectome of a portion of an adult female fly brain (hemibrain) provides connectivity data for ~22,500 neurons. Among them, ~2600 neurons have axons or dendrites in the MB, while ~1500 neurons are directly downstream of MBONs (using a threshold of 10 synapses from each MBON to each downstream target) and ~3200 are upstream of MB dopaminergic neurons (using a threshold of five synapses from each upstream neuron to each DAN). Thus this study will consider approximately one-third of the neurons in the central brain in this analysis of the MB ensemble (Li, 2020).

Throughout the paper synaptic thresholds were set in order to focus the descriptions and analyses on the most strongly connected neurons. In the above analysis, a higher threshold for MBON connections was chosen to downstream targets than for DAN inputs because the typical MBON has many more output synapses than a DAN has input synapses. At a thresholds of five synapses, DANs have a median of 31 different input neurons, but if the threshold to 10 synapses were increased this would decrease to only six different neurons. In contrast, at the threshold of 10 synapses, MBONs are connected to a median of 90 downstream neurons. There were some limitations resulting from not having a wiring diagram of the full central nervous system, as complete connectivity information for neurons with processes that extended outside the hemibrain volume was lacking. It was generally possible to mitigate these limitations by identifying the corresponding neurons in other EM or light microscopic datasets when the missing information was important for the analyses. Thus the hemibrain dataset was able to support a nearly comprehensive examination of the full neural network underlying the MB ensemble (Li, 2020).

Studies of the larval MB are providing parallel information on the structure and function of an MB with most of the same cell types, albeit fewer copies of each. The microcircuits inside three MB compartments in the adult were previously described and this study reports that the overall organization of these three compartments is conserved in a second individual of a different gender. More importantly, this study extends the analysis of microcircuits within the MB lobes to all 15 compartments, revealing additional aspects of spatial organization within individual compartments (Li, 2020).

The current study led to discovery of new morphological subtypes of KCs and to determine the sensory inputs delivered to the dendrites of each of the ~2000 KC. Considerable structure was found in the organization of those inputs and unexpectedly high levels of visual input, which was the majority sensory input for two classes of KCs. This segregation of distinct sensory representations into channels is maintained across the MB, such that MBONs, by sampling from different KCs, have access to different sensory modalities and representations. A new class of 'atypical' MBONs was discovered, consisting of 14 cell types, that have part of their dendritic arbors outside the MB lobes, allowing them to integrate input from KCs with other information; at least five of them make strong, direct synaptic contact onto descending neurons that drive motor action. How MBONs from different compartments interact with each other to potentially integrate and transform the signals passed from the MB to the rest of the brain is described, revealing a number of circuit motifs including multi-layered MBON-to-MBON feedforward networks and extensive convergence both onto common targets and onto each other through axo-axonal connections. Finally, this study analyzed the inputs to all 158 DANs that innervate the MB. Extensive direct feedback was found from MBONs to the dendrites of DANs, providing a mechanism of communication within and between MB compartments. Groups of DANs were found that share common inputs, providing mechanistic insights into the distributed parallel processing of aversive and appetitive reinforcement and other experimental observations (Li, 2020).

The hemibrain dataset contains the most comprehensive survey of the cell types and connectivity of MB neurons available to date. This connectome allowed probing in fine detail the circuitry underlying canonically proposed functions of the MB, including the representation of olfactory information by KCs, computation of valence by MBONs, and reinforcement of associations by DANs. Patterns in the input to DANs, MBON-to-DAN and MBON-to-MBON connectivity were found that suggest how associative learning in the MB can affect both the acquisition of new information through learning and the expression of previously learned responses. The connectome also reveals circuitry that supports non-canonical MB functions, including selective structure in non-olfactory pathways, a network of atypical MBONs, extensive heterogeneity in DAN inputs, and connections to central brain areas involved in navigation and movement (Li, 2020).

This analysis of the hemibrain connectome relied heavily on an extensive catalog of previously identified and genetically isolated cell types and on decades of study illuminating the link between MB physiology and fly behavior. It is worth emphasizing the interdependency of anatomy, physiology and behavior at the beginnning the post-connectomic era in fly research. Some of the neurons described that appear similarly connected may turn out to have diverse functions due to different physiology and, conversely, neurons that are morphologically distinct may turn out to have similar functions. In addition, a set of inputs with high synapse counts might appear, at the connectome level, to represent a major pathway for activating a particular neuron, but this will not be true if these inputs rarely fire at the same time. Likewise, a set of highly correlated inputs can be effective even if their individual synapse counts are modest. Lack of knowledge about correlated activity is probably the most significant uncertainty when attempting to map synapse counts onto circuit function, likely larger than possible errors in synapse identification and the effects of imposing various thresholds (see below). Finally, the connectome does not reveal gap junction connections or identify more distant non-synaptic modulation. These caveats should be kept in mind when interpreting connectome data (Li, 2020).

A number of these studies involved imposing a cutoff on the number of synapses required to include a particular connection in the analysis. The intent of these thresholds is to focus the analysis on what are likely to be the strongest inputs and outputs. Whether this cutoff should be based on a fixed number of synapses or on a percentage of total synapse counts is open to debate as, of course, is the actual value that the cutoff should take. Neurons differ widely in their numbers of inputs and outputs and these differences need to be taken into account when choosing thresholds. That is why information is presented about percentage of total inputs and outputs as well as synapse numbers. An approach to thresholding that takes such information into account has been used in a previous stdu; in the case of MBON to CX connections explored both here and in the previous study, differences in the connectivity map obtained with different thresholding methods were limited to the weakest connections (Li, 2020).

Most neurons make a large number of weak connections to other neurons, often involving just a single synapse. Some cases of low synapse number may be the result of incomplete reconstruction of neuronal arbors. In the MB lobes, an extensive effort was made to fully reconstruct all neurites and >80% of all computationally predicted synapses were assigned to identified neurons. Likewise, the arbors of MBONs and DANs outside the MB were extensively proofread. However, in other brain regions, where MBON outputs and DAN inputs lie, reconstruction was often much less complete. In such regions, it is difficult to estimate the extent to which the mapped synapses are representative of the full set of connections. A previous study reports an analysis of connectivity determined at two stages of reconstruction in the CX and, while the number of assigned synapses increased with additional proofreading, there was little difference in the connectivity maps. Errors in either synapse prediction or developmental wiring are also likely to produce some false connections represented by only one or two synapses. Of course, real, but sparse, connections might still be impactful if they fire concurrently, but this is not something that can be judged from the available information. Therefore, it seemed reasonable to ignore weak connections in this analyses. Thresholds must obviously be chosen with care and the effects of any particular cutoff value on results and conclusions should be assessed, preferably in conjunction with experimental data (Li, 2020).

Fly research has been greatly facilitated by the development of numerous fly lines that provide cell-type specific genetic access. This analysis has revealed, particularly in the case of DANs, subtypes within groups of neurons that would previously have been considered a single type. Thus, the higher resolution view of cell types that connectomics provides points out the need to develop driver lines or other experimental methods for more fine-tuned genetic access (Li, 2020).

The cell type constituents and circuit motifs of the MB in the adult fly have many similarities with its precursor at the larval stage of development. Both the larval and adult MBs support associative learning and, in both, PNs from the antennal lobe that convey olfactory information provide the majority of the sensory input, complemented by thermal, gustatory and visual sensory information that is segregated into distinct KC populations. However, the multi-layered organization of non-olfactory inputs in the main and accessory calyces (including integration of diverse input sources by LVINs) suggests that the KC representation in the adult is more highly enriched and specialized for non-olfactory sensory features. It is worth noting that the earliest born types of each of the three main adult KC classes (KCγs, KCγd, KCγt, KCα'β'ap1, KCαβp) appear to be specialized for non-olfactory sensory cues, and in most cases their dendrites lie in the accessory calyces (Li, 2020).

In the first instar larval MB, the only larval stage for which a connectome is available, there are roughly 70 mature KCs, which is increased nearly 30-fold in the adult. This enrichment likely increases odor discrimination and olfactory memory capacity. The larval MB has only eight compartments in its horizontal and vertical lobes. Although the increase in the adult to 15 compartments is only about a factor of two, the extent of their DAN modulation is greatly expanded. The larva has only seven confirmed DANs and five additional cells of unknown neurotransmitter thought to provide modulatory input, a factor of 15-fold less than the adult. Whereas each larval KC innervates all eight compartments, individual adult KCs innervate only five out of 15 compartments. Therefore the DANs in the larva are capable of modulating all KCs, whereas in the adult, DANs in different compartments modulate specific subtypes of KCs. The expansion of the number of DANs within many compartments in the adult MB, and subcompartmental targeting of individual DANs within a compartment, further increases the difference in granularity of DAN modulation between the larva and adult (Li, 2020).

MBONs feedback onto DANs and converge onto common downstream targets in both the larva and adult, implying some shared computational strategies. However, the greatly increased DAN complexity in the adult fly and the presence of subcompartmental organization of DAN axon targets not present in the larva suggest a substantial increase in the specificity of the learning signals involved in memory formation and raises the possibility of modality-specific learning signals to complement the multimodal KC representation (Li, 2020).

Previous theoretical work has emphasized the advantages of mixing sensory input to the KCs so that they provide a high-dimensional representation from which MBONs, guided by DAN modulation, can derive associations between sensory input and stimulus-valence. The connectome data modifies this viewpoint in two ways. First, although olfactory input to the KCs is highly mixed, various structural features reduce the dimensionality of the KC odor representation. A recent analysis of EM data from an adult Drosophila MB identified groups of PNs thought to represent food odors that are preferentially sampled by certain KCs. Consistent with this observation, the current analysis revealed subtype-specific biases in PN sampling by KCs, including an overrepresentation of specific glomeruli by α/β and α'/β' KCs. These biases, as well as other structural features, appear to arise from the stereotyped arrangement of PN axons within the calyx and their local sampling by KCs. This may reflect a developmental strategy by which the KC representation is organized to preferentially represent particular PN combinations. Further analyses of KC connectomes across hemispheres and animals, as well as experimental studies, will help evaluate the impact of this structure (Li, 2020).

The hemibrain connectome also revealed a second, more dramatic, structural feature of sensory input to the MB: non-olfactory input streams corresponding to visual and thermo-hygro sensation are strongly segregated. This organization may reflect the nature of stimulus-valence associations experienced by flies. In purely olfactory learning, when valence is associated with particular sets of olfactory receptors, MBONs need to be able to sample those combinations to successfully identify the stimulus. This requires that KCs mix input from multiple glomeruli within the olfactory stream. However, KCs that mix across streams corresponding to different sensory modalities may not be necessary if each modality can be used separately to identify valence. For example, either the visual appearance of an object or its odor may individually be sufficient to identify it as a food source. This study asked whether there is an advantage in having separate modality-specific sensory pathways, as seen in the connectome data, when valences can be decomposed in this way (Li, 2020).

To address this question a model is considered in which KC input is divided into two groups, visual and olfactory. In one version of the model ('shared KC population'), all KCs receive sparse random input from both types of PNs, corresponding to a high degree of mixing across modalities. In the other ('separate KC populations'), half of the KCs receive sparse random input exclusively from visual PNs and half only from olfactory PNs, corresponding to no cross-modal mixing. Two tasks were defined that differ in the way valences are defined. In the first (factorizable) version of the task, each olfactory stimulus and each visual stimulus is assigned a positive or neutral subvalence, and the net valence of the combined stimulus is positive if either of these components is positive (olfactory OR visual). In the second (unfactorizable) version, a valence is randomly assigned independently to each olfactory+visual stimulus pair. The ability of a model MBON, acting as a linear readout, was evaluated to determine valence from these two different KC configurations. Separate modality-specific KC populations are indeed beneficial when the valence can be identified from either one modality or the other. Dedicating different KC subtypes to distinct sensory modalities allows the predictive value of each modality to be learned separately. This result suggests that the divisions into KCs specialized for visual, thermo/hygro and olfactory signals may reflect how natural stimuli of different modalities are predictive of valence (Li, 2020).

On the basis of light-level studies, DAN modulation of KC-to-MBON synapses has been considered to operate at the resolution of MB compartments. However, taken with other recent studies, the morphology and connectivity data indicate that functionally distinct PAM-DAN subtypes operate within a MB compartment. DAN subtypes receive different inputs and likely modulate different KC-MBON synapses within a compartment (Li, 2020).

A prior analysis showed that PAM01-fb DANs were required to reinforce the absence of expected shock during aversive memory extinction, whereas a different set of γ5 DANs were needed for learning with sugar reinforcement. It is conceivable that another γ5 DAN subtype is required for male flies to learn courtship rejection. The connectome data revealed DAN subtypes in every compartment that is innervated by PAM DANs (Li, 2020).

The γ3 compartment provides an interesting example of subcompartmental targeting of modulation by DANs and of KC input onto MBONs. There are two subtypes of PAM DANs and three types of MBONs in the γ3 compartment. MBON09 and MBON30 primarily receive olfactory information from γm KCs whereas MBON33 primarily receives visual information from γd KCs. PAM12-dd and PAM12-md DANs appear to modulate KC inputs to MBON09/MBON30 or MBON33, respectively. Although existing driver lines do not separate the dd and md subtypes, PAM-γ3 DAN activity is suppressed by sugar and activated by electric shock. This study found that PAM12-dd DANs share input with PPL1 DANs conveying punishment signals, while PAM12-md DANs are co-wired with PAM08 (γ4) DANs conveying reward signals. Thus, synaptic transmission from two sets of modality-specific KCs to different MBONs can be independently modulated by DANs signaling different valences, all within a single compartment (Li, 2020).

To explore the implications of DAN modulation that is specific to sensory modality,the model presented above was expanded by including two DANs, one conveying the visual component of valence and the other the olfactory component. KCs were divided into visual and olfactory modalities, and two configurations for DAN modulation, one, 'shared reward signal', that is compartment-wide and non-specific, and the other, 'separate reward signals' were considered, in which each DAN only induces plasticity onto KC synapses matching its own modality. This latter case models a set of DANs that affect synapses from visual KCs onto the MBON and another set that affect olfactory synapses (alternatively, it could model two MBONs in different compartments that are modulated independently and converge onto a common target). This study found that, when KCs are divided into separate populations and separate modalities can be used to identify valence, learning is more efficient if the pathways are modulated individually (Li, 2020).

This analysis revealed that DANs receive very heterogeneous inputs but, nonetheless, some DANs both within and across compartments often share common input. This combination of heterogeneity and commonality provides many ways of functionally combining different DAN subtypes. For example, it is expected that this combination allows DANs to encode many different combinations of stimuli, actions and events in a state-dependent manner and to transmit this information to specific loci within the MB network (Li, 2020).

In addition to heterogeneous inputs from a variety of brain regions, the DAN network receives a complex arrangement of within and across-compartment monosynaptic input from a variety of MBONs, using both excitatory and inhibitory neurotransmitters. It was found that nearly all MB compartments contain at least one direct within-compartment MBON-DAN feedback connection. MBON feedback onto these DAN subtypes allows previously learned associations that modify MBON activity to affect future learning. MBONs that feedback onto the same DANs that modulate them could, if the result of learning is reduced DAN activity, prevent excess plasticity for an already learned association. In cases where MBON activity excites the DAN, the self-feedback motif could assure that learning does not stop until the MBON has been completely silenced. More generally, MBON inputs to DANs imply that dopaminergic signals themselves reflect learned knowledge and the actions it generates. This could, in turn, allow MBON modulation of DAN activity to support a number of learning paradigms beyond pure classical conditioning, including extinction, second-order conditioning, operant conditioning and reinforcement learning (Li, 2020).

Flies can perform second-order conditioning, in which a stimulus that comes to be associated with reinforcement may itself act as a pseudo-reinforcement when associated with other stimuli. This computational motif of learning the value of sensory states and using the inferred value as a surrogate reinforcement to guide behavioral learning is the core principle behind a class of machine learning techniques known as actor-critic algorithms. These algorithms consist of two modules: the 'actor', whose job is to map sensory inputs to behavioral outputs and the 'critic', whose job is to map sensory inputs to their inferred values and provide these values as a learning signal for the actor. In the mammalian basal ganglia, the dorsal and ventral striatum, the latter of which strongly influences the activity of dopamine neurons in areas including VTA, have been proposed to represent actor and critic modules, respectively. In the MB, this perspective suggests a possible additional 'critic' function for some MBONs beyond their known 'actor' role in directly driving behaviors. Consistent with this view, activation of individual MBONs can excite DANs in other compartments. This study found strong direct monosynaptic connections between some MBONs and DANs in other compartments. Functional studies will be needed to determine which MBONs, if any, participate in an actor-critic arrangement, and which circuit mechanisms-for example, release from inhibition, or reduction of excitation-are at work (Li, 2020).

Another potential role of cross-compartment MBON-DAN feedback is to gate the learning of certain associations so that the learning is contingent on other associations having already been formed. Such a mechanism could support forms of memory consolidation in which long-term memories are only stored after repeated exposure to a stimulus and an associated reward or punishment. Prior studies have linked plasticity in the γ1 and γ2 compartments to short-term aversive memory and plasticity in the α2 and α3 compartments to long-term aversive memory. The cross-compartmental MBON-to-DAN connections observed in this study suggest an underlying cicuit mechanism for this 'transfer' of short to long term memory. Aversion drives PPL101 (γ1pedc) and depresses the conditioned odor-drive to the GABAergic MBON11 (γ1pedc>α/β). MBON11 is strongly connected with PPL1-γ1pedc and is more weakly connected with PPL105 and PPL106. Depression of MBON11 will therefore also release the PPL105 and PPL106 DANs from MBON11-mediated inhibition, increasing their activity in response to the conditioned odor and making them more responsive during subsequent trials. The net result is that short-term aversive learning by MBON11 (γ1pedc>α/β) promotes long-term learning in MBON18 (α2sc) and MBON14 (α3) by releasing the inhibition on the dopamine neurons that innervate the α2 and α3 compartments. Indeed pairing inactivation of MBON11 with odor presentation can form an aversive memory that requires output from PPL1-DANs and optogenetic stimulation of MBON11 during later trials of odor-shock conditioning impairs long-term memory formation (Li, 2020).

Cross-compartment MBON-DAN feedback may also enable context-dependent valence associations, such as the temporary association of positive valence with neutral stimuli when a fly is repeatedly exposed to aversive conditions. Multiple, spaced aversive conditioning trials were recently shown to form, in addition to an aversive memory for the shock-paired odor, a slowly emerging attraction, a 'safety' memory, for a second odor that was presented over the same training period without shock. The GABAergic MBON09 (γ3β'1) appears to play a critical role in the formation of this safety memory. The synapses from KCs conveying the shock-paired odor are depressed in that portion of MBON09's dendrite that arborizes in γ3, while the synapses from KCs conveying the safety odor are depressed in its β'1 arbor. These combined modulations should gradually release downstream neurons from MBON09 feedforward inhibition, consistent with the proposed mechanism for PAM13/14 (β'1) and PAM05/06 (β'2m and β'2p) DANs becoming more responsive to the safe odor. The connectome data revealed that MBON09 is directly connected to the PAM13/14 (β'1ap) and PAM05 (β'2p) DANs, consistent with release from inhibition underlying the delayed encoding of safety (Li, 2020).

DANs also make direct connections with MBONs in each MB compartment, and optogenetic activation of PAM11 DANs can directly excite MBON07 with slow dynamics. This might provide a mechanism to temporarily suppress memory expression without impairing the underlying memory, which is stored as depressed KC-to-MBON synapses. For example, DANs are known to control hunger and thirst-dependent memory expression, and their excitation of MBONs could provide a possible mechanism (Li, 2020).

Feedforward MBON-MBON connections were postulated, based on behavioral and light microscopic anatomical observations, to propagate local plasticity between compartments. The GABAergic MBON11 is poised to play a key role in such propagation as it is connected with 17 other MBONs (at a threshold of 10 synapses). Thus local depression of KC-MBON11 synapses by the shock responsive PPL1-γ1pedc DAN would be expected to result in disinhibition of MBONs in other compartments. Indeed enhanced CS+ responses of MBON01 and MBON03 after odor-shock conditioning have been ascribed to release from MBON11 inhibition. The consequence of releasing other strongly connected MBONs (MBON07, MBON14 and MBON29) from MBON11 inhibition awaits future study. As discussed above, MBON11 is also connected to DANs innervating its cognate compartment and to DANs innervating other compartments. MBON11 may therefore coordinate MBON network activity via both direct and indirect mechanisms. Analogous feedforward inhibitory connections were found from MBON09 (γ3β'1) to MBON01 and MBON03. Aversive learning therefore reduces both MBON11- and MBON09-mediated inhibition of these MBONs, which further skews the MBON network toward directing avoidance of the previously punished odor (Li, 2020).

Disinhibition likely also plays an important role in appetitive memory. The connectivity of MBON06 (β1>α) revealed here indicates that local plasticity in the β1 compartment can propagate to other MBONs. Similar to MBON11, MBON06 is directly connected with nine other MBONs with a threshold of 10 synapses. MBON06 gradually increases its response to repeated odor exposure and odor-evoked responses of β1 DANs vary with metabolic state. More compellingly, artificially triggering PAM10 (β1) DANs can assign appetitive valence to odors. However, the role of MBON06 in appetitive memory and how the PAM10 (β1) DANs modulate KC synapses to MBON06 has not been investigated. The glutamatergic MBON06 (β1>α) makes a large number of reciprocal axoaxonic connections with the GABAergic MBON11, whose activation favors approach. This reciprocal network motif and the positive sign of behavior resulting from β1 DAN-driven memory suggests that MBON06 released glutamate is likely to be inhibitory to MBON11 and its other downstream targets. MBON06 also makes twice as many connections onto the glutamatergic MBON07 and the cholinergic MBON14 as does MBON11. MBON14 (α3) and MBON07 (α1) have established roles in appetitive memory. Assuming that PAM10 (β1) DANs encode appetitive memory by depressing synapses from odor-specific KCs onto MBON06 (β1>α), MBON06 suppression will release the feedforward inhibition of MBON06 onto MBON07 (α1), freeing it to participate in driving the PAM11-aad (α1) DANs. Release of MBON06 inhibition should also simultaneously potentiate the responses of MBON14 (α3) to the conditioned odor. Lastly, releasing the strong inhibition from MBON06 frees MBON11 to provide weaker inhibition that would further favor odor-driven approach (Li, 2020).

Aversive learning will also alter the function of the MBON06:MBON11 network motif. Aversive reinforcement through the PPL101 (γ1pedc) DAN depresses KC-MBON11 connections. This depression releases MBON06 from MBON11-mediated suppression and allows MBON06 to then suppress output through MBON07 and MBON11, further favoring odor-driven avoidance (Li, 2020).

Several studies have described the influence of internal states such as hunger and thirst on the function and physiology of the MB. In essence, states appear to modulate the DAN-MBON network so that the fly preferentially engages in the pursuit of its greatest need. Since current knowledge suggests these deprivation states employ volume release of modulatory peptides or monoamines to control specific DANs and their downstream MBONs, the connectome analyzed in this study does not provide a complete description of this circuit. Nevertheless, direct connectivity does provide some interesting new insight concerning hunger and thirst-dependent control (Li, 2020).

The MBON06 (β1>α):MBON11 (γ1pedc>α/β) cross-inhibitory network motif is likely to be relevant for the dependence of learning and memory expression on hunger state. PPL101 (γ1pedc) DANs and MBON11 (γ1pedc>α/β) are sensitive to nutrient/satiety status, with MBON11 being more responsive in hungry flies. Therefore, the hungry state favors the activity of the MBONs that are normally repressed by MBON06 (Li, 2020).

Thirst-dependent seeking of water vapor requires the activity of DANs innervating the β'2 compartment. The current work shows that these DANs are likely to directly modulate thermo/hygrosensory KCs. In addition, a recent study showed that thirst-dependent expression of water memory required peptidergic suppression of the activity of both the PPL103(γ2α'1) and PAM02 (β'2a) DANs. Interestingly, blocking the PAM02 (β'2a) DANs released memory expression in water-sated flies whereas blocking the PPL103 (γ2α'1) DANs had no effect. However, if the PPL103 (γ2α'1) DANs were blocked together with PAM02 (β'2a) they further facilitated water memory expression. The connectome suggests circuit mechanisms that could reconcile these observations: MBON12 (γ2α'1) provides strong cholinergic input to the PAM02 (β'2a) DANs, suggesting that PPL103 (γ2α'1) DANs might facilitate water memory expression by suppressing MBON12's excitatory input onto PAM02 (β'2a) DANs (Li, 2020).

A role for the MB in guiding locomotion and navigation in ants and other insects has been proposed. The strong and direct connections observed from the majority of MBONs to the CX, the fly's navigation and locomotion control center, provide one circuit path for the MB to exert influence on motor behaviors. Discovering how this input is utilized by the CX will require additional experimental work (Li, 2020).

Optogenetic activation of Drosophila MBONs can promote attraction or avoidance by influencing turning at the border between regions with and without stimulating light. The effect of MBON activation is additive: coactivation of positive-valence MBONs produced stronger attraction, whereas coactivation of positive and negative-valence MBONs cancelled each other out. Because the fly needs to balance the outputs of different compartments, it is expected that those downstream neurons that integrate inputs from multiple MBONs will have a privileged role in motor control (Li, 2020).

The activity of some DANs has been shown to correlate with motor activity, and the optogenetic activation of PAM-β'2 or PPL1-α3 DANs can attract flies, indicating that DAN activity can itself in some circumstances drive motor behavior. The circuit mechanisms generating the correlation between DAN activity and motor behavior remain to be discovered. Downstream targets of MBONs provide extensive input to DANs, and this study found that neurons downstream of multiple MBONs are twice as likely as other MBON targets to provide such direct input to DANs (Li, 2020).

This study also discovered a direct pathway mediated by atypical MBONs that connects to the descending neurons (DNs) that control turning, an observation that provides additional support for the importance of the MB in the control of movement. The connections of these MBONs appear to be structured so as to promote directional movement, often involving a push-pull arrangement of MBONs signaling approach and avoidance. In addition to direct connections to DNs, there is a network of connections mediated by both local LAL interneurons and interneurons that connect the right and left hemisphere LALs. The atypical MBONs that connect directly to the descending steering system, MBON26, MBON27, MBON31 and MBON32, appear to be among the most integrative neurons in the MB system in the sense that they combine direct KC input from the MB compartments with both input from many other MBONs and non-MB input. At the level of the descending neurons, the highly processed signals from these MBONs are combined with inputs from many other sources, including the central complex, to affect a decision to turn. This high degree of integration presumably reflects the complexity and importance of this decision, with many factors involved that might act individually or in combination (Li, 2020).

Visual input to the MB is over-represented in the output to the descending neurons, predominantly through MBON27. Short- and long-term learning based on features in a visual scene has been reported to involve the CX. Plasticity in the CX enables visual feature input from the sky and surrounding scenery to be mapped flexibly onto the fly's internal compass. The visual input conveyed to the MB and, presumably, the learning at the synapses between visual KCs and MBON27, may be of lower resolution, encoding broader features such as color and contrast. An early study demonstrated that the MB is dispensable for flying Drosophila to learn shapes but that it is required for them to generalize their learning if the visual context changes between training and testing. Memory of visual features and the ability to generalize context could allow visual landmarks to help guide navigation either through the CX or by directly influencing descending neurons. The thermo/hygrosensory features conveyed by MBON26 could play a similar role, as could the large amount of odor-related information present in this MBON output pathway (Li, 2020).

The MB has an evolutionarily-conserved circuit architecture and uses evolutionarily-conserved molecular mechanisms of synaptic plasticity. The dense connectome analysed in this report has uncovered many unanticipated circuit motifs and suggested potential circuit mechanisms that now need to be explored experimentally. Drosophila provides access to many of the required tools such as cell-type-specific driver lines, genetically encoded sensors and microscopy methods to observe whole-brain neuronal activity and fine ultrastructure. These features make the fly an excellent system in which to study many general issues in neuroscience, including: the functional diversity of dopaminergic neurons that carry distributed reinforcement signals, the interactions between parallel memory systems, and memory-guided action selection, as well as the mechanisms underlying cell-type-specific plasticity rules, memory consolidation, and the influence of internal state. It is expected that studies of the MB will provide insight into general principles used for cognitive functions across the animal kingdom (Li, 2020).

As mentioned in the introduction, the MB shares many features with the vertebrate cerebellum, and the results should be informative for studies of the cerebellum proper as well as other cerebellum-like structures such as the dorsal cochlear nucleus and the electrosensory lobe of electric fish. A distinctive feature of these systems, and of the MB, is that learning is driven by a particular mechanism; for example DAN modulation in the MB or complex spiking driven by climbing fiber input in the cerebellum. Studies of learning in cortical circuits have traditionally focused on Hebbian forms of learning driven by the ongoing input and output activity of a neuron. However, recent results from both hippocampal circuits have stressed the importance of plasticity that is driven by dendritic plateau potentials or bursts that resemble the distinct learning events seen in cerebellar and MB circuits. Thus, the form of plasticity seen in the MB and its control by output and modulatory circuits may inform studies of learning in the cerebral cortex as well (Li, 2020).

Dopaminergic neurons (DANs) drive learning across the animal kingdom, but the upstream circuits that regulate their activity and thereby learning remain poorly understood. This study provides a synaptic-resolution connectome of the circuitry upstream of all DANs in a learning center, the mushroom body of Drosophila larva. Afferent sensory pathways and a large population of neurons were discovered that provide feedback from mushroom body output neurons and link distinct memory systems (aversive and appetitive). This was combined with functional studies of DANs and their presynaptic partners and with comprehensive circuit modeling. It was found that DANs compare convergent feedback from aversive and appetitive systems, which enables the computation of integrated predictions that may improve future learning. Computational modeling reveals that the discovered feedback motifs increase model flexibility and performance on learning tasks. This study provides the most detailed view to date of biological circuit motifs that support associative learning (Eschbach, 2020).

To behave adaptively in an ever-changing environment, animals must be able to learn new associations between sensory cues (conditioned stimuli, CS) and rewards or punishments (aversive and appetitive unconditioned stimuli, US), and continuously update previous memories, depending on their relevance and reliability (Eschbach, 2020).

Modulatory neurons (for example, DANs) convey information about rewards and punishments, and provide the teaching signals for updating the valence associated with CS in learning circuits across the animal kingdom (for example, the vertebrate basal ganglia or the insect mushroom body, MB). The co-occurrence of CS and modulatory neuron activity tuned only to the received US can support simple associative memory formation. To account for more complex behavioral phenomena, theories have been developed in which learning can be regulated by previously formed associations. According to reinforcement learning theories, learning is driven by errors between predicted and actual US (prediction errors) which are thought to be represented by the activity of DANs. Indeed, in many model organisms, the responses of modulatory neurons have been shown to be adaptive, including monkeys, rodents and insects. Despite recent progress the basic principles by which DAN activity is adaptively regulated and teaching signals are computed are not well understood (Eschbach, 2020).

A prerequisite for the adaptive regulation of modulatory neuron activity is convergence of afferent pathways that convey information about received US with feedback pathways that convey information about previous experiences. A comprehensive synaptic-resolution connectivity map of the feedback circuits that regulate modulatory neurons would provide a basis for understanding how learning is adaptively regulated by prior learning, but it has previously been out of reach (Eschbach, 2020).

Insects, especially in their larval stages, have small and compact brains that have recently become amenable to large-scale electron microscopy (EM) circuit mapping. Both adult and larval insect stages possess a brain center that is essential for associative learning, the MB. The MB contains neurons called Kenyon cells (KCs) that sparsely encode CS, MB modulatory neurons (collectively called MB input neurons, MBINs) that provide the teaching signals and MB output neurons (MBONs) whose activity represents learnt valences of stimuli. In the Drosophila larva, most modulatory neurons are DANs, some are octopaminergic neurons (OANs) and some have unidentified neurotransmitters (simply called MBINs). Modulatory neurons and MBONs project axon terminals and dendrites, respectively, onto the KC axons in a tiled manner, defining MB compartments, in both adult and larval Drosophila. In adult Drosophila, it has been shown that coactivation of KCs and DANs reduces the strength of the KC-MBON synapse in that compartment. Different compartments have been implicated in the formation of distinct types of memories, for example, aversive and appetitive, or short and long term. However, despite a good understanding of the structure and function of the core components of the MB in both adult and larval Drosophila, the circuits presynaptic to modulatory neurons that regulate their activity have remained relatively uncharacterized (Eschbach, 2020).

This study therefore reconstructed all neurons presynaptic to all modulatory neurons in an EM volume that spans the entire nervous system of a first instar Drosophila larva, in which all the core components of the MB were previously reconstructed. This study also determined which individual modulatory neurons are activated by punishments and reconstructed their afferent US pathways from nociceptive and mechanosensory neurons. The neurotransmitter profiles of some of the neurons in the network were characterized and some of the identified structural connections were functionally confirmed. Finally, a model was developed of the circuit constrained by the connectome, the neurotransmitter data and the functional data, and it was used to explore the computational advantages offered by the recently discovered architectural motifs for performing distinct learning tasks (Eschbach, 2020).

Modulatory neurons (for example, DANs) are key components of higher-order circuits for adaptive behavioral control, and they provide teaching signals that drive memory formation and updating. This study provides a synaptic-resolution connectivity map of a recurrent neural network that regulates the activity of modulatory neurons in a higher-order learning center, the Drosophila larval MB. Some of the recently identified structural pathways were functionally tested, and a model of the circuit was developed to explore the roles of these motifs in different learning tasks (Eschbach, 2020).

A large population was discovered of 61 feedback neuron pairs that provide one- and/or two-step feedback from the MBONs to modulatory neurons. Strikingly, it was found that many modulatory neurons receive more than 50% of their total dendritic input from feedback pathways. These results suggest that prior memories as represented by the pattern of MBON activity can strongly influence modulatory neuron activity (Eschbach, 2020).

Learning and memory systems in vertebrates and insects are often organized into distinct compartments implicated in forming distinct types of memories (for example, aversive and appetitive or short and long term). Interestingly, it was found that the majority of the discovered feedback pathways link distinct memory systems, suggesting that the MB functions as an interconnected ensemble during learning. Thus, prior memories formed about an odor in one compartment can influence the formation and updating of distinct types of memories about that odor in other compartments (Eschbach, 2020).

In adult Drosophila, functional connections between some MBONs and DANs have been reported, and some have been shown to play a role in short-term memory formation, long-term memory consolidation, extinction and reconsolidation, or in synchronizing DAN ensemble activity in a context-dependent manner. In some cases, direct MBON-to-DAN connections have been demonstrated. Although direct connections from several MBONs onto DANs exist in the larva, this study found that indirect connections via the feedback neurons account for a much larger fraction of a modulatory neuron's dendritic input than direct MBON synapses. This suggests that adaptive DAN responses may be largely driven by such indirect feedback (Eschbach, 2020).

Some of the one-step within-compartment feedback motifs that were found are analogous to the feedback motifs so far described for the DANs in the vertebrate midbrain. Although the diversity and the inputs of striatal feedback neurons have not yet been fully explored, in the future it will be interesting to determine whether many of the striatal feedback neurons also link distinct memory systems (Eschbach, 2020).

The use of internal predictions can dramatically increase the flexibility of a learning system. This study reveals candidate circuit motifs that could compute integrated predicted value signals across appetitive and aversive memory systems. A prominent motif that was identified is convergence of excitatory and inhibitory connections from MBONs from compartments of opposite valence onto DANs. In naive animals, odor-evoked MBON excitation in all compartments is thought to be similar. However, associative learning selectively depresses conditioned odor drive to MBONs in compartments where modulatory neuron activation has been paired with the odor. It is proposed that by comparing the conditioned odor-evoked MBON excitation in compartments of opposite valence via cross-compartment feedback connections, modulatory neurons compute an integrated predicted value signal across appetitive and aversive domains (Eschbach, 2020).

Convergence of feedback and US pathways could allow the computation of prediction errors An important aspect of reinforcement learning theories is the idea that modulatory neurons compare predicted and actual US (to compute so-called prediction errors) and drive memory formation or extinction depending on the sign of the prediction error. Although Drosophila modulatory neurons have not yet been directly shown to represent prediction errors, adult and larval Drosophila are capable of extinction, and the current study reveals candidate motifs that could support the comparison of expected and actual US. Modulatory neurons were found to receive convergent input from feedback pathways from MBONs and from US pathways. Modulatory neurons could therefore potentially compute prediction errors by comparing inhibitory drive from the feedback pathways with the excitatory drive from the US pathways, or vice versa. Consistent with this idea, some DANs were observed in this model that are inhibited by US alone and activated by CS+ alone, or vice versa (Eschbach, 2020).

This study also reveals that US pathways and feedback pathways converge at two levels: not only at the modulatory neurons themselves, but also at the FB2Ns. Actual and expected outcomes could therefore also be compared by FB2Ns. A recent study in the mouse ventral tegmental area has found that some pre-DAN neurons encoded only actual or only expected reward, whereas others encoded both variables. Thus, both in vertebrates and in insects, comparing predicted and actual outcomes may be a complex computation involving multiple levels of integration that eventually converge onto an ensemble of modulatory neurons (Eschbach, 2020).

An assumption in many reinforcement learning models is that all modulatory neurons receive a global scalar reward prediction error signal. The current study was able to analyze the comprehensive set of inputs of every individual uniquely identifiable modulatory neuron in a learning center. This revealed that each modulatory neuron receives a unique set of feedback inputs that could enable each neuron to compute a unique set of features. Consistent with this, a diversity of adaptive response types in the modulatory neurons was observed in this model This suggests that instead of computing a single global reward prediction error that is distributed to all modulatory neurons, the network uses a range of distinct compartmentalized and distributed teaching signals (Eschbach, 2020).

The connectivity and modeling studies revealed two architectural features of the circuit that provide input to the modulatory neurons that increase its performance and flexibility on learning tasks. The first is the multilevel feedback architecture that includes not only the previously known direct MBON feedback, but also multiple levels of indirect feedback. The second is the extensive set of cross-compartment connections. Modeling suggests that these motifs support improved performance on complex tasks that require the computation of variables such as predictions, prediction errors and context (Eschbach, 2020).

In summary, this study presents a complete circuit diagram of a recurrent network that computes teaching signals in a biological system, providing insights into the architectural motifs that increase its computational power and flexibility. The connectome-constrained model provides numerous predictions that can be tested in the future in a tractable model organism, for which genetic tools can be generated to monitor and manipulate individual neurons. The connectome, together with the functional and modeling studies, therefore provides exciting opportunities for elucidating the biological implementation of reinforcement learning algorithms (Eschbach, 2020).

Many neurons show compartmentalized activity, in which activity does not spread readily across the cell, allowing input and output to occur locally. However, the functional implications of compartmentalized activity for the wider neural circuit are often unclear. This problem was addressed in the Drosophila mushroom body, whose principal neurons, Kenyon cells, receive feedback inhibition from a non-spiking interneuron called APL. This study used local stimulation and volumetric calcium imaging to show that APL inhibits Kenyon cells' dendrites and axons, and that both activity in APL and APL's inhibitory effect on Kenyon cells are spatially localized (the latter somewhat less so), allowing APL to differentially inhibit different mushroom body compartments. Applying these results to the Drosophila hemibrain connectome predicts that individual Kenyon cells inhibit themselves via APL more strongly than they inhibit other individual Kenyon cells. These findings reveal how cellular physiology and detailed network anatomy can combine to influence circuit function (Amin, 2020).

Different types of Drosophila dopaminergic neurons (DANs) reinforce memories of unique valence and provide state-dependent motivational control. Prior studies suggest that the compartment architecture of the mushroom body (MB) is the relevant resolution for distinct DAN functions. This study used a recent electron microscope volume of the fly brain to reconstruct the fine anatomy of individual DANs within three MB compartments. The 20 DANs of the γ5 compartment, at least some of which provide reward teaching signals, can be clustered into 5 anatomical subtypes that innervate different regions within γ5. Reconstructing 821 upstream neurons reveals input selectivity, supporting the functional relevance of DAN sub-classification. Only one PAM-γ5 DAN subtype γ5(fb) receives direct recurrent feedback from γ5β'2a mushroom body output neurons (MBONs) and behavioral experiments distinguish a role for these DANs in memory revaluation from those reinforcing sugar memory. Other DAN subtypes receive major, and potentially reinforcing, inputs from putative gustatory interneurons or lateral horn neurons, which can also relay indirect feedback from MBONs. The single aversively reinforcing PPL1-γ1pedc DAN was similarly reconstructed. The γ1pedc DAN inputs mostly differ from those of γ5 DANs and they cluster onto distinct dendritic branches, presumably separating its established roles in aversive reinforcement and appetitive motivation. Tracing also identified neurons that provide broad input to γ5, β'2a, and γ1pedc DANs, suggesting that distributed DAN populations can be coordinately regulated. These connectomic and behavioral analyses therefore reveal further complexity of dopaminergic reinforcement circuits between and within MB compartments (Otto, 2020).

An adaptive transition from exploring the environment in search of vital resources to exploiting these resources once the search was successful is important to all animals. The neuronal circuitry that allows larval Drosophila melanogaster of either sex to negotiate this exploration-exploitation transition was examined. This was done by combining Pavlovian conditioning with high-resolution behavioral tracking, optogenetic manipulation of individually identified neurons, and EM data-based analyses of synaptic organization. Optogenetic activation of the dopaminergic neuron DAN-i1 was found to both establish memory during training and acutely terminate learned search behavior in a subsequent recall test. Its activation leaves innate behavior unaffected, however. Specifically, DAN-i1 activation can establish associative memories of opposite valence after paired and unpaired training with odor, and its activation during the recall test can terminate the search behavior resulting from either of these memories. These results further suggest that in its behavioral significance DAN-i1 activation resembles, but does not equal, sugar reward. Dendrogram analyses of all the synaptic connections between DAN-i1 and its two main targets, the Kenyon cells and the mushroom body output neuron MBON-i1, further suggest that the DAN-i1 signals during training and during the recall test could be delivered to the Kenyon cells and to MBON-i1, respectively, within previously unrecognized, locally confined branching structures. This would provide an elegant circuit motif to terminate search on its successful completion (Schleyer, 2020).

The patterns of neuronal connectivity underlying multisensory integration, a fundamental property of many brains, remain poorly characterized. The Drosophila melanogaster mushroom body-an associative center-is an ideal system to investigate how different sensory channels converge in higher order brain centers. The neurons connecting the mushroom body to the olfactory system have been described in great detail, but input from other sensory systems remains poorly defined. This study used a range of anatomical and genetic techniques to identify two types of input neurons that connect visual processing centers-the lobula and the posterior lateral protocerebrum-to the dorsal accessory calyx of the mushroom body. Together with previous work that described a pathway conveying visual information from the medulla to the ventral accessory calyx of the mushroom body, this study defines a second, parallel pathway that is anatomically poised to convey information from the visual system to the dorsal accessory calyx (Li, 2020).

Sensory systems use different strategies to detect specific physical features of the outside world. For instance, the olfactory system contains many different types of sensory neuron that are each specialized in detecting a specific class of volatile chemicals. Through only two neuronal layers, olfactory information-the identity of an odor and its concentration-is relayed to higher brain centers. In contrast, the visual system contains far fewer types of sensory neuron, but through numerous neuronal layers, it relays a range of highly processed information-for instance, color, brightness, motion, and shape-to higher brain centers. Thus, higher brain centers have to integrate different types of processed information, bind that information into a coherent representation of the outside world, and use such representations to guide behavior. How higher brain centers achieve this feat remains largely unknown. This gap in knowledge mainly stems from the fact that higher brain centers are formed by a large number of neurons and that the projection neurons conveying information from different sensory systems to these centers often remain poorly characterized. This makes it difficult to understand whether there are specific patterns of neuronal connectivity that enable multisensory integration and what the nature of these patterns are. Deciphering the fundamental neuronal mechanisms that underlie multisensory integration requires a model system such as the Drosophila melanogaster mushroom body, which consists of a relatively small number of neurons whose connections can be charted reliably (Li, 2020).

The Drosophila mushroom body is formed by ∼2,000 neurons-called the Kenyon cells-and has long been studied for its essential role in the formation of olfactory associative memories. The identity of the projection neurons that connect the olfactory system to the mushroom body-and the way Kenyon cells integrate input from these neurons-has been characterized in great detail, highlighting fundamental connectivity patterns that enable this higher brain center to represent olfactory information efficiently. Evidence in Drosophila melanogaster shows that the mushroom body is more than an olfactory center, as it is also required for the formation of visual and gustatory associative memories. However, the identity of the neurons that connect the mushroom body to other sensory systems remains poorly characterized. Thus, a first step toward understanding how the mushroom body integrates multisensory information is to identify such non-olfactory mushroom body input neurons and the genetic tools necessary to manipulate these neurons (Li, 2020).

The mushroom body receives its input through its calyx and sends its output through its lobes. The calyx-a morphologically distinct neuropil containing the synapses formed between projection neurons and Kenyon cells-can be divided into four, non-overlapping regions: one main calyx as well as three accessory calyces named the dorsal, lateral, and ventral accessory calyces. The five output lobes-the α, α', β, β', and γ lobes-contain the synapses formed between Kenyon cells, mushroom body output neurons, and dopaminergic neurons. With respect to these input and output regions, Kenyon cells can be divided into seven distinct types. Of these seven types, five types-the α/βc, α/βs, α'/β'ap, α'/β'm, and γmain Kenyon cells-extend their dendrites only into the main calyx and their axons along one or two lobes. Most of the neurons that project to the main calyx emerge from the antennal lobe, the primary olfactory center in the Drosophila brain. Thus, α/βc, α/βs, α'/β'ap, α'/β'm, and γmain Kenyon cells receive input primarily from the olfactory system (Li, 2020).

In contrast, the two other classes of Kenyon cells do not extend their dendrites into the main calyx. Instead, the α/βp Kenyon cells extend their dendrites into the dorsal accessory calyx-avoiding completely the main, lateral, and ventral accessory calyces-and their axons along the α and β lobes. Likewise, the γd Kenyon cells extend their dendrites exclusively into the ventral accessory calyx and their axons along the γ lobe. Thus, both the α/βp and γd Kenyon cells are anatomically poised to receive non-olfactory input. There is evidence suggesting that the ventral accessory calyx receives input from the medulla, a region of the optic lobe that specializes in processing brightness and color. Furthermore, a recent study suggests that the dorsal accessory calyx is a multisensory center that integrates input from multiple sensory pathways, including the olfactory, gustatory, and visual systems (Li, 2020).

This study reports a strategy that uses a combination of genetic tools-including transgenic lines that drive expression in few neurons and a photo-labeling technique used to identify individual neurons and their pre-synaptic partners-to characterize the input neurons of the α/βp Kenyon cells. Two types of mushroom body input neuron were identified in that, together, form about half of the total input the α/βp Kenyon cells receive in the dorsal accessory calyx. The first neuronal type-henceforth referred to as LOPNs-consists of a neuron that projects from the lobula, a region of the optic lobe specialized in detecting visual features, such as shape and motion. The second type of neuron-henceforth referred to as PLPPNs-consists of projection neurons that emerge from the posterior lateral protocerebrum, a brain region that receives input from the optic lobe. Interestingly, LOPN and PLPPNs do not project to the ventral accessory calyx and do not connect to the γd Kenyon cells. Based on these findings, it is concluded that there are two parallel pathways that convey visual information to the mushroom body: a pathway projecting from the medulla to the γd Kenyon cells and another pathway projecting from the lobula and posterior lateral protocerebrum to the α/βp Kenyon cells (Li, 2020).

This study has identified and characterized neurons projecting to the dorsal accessory calyx of the mushroom body and show that these neurons are pre-synaptic to the α/β_p Kenyon cells. Using a combination of genetic and anatomical techniques, it was possible to distinguish two different types of projection neuron: _LOPN projecting from the lobula-an area of the optic lobe processing visual features, such as shape and motion-and the PLPPNs projecting from the posterior lateral protocerebrum. Although the posterior lateral protocerebrum remains poorly characterized in D. melanogaster, evidence from other insects shows that this brain region receives input from the optic lobe. Interestingly, it was found that the dendrites formed by the _PLPPNs in the posterior lateral protocerebrum are in close proximity to neurons that project from the ventral medulla. Based on these results-and considering insights from the connectome-it is estimated that LOPNs and PLPPNs account for half of total input that α/βp Kenyon cells receive in the dorsal accessory calyx. LOPNs and PLPPNs do not extend axonal terminals into the ventral accessory calyx, the other calyx known to receive visual input, but rather extend axonal terminals into the dorsal accessory calyx and into the superior lateral protocerebrum. Likewise, the α/βp Kenyon cells do not connect to the visual projection neurons that are associated with the ventral accessory calyx. These findings suggest that the visual system is connected to the mushroom body via two parallel pathways: the α/βp Kenyon cells receive input from the lobula and the posterior lateral protocerebrum, whereas the γd Kenyon cells receive input directly from the medulla. Further functional studies are necessary to determine what kind of visual information is processed by the α/βp Kenyon cells (Li, 2020).

In Drosophila melanogaster, the mushroom body has long been studied as an olfactory processing center. However, evidence from many insects, including the honeybee Apis mellifera, shows that the mushroom body integrates sensory information across different modalities. In honeybees, the input region of the mushroom body, also called the calyx, is divided into different layers, and each layer receives input from either the olfactory or visual system. Because the dendrites of Kenyon cells are also restricted to specific layers, it has been suggested that, in the honeybee, multisensory integration does not occur at the level of individual Kenyon cells but rather at the population level. Although the honeybee mushroom body differs greatly from the Drosophila mushroom body-it contains about a hundred times as many Kenyon cells and its input region is divided in multiple complex layers-it appears that both mushroom bodies share a common fundamental connectivity principle: the segregation of input based on sensory modality. This connectivity mechanism is immediately apparent in the structural organization of the Drosophila melanogaster mushroom body: the Kenyon cells receiving input from the olfactory system all extend their dendrites into the main calyx, whereas the Kenyon cells receiving input from the visual system extend their dendrites either in the dorsal accessory calyx or the ventral accessory calyx. Many studies have demonstrated that the Kenyon cells that process olfactory information-those associated with the main calyx-integrate input broadly across the different types of olfactory projection neuron. Interestingly, it appears that the Kenyon cells that process visual information are wired differently (Li, 2020).

A thorough understanding is available of how olfactory Kenyon cells integrate input from the antennal lobe: most Kenyon cells receive, on average, input from seven projection neurons, and the projection neurons connecting to the same Kenyon cell share no apparent common features. Theoretical studies have shown that this random-like connectivity pattern enables the mushroom body to form sparse and decorrelated odor representations and thus maximizes learning. Randomization of sensory input is a connectivity pattern that is well suited for representing olfactory information-as an odor is encoded based on the ensemble of olfactory receptors it activates-and might not be suitable for representing visual information. Indeed, the results of this study suggest that specific visual features-the signals processed by the medulla and the ones processed by the lobula and the posterior lateral protocerebrum-need to be represented by two separate subpopulations of Kenyon cells. This observation mirrors anatomical studies of the honeybee brain: the neurons projecting from the lobula terminate in a different layer than the neurons projecting from the medulla. This arrangement might be essential to preserve distinct visual features when forming associative memories. Functional and behavioral studies are required to determine whether indeed the mushroom body represents multisensory stimuli in this manner (Li, 2020).

Go to top

Many neurons show compartmentalized activity, in which activity does not spread readily across the cell, allowing input and output to occur locally. However, the functional implications of compartmentalized activity for the wider neural circuit are often unclear. This problem was addressed in the Drosophila mushroom body, whose principal neurons, Kenyon cells, receive feedback inhibition from a non-spiking interneuron called APL. This study used local stimulation and volumetric calcium imaging to show that APL inhibits Kenyon cells' dendrites and axons, and that both activity in APL and APL's inhibitory effect on Kenyon cells are spatially localized (the latter somewhat less so), allowing APL to differentially inhibit different mushroom body compartments. Applying these results to the Drosophila hemibrain connectome predicts that individual Kenyon cells inhibit themselves via APL more strongly than they inhibit other individual Kenyon cells. These findings reveal how cellular physiology and detailed network anatomy can combine to influence circuit function (Amin, 2020).

This study has shown that activity in APL is spatially restricted intracellularly for both sensory-evoked and local artificial stimulation. This local activity in APL translates into a spatially non-uniform inhibitory effect on Kenyon cells that is strongest locally and becomes weaker farther from the site of APL stimulation. Finally, combining physiological and anatomical data -- APL's estimated space constant and the spatial arrangement of KC-APL and APL-KC synapses -- predicts that each Kenyon cell disproportionately inhibits itself more than other individual Kenyon cells (Amin, 2020).

This study showed that APL can locally inhibit Kenyon cells in multiple locations in the mushroom body. Remarkably, inhibition of Ca2+ influx in Kenyon cell axons could be stronger closer to the axon initial segment (in the posterior peduncle than farther from it, for example, stronger at the lower vertical lobe than the tip of the vertical lobe, both when stimulating APL with ATP or when directly applying GABA. At first glance, this observation is puzzling given the direction of action potential travel. GABA should act on Kenyon cells by suppressing depolarization through shunting inhibition in both dendrites and axons, as the GABAA receptor Rdl is expressed in both. Therefore, the observation of stronger proximal than distal inhibition appears to suggest that as action potentials travel from the axon initial segment toward the distal tip of the axon, they can enter a zone of shunting inhibition and be suppressed, yet recover on the other side. Such a scenario could occur if depolarization is suppressed enough to reduce voltage-gated Ca2+ influx in a local zone, yet remains sufficient on the other side of the inhibitory zone to trigger enough voltage-gated Na+ channels to regenerate the action potential. However, this interpretation seems unlikely. More likely, local inhibition might particularly suppress Ca2+ influx rather than depolarization, perhaps by acting through GABAB receptors; Kenyon cells express GABABR1 and GABABR2 and APL inhibits KCs partly via GABAB receptors, although their subcellular localization is unknown. Given that synaptic vesicle release requires Ca2+ influx, such inhibition would still locally suppress Kenyon cell synaptic output (Amin, 2020).

Local inhibition of Kenyon cell output predicts that activity of MBONs near the site of APL activation would be more strongly inhibited than MBONs far away. This prediction may be tested in future experiments, for example locally stimulating APL in the tip of the vertical lobe and comparing the inhibitory effect on MBONs in nearby compartments like α3 and α'3 vs. on MBONs in distant compartments like γ5 and β2 (Amin, 2020).

While APL inhibits Kenyon cells locally, the inhibitory effect spreads somewhat more widely than APL's own activity, though weakly. Local GABA application produces similar results to locally activating APL with ATP, suggesting that GCaMP6f signals in APL accurately predict GABA release. How can APL inhibit Kenyon cells where it itself is not active? Wider inhibition when inhibiting Kenyon cell dendrites can be easily explained as blocking action potentials, but the wider inhibition when inhibiting the axons is more puzzling. It is speculated that this result might arise from wider network activity. For example, Kenyon cells form recurrent connections with DPM and dopaminergic neurons and form synapses and gap junctions on each other. Through such connections, an odor-activated Kenyon cell might excite a neighboring Kenyon cell's axon; the neighbor might passively spread activity both forward and backward or, not having fired and thus not being in a refractory period, it might even be excited enough to propagate an action potential back to the calyx. Alternatively, Kenyon cells indirectly excite antennal lobe neurons in a positive feedback loop. In these scenarios, locally puffing GABA or activating APL artificially in the vertical lobe tip would block the wider network activity, thus reducing Ca2+ influx in the calyx. Although the calyx was not activated when stimulating Kenyon cells in the vertical lobe tip, this might be because simultaneous activation of all Kenyon cells (but not odor-evoked activation of ~10% of Kenyon cells) drives strong-enough local APL activation to block wider network activity. Future experiments may test these possibilities (Amin, 2020).

By combining physiological measurements of localized activity with the detailed anatomy of the hemibrain connectome, a model was built that predicts that the average Kenyon cell inhibits itself more than it inhibits other individual Kenyon cells. This prediction is supported by previous experimental results that some Kenyon cells can inhibit some Kenyon cells more than others, and that an individual Kenyon cell can inhibit itself. The model goes beyond these results in predicting that the average Kenyon cell actually preferentially inhibits itself, and by explaining how differential inhibition can arise from local activity in APL and the spatial arrangement of KC-APL and APL-KC synapses (Amin, 2020).

Note that these results do not contradict previous findings that Kenyon cell lateral inhibition is all-to-all. The finding of all-to-all inhibition was based on findings that whereas blocking synaptic output from all Kenyon cells vastly increases Kenyon cell odor responses (by blocking negative feedback via APL), blocking output from one subset of Kenyon cells has no effect on (for α'β' and γ Kenyon cells), or only weakly increases (for αβ Kenyon cells), odor responses of that subset. This finding is consistent with findings that predict only preferential, not exclusive, self-inhibition. For example, while blocking output from only γ Kenyon cells would remove γ-to-γ inhibition, γ Kenyon cells (only 1/3 of all Kenyon cells) should still get enough lateral inhibition from the other 2/3 of Kenyon cells to suppress their activity to normal levels (Amin, 2020).

The findings show that localized activity within APL has two broader implications for mushroom body function. First, local activity in APL leads to stronger inhibition of Kenyon cells nearby than far away, to the point that different compartments of the mushroom body lobes defined by dopaminergic and output neuron innervation can receive different inhibition. This local inhibition would allow the single APL neuron to function effectively as multiple inhibitory interneurons, much like mammalian and fly amacrine cells. Each 'sub-neuron' of APL could locally modulate the function of one mushroom body 'compartment', that is, a unit of dopaminergic neurons/Kenyon cells / mushroom body output neurons. Different compartments are innervated by different dopaminergic neurons (e.g., reward vs. punishment) and different mushroom body output neurons (e.g., avoid vs. approach), and they govern synaptic plasticity and hence learning by different rules (e.g., different speeds of learning/forgetting). Thus, APL could locally modulate different compartment-specific aspects of olfactory learning, especially given that different regions of APL respond differently to dopamine and electric shock punishment. If such local inhibition is important for learning, the fact that spatial attenuation of APL's inhibitory effect is gradual and incomplete could explain why mushroom body compartments are arranged in their particular order, with reward and punishment dopaminergic neurons segregated into the horizontal and vertical lobes, respectively. Under this scenario, APL would serve two distinct, spatially segregated functions: enforcing Kenyon cell sparse coding in the calyx, and modulating learning in the compartments of the lobes (Amin, 2020).

Second, the finding of disproportionate self-inhibition compared to other-inhibition provides a new perspective on APL's function. Inhibition from APL sparsens and decorrelates Kenyon cell odor responses to enhance learned odor discrimination but in general, decorrelation is better served by all-to-all lateral inhibition than by self-inhibition. Self-inhibition does help decorrelate population activity by pushing some neurons' activity below spiking threshold; this could occur if APL can be activated by subthreshold activity in Kenyon cells, for example, from KC-APL synapses in Kenyon cell dendrites. However, this effect of self-inhibition is better thought of, not as decorrelation per se, but rather as gain control, effectively equivalent to adjusting the threshold or gain of excitation according to the strength of stimulus. Of course, lateral inhibition is still a strong force in the mushroom body: given that there are ~2000 Kenyon cells, the sum total lateral inhibition that an individual Kenyon cell receives would still be stronger than its own self-inhibition, even with the predicted imbalance. Why might the predicted 'bonus' self-inhibition be useful? Beyond its role in sparse coding, APL inhibition is also thought to function as a gating mechanism to suppress olfactory learning; for such a function it would make sense for Kenyon cells to disproportionately inhibit themselves. Future work will address how lateral inhibition interacts with other functions for APL in this local feedback circuit (Amin, 2020).

It has been evidenced that vision-based decision-making in Drosophila consists of both simple perceptual (linear) decision and value-based (non-linear) decision. This paper proposes a general computational spiking neural network (SNN) model to explore how different brain areas are connected contributing to Drosophila linear and nonlinear decision-making behavior. First, the SNN model could successfully describe all the experimental findings in fly visual reinforcement learning and action selection among multiple conflicting choices as well. Second, the computational modeling shows that dopaminergic neuron-GABAergic neuron-mushroom body (DA-GABA-MB) works in a recurrent loop providing a key circuit for gain and gating mechanism of nonlinear decision making. Compared with existing models, this model shows more biologically plausible on the network design and working mechanism, and could amplify the small differences between two conflicting cues more clearly. Finally, based on the proposed model, the unmanned aerial vehicle (UAV) could quickly learn to make clear-cut decisions among multiple visual choices and flexible reversal learning resembling to real fly. Compared with linear and uniform decision-making methods, the DA-GABA-MB mechanism helps UAV complete the decision-making task with fewer steps (Zhao, 2020).

In pursuit of food, hungry animals mobilize significant energy resources and overcome exhaustion and fear. How need and motivation control the decision to continue or change behavior is not understood. Using a single fly treadmill, this study shows that hungry flies persistently track a food odor and increase their effort over repeated trials in the absence of reward suggesting that need dominates negative experience. It was further shown that odor tracking is regulated by two mushroom body output neurons (MBONs) connecting the MB to the lateral horn. These MBONs, together with dopaminergic neurons and Dop1R2 signaling, control behavioral persistence. Conversely, an octopaminergic neuron, VPM4, which directly innervates one of the MBONs, acts as a brake on odor tracking by connecting feeding and olfaction. Together, these data suggest a function for the MB in internal state-dependent expression of behavior that can be suppressed by external inputs conveying a competing behavioral drive (Sayin, 2019).

Flexibility is an important factor in an ever in-flux environment, where scarcity and competition are the norm. Without persistence to achieve its goals, however, an animal's strive to secure food, protect its offspring, or maintain its social status is in jeopardy. Therefore, sensory cues related to food or danger often elicit strong impulses. However, these impulses must be strictly controlled to allow for coherent goal-directed behavior and to permit behavioral transitions when sensible. Inhibition of antagonistic behavioral drives at the cognitive and physiological level has been proposed as a major task of a nervous system. Which sensory cues and ultimately which behaviors are prioritized and win depends on the animal's metabolic state, internal motivation, and current behavioral context. How this is implemented at the level of individual neurons, circuit motifs, and mechanisms remains an important open question (Sayin, 2019).

Like most animals, energy-deprived flies prioritize food seeking and feeding behavior. To find food, flies can follow olfactory or visual cues over long distances. External gustatory cues provide information about the type and quality of the eventually encountered food. However, only internal nutrient levels will provide reliable feedback about the quality and quantity of a food source and ultimately suppress food-seeking behaviors. Therefore, food odor, the taste of food, and post-ingestive internal feedback signals induce sequential and partly antagonistic behaviors. Interestingly, chemosensory and internal feedback systems typically mediated by distinct neuromodulators appear to converge in the mushroom body (MB). How neurons and neural circuits signal and combine external and internal cues to maintain or suppress competing behavioral drives is not well understood (Sayin, 2019).

In mammals, norepinephrine (NE) released by a brain stem nucleus, the locus coeruleus, has been implicated in controlling the balance between persistence and action selection. The potential functional counterpart of NE in insects could be octopamine (OA). Flies lacking OA indeed show reduced arousal, for instance upon starvation. Additionally, OA neurons (OANs) gate appetitive memory formation of odors and also modulate taste neurons and feeding behavior. OANs are organized in distinct clusters and project axons to diverse higher brain regions in a cell type-specific manner. The precise roles and important types of OA and NE neurons in state-dependent action selection remain to be elucidated (Sayin, 2019).

Similar to NE and OA, dopamine (DA) is being studied in many aspects of behavioral adaptation and flexibility. Different classes of DA neurons (DANs) innervating primarily the MB signal negative or positive context, or even wrong predictions (Sayin, 2019 and references therein).

This study took advantage of the small number and discrete organization of neuromodulatory neurons in the fly brain to analyze the mechanistic relationship between motivation-dependent persistence in one behavior and the decision to disengage and change to another behavior. Using a single fly spherical treadmill assay, this study found that hungry flies increase their effort to track a food odor with every unrewarded trial. MB output through two identified MBONs (MBON-γ1pedc>α/β and MBON-α2sc) is required for persistent odor tracking. MBON-α2sc provides a MB connection to the lateral horn (LH), where it can modify innate food odor attraction. Furthermore, this study pinpoints a specific type of OAN, VPM4 (ventral paired medial), which connects feeding centers directly to MBON-γ1pedc>α/β and disrupts food odor tracking. Finally, the experimental data suggest that persistent tracking depends on DANs, including PPL1-γ1pedc, and signaling through dopamine receptor Dop1R2 in αβ-type KCs. Based on these results, it is proposed that MB output and a direct external input, depending on internal state and motivation, gradually promote or interrupt ongoing behavior (Sayin, 2019).

What drives gradually increasing persistence in behavior? For the fly, a model is proposed by which a circuit module of KCs, MBONs, and DANs drive gradually increasing odor tracking, which can be efficiently suppressed by extrinsic MBON-innervating feeding-related OANs. Behavioral persistence has been previously analyzed in flies in a different context. For instance, courtship of fly males and copulation with a female are maintained by dopaminergic neurons in the ventral nerve cord, where they counteract GABAergic neurons. In that scenario, DANs in the ventral nerve cord maintain an ongoing behavior and prevent male premature disengagement before successful insemination (Sayin, 2019).

The experimental data also implicate DANs, primarily from within the PPL1 (e.g., PPL1-γ1pedc) and PPL2ab clusters, and Dop1R2 signaling. In particular, inactivation of synaptic output of DANs positive for TH-Gal4 as well as loss of Dop1R2 in αβ-type KCs reduced the increase in odor tracking from trial to trial, while not affecting the speed at first odor stimulation. These data suggest that TH+ DANs promote goal-directed movement, i.e., odor tracking, through a Dop1R2-dependent mechanism in KCs (Sayin, 2019).

MBON-γ1pedc>αβ, which receives dopaminergic input by PPL1-γ1pedc, is required for odor tracking. Moreover, this study also observed a trial-to-trial decrease in odor response of this MBON, matching the dopamine-induced synaptic depression previously observed in MBONs upon learning. Notably, PPL1-γ1pedc activates Dop1R2 in MBON-γ1pedc>αβ, a signal recently found to be critical for appetitive long-term memory. Nevertheless, it appears that, in addition to PPL1-γ1pedc, other DANs regulate behavioral persistence by modulating in particular αβ-KCs. It is intriguing to speculate about a common function of Dop1R2 in the formation of long-lasting aversive memory induced by repeatedly pairing odor with an aversive experience and the behavior examined in this study: increased and persistent expression of a behavior induced by the experience of repeated failure to reach a goal (Sayin, 2019).

The experimental data further implicated MBON-α2sc, which is connected to MBON-γ1pedc>αβ. Calcium imaging data are consistent with an inhibitory interaction between the two MBONs. However, some of the behavioral data and prior imaging data do not support an inhibitory connection. Furthermore, MBON-γ1pedc>αβ projects to other brain regions and downstream targets, and similarly MBON-α2sc receives additional inputs—all of which could be equally or more important for persistent behavior than a direct connection between these two MBONs. Finally, some DANs respond to movement, including PPL1-γ2α'1/MV1. Although no essential role of this particular neuron was found in odor tracking persistence, movement might contribute to the activity of MBONs responding the odorant (Sayin, 2019).

Remarkably, MBON-α2sc connects the MB to neurons within the LH. Thus, it is speculated that the LH might assign an odor to its corresponding behavioral category, such as 'food-related' for vinegar, while the MB acts as a top-down control to gauge the expression of an innate behavior (i.e., tracking an appetitive odor) according to state and experience (Sayin, 2019).

The behavioral data led to the proposal of a circuit model. Using computational modeling, this study tested whether the MB network including DANs and MBONs could, in theory, produce the observed behavior. Indeed, it was found that a simplified recurrent circuit of KCs, DANs, and MBONs can account for the observed behavioral persistence and also the measured MBON-γ1pedc>αβ odor responses. While this model cannot replace experimental evidence, it forms a useful theoretical framework for future studies on the role of the MB in behavioral persistence (Sayin, 2019).

Based on the present data and computational predictions, a model is proposed by which the recurrent circuit architecture of the MB, in addition to storing information for future behavior, is ideally suited to maintain and gradually change ongoing behavior, for instance by modulating output of the LH, according to the animal's internal state and needs (Sayin, 2019).

The use of an olfactory treadmill has allowed dissection of the different aspects of a food search. In particular, how does food and feeding suppress food search if the sensory cue, the odor, is still present? OA-VPM4 connects feeding centers (i.e., SEZ) directly with odor tracking-promoting MBON-γ1pedc>αβ and inhibits its activity suggesting an inhibitory connection between VPM4 and the MBON. Nevertheless, it cannot be excluded that OA-VPM4 signals through multiple mechanisms including OA and possibly other neurotransmitters. In addition, a recent study showed that activation of VPM4 promotes proboscis extension to sugar. Although a direct role in taste detection through pharynx or labellum appears unlikely, it is possible that feeding behavior itself (e.g., lymphatic sugar, food texture, activity of feeding muscles) are detected and/or promoted by these neurons and then brought to the MB. It is proposef that VPM4 is a direct mediator between olfactory-guided food search and the rewarding experience of feeding and related behavior (Sayin, 2019).

The data provide a neural circuit mechanism empowering flies to express and prioritize behavior in a need- and state-dependent manner. It is exciting to speculate that fundamentally similar circuit motifs might exist in NE and DA neuron-containing circuits in the mammalian brain, governing the organization of behavior in a flexible and context-dependent manner by integrating internal and external context. For instance, noradrenergic neurons of the brainstem nucleus of the solitary tract (NST) receive taste information, and input from the gastrointestinal tracts, lungs, and heart. Neurons in the NST project to multiple brain regions including the amygdala, hypothalamus, and insular cortex, all of which receive internal state as well as other sensory information (Sayin, 2019).

The data in the fly provide an experimental and theoretical framework for a better understanding of the fundamental circuit mechanisms underpinning neuromodulation of context-dependent behavioral persistence and withdrawal (Sayin, 2019).

Animals employ diverse learning rules and synaptic plasticity dynamics to record temporal and statistical information about the world. However, the molecular mechanisms underlying this diversity are poorly understood. The anatomically defined compartments of the insect mushroom body function as parallel units of associative learning, with different learning rates, memory decay dynamics and flexibility. This study shows that nitric oxide (NO) acts as a neurotransmitter in a subset of dopaminergic neurons in Drosophila. NO's effects develop more slowly than those of dopamine and depend on soluble guanylate cyclase in postsynaptic Kenyon cells. NO acts antagonistically to dopamine; it shortens memory retention and facilitates the rapid updating of memories. The interplay of NO and dopamine enables memories stored in local domains along Kenyon cell axons to be specialized for predicting the value of odors based only on recent events. These results provide key mechanistic insights into how diverse memory dynamics are established in parallel memory systems (Aso, 2019).

An animal's survival in a dynamically changing world depends on storing distinct sensory information about their environment as well as the temporal and probabilistic relationship between those cues and punishment or reward. Thus it is not surprising that multiple distributed neuronal circuits in the mammalian brain have been shown to process and store distinct facets of information acquired during learning. Even a simple form of associative learning such as fear conditioning induces enduring changes, referred to as memory engrams, in circuits distributed across different brain areas. Do these multiple engrams serve different mnemonic functions, what molecular and circuit mechanisms underlie these differences, and how are they integrated to control behavior? Localizing these distributed engrams, understanding what information is stored in each individual memory unit and how units interact to function as one network are important but highly challenging problems. The Drosophila mushroom body (MB) provides a well-characterized and experimentally tractable system to study parallel memory circuits. Olfactory memory formation and retrieval in insects requires the MB. In associative olfactory learning, exposure to an odor paired with a reward or punishment results in formation of a positive- or negative-valence memory, respectively. In the MB, sensory stimuli are represented by the sparse activity of ~2,000 Kenyon cells (KCs). Each of 20 types of dopaminergic neurons (DANs) innervates compartmental regions along the parallel axonal fibers of the KCs. Similarly, types of mushroom body output neurons (MBONs) arborize their dendrites in specific axonal segments of the KCs; together, the arbors of the DANs and MBONs define the compartmental units of the MB. Activation of individual MBONs can cause behavioral attraction or repulsion, depending on the compartment in which their dendrites arborize, and MBONs appear to use a population code to govern behavior (Aso, 2019).

A large body of evidence indicates that these anatomically defined compartments of the MB are also the units of associative learning. Despite the long history of behavioral genetics in fly learning and memory, many aspects of the signaling pathways governing plasticity -- especially whether they differ between compartments -- remain poorly understood. Nevertheless, dopaminergic neurons and signaling play a key role in all MB compartments, and flies can be trained to form associative memories by pairing the presentation of an odor with stimulation of a single dopaminergic neuron. Punishment or reward activates distinct sets of DANs that innervate specific compartments of the MB. Activation of the DAN innervating a MB compartment induces enduring depression of KC-MBONs synapses in those specific KCs that were active in that compartment at the time of dopamine release. Thus, which compartment receives dopamine during training appears to determine the valence of the memory, while which KCs were active during training determines the sensory specificity of the memory (Aso, 2019).

Compartments operate with distinct learning rules. Selective activation of DANs innervating specific compartments has revealed that they can differ extensively in their rates of memory formation, decay dynamics, storage capacity, and flexibility to learn new associations. For instance, the dopaminergic neuron PAM-α1 can induce a 24h memory with a single 1-minute training session, whereas PPL1-α3 requires ten repetitions of the same training to induce a 24h memory. PPL1-γ1pedc (aka MB-MP1) can induce a robust short-lasting memory with a single 10-second training, but cannot induce long-term memories even after 10 repetitions of a 1-minute training. PAM-α1 can write a new memory without compromising an existing memory, whereas PPL1-γ1pedc extinguishes the existing memory when writing a new memory. What molecularand cellular differences are responsible for the functional diversity of these compartments? Some differences might arise from differences among KC cell types, but memory dynamics are different even between compartments that lie along the axon bundles of the same Kenyon cells (for example, α1 and α3). This paper shows that differences in memory dynamics between MB compartments can arise from the deployment of distinct cotransmitters by the DAN cell types that innervate them (Aso, 2019).

Evidence from a wide range of organisms establishes that dopaminergic neurons often release a second neurotransmitter, but the role of such cotransmitters in diversifying neuronal signaling is much less clear. In rodents, subsets of dopaminergic neurons co-release glutamate or GABA. In mice and Drosophila, single-cell expression profiling reveals expression of diverse neuropeptides in dopaminergic neurons. EM connectome studies of the mushroom body in adult and larval Drosophila reveal the co-existence of small-clear-core and large- dense-core synaptic vesicles in individual terminals of dopaminergic neurons; moreover, the size of the observed large-dense-core 02 vesicles differs between DAN cell types (Aso, 2019).

This study found that NOS, the enzyme that synthesizes NO, was located in the terminals of a subset of DAN cell types. NOS catalyzes the production of nitric oxide (NO) from L-arginine. Drosophila NOS is regulated by Ca2+/calmodulin, raising the possibility that NO synthesis might be activity dependent. Furthermore, the localization of the NOS1 protein in the axonal terminals of DANs is consistent with NO serving as a cotransmitter. The conclusion that NO acts as a neurotransmitter is supported by the observation that NO signaling requires the presence of a putative receptor, soluble guanylate cyclase, in the postsynaptic Kenyon cells. This role contrasts with the proposed cell-autonomous action of NOS in the ellipsoid body, in which NO appears to target proteins within the NOS-expressing ring neurons themselves, rather than conveying a signal to neighboring cells. The valence-inversion phenotype observed when PPL1-γ1pedc was optogenetically activated in a dopamine-deficient background can be most easily explained if NO induces synaptic potentiation between odor-activated KCs and their target MBONs. Modeling work is consistent with this idea, but testing this idea and 19 other possible mechanisms for NO action will require physiological experiments (Aso, 2019).

During olfactory learning, the concentration of Ca2+ in KC axons represents olfactory information. The coincidence of a Ca2+ rise in spiking KCs and activation of the G-protein-coupled Dop1R1 dopamine receptor increases adenylyl cyclase activity. The resultant cAMP in turn activates protein kinase A, a signaling cascade that is important for synaptic plasticity and memory formation throughout the animal phyla. In contrast, when DANs are activated without KC activity, and thus during low intracellular Ca2+ in the KCs, molecular pathways involving the Dop1R2 receptor, Rac1 and Scribble facilitate decay of memory (Aso, 2019).

This study found that NOS in PPL1-γ1pedc shortens memory retention, while facilitating fast updating of memories in response to new experiences. These observations could be interpreted as indicating that NO regulates forgetting. Indeed, NO-dependent effect requires scribble in KCs, a gene previously reported as a component of active forgetting. However, it is an open question whether the signaling pathways for forgetting, which presumably induce recovery from synaptic depression, are related to signaling cascades downstream of NO and guanylate cyclase, which appear to be able to induce memory without prior induction ofsynaptic depression by dopamine. Lack of detectable 1-day memory formation after spaced training with PPL1-γ1pedc can be viewed as a balance between two distinct, parallel biochemical signals, one induced by dopamine and the other by NO, rather than the loss of information (that is, forgetting). Confirming this interpretation will require better understanding of the signaling pathways downstream of dopamine and NO. The search for such pathways will be informed by the prediction from modeling that dopamine and NO may alter two independent parameters that define synaptic weights with a multiplicative interaction (Aso, 2019).

In the vertebrate cerebellum, which has many architectural similarities to the MB, long-term-depression at parallel fiber-Purkinje cell synapses (equivalent to KC-MBON synapses) induced by climbing fibers (equivalent to DANs) can coexist with long-term-potentiation by NO. In this case, the unaltered net synaptic weight results from a balance between coexisting LTD and LTP rather than recovery from LTD. This balance was suggested to play an important role in preventing memory saturation in the cerebellum 59 and allowing reversal of motor learning. In the Drosophila MB, a similar facilitation was observed of reversal learning by NO. The antagonistic roles of NO and synaptic depression may be a yet another common feature of the mushroom body and the cerebellum (Aso, 2019).

Opposing cotransmitters have been observed widely in both invertebrate and vertebrate neurons. A common feature in these cases is that the transmitters have distinct time courses of action. For instance, hypothalamic hypocretin-dynorphin neurons that are critical for sleep and arousal synthesize excitatory hypocretin and inhibitory dynorphin. When they are released together repeatedly, the distinct kinetics of their receptors result in an initial outward current, then little current, and then an inward current in the postsynaptic cells. In line with these observations, this study found that dopamine and NO show distinct temporal dynamics: NO-dependent memory requires repetitive training and takes longer to develop than dopamine-dependent memory. What molecular mechanisms underlie these differences? Activation of NOS may require stronger or more prolonged DAN activation than does dopamine release. Alternatively, efficient induction of the signaling cascade in the postsynaptic KCs might require repetitive waves of NO input. Direct measurements of release of dopamine and NO, and downstream signaling events by novel sensors will be needed to address these open questions (Aso, 2019).

Decades of behavioral genetic studies have identified more than one hundred genes underlying olfactory conditioning in Drosophila. Mutant andtargeted rescue studies have been used to map the function of many memory-related genes encoding synaptic or intracellular signaling proteins (for example, rutabaga, DopR1/dumb, DopR2/DAMB, PKA-C1/DC0, Synapsin, Bruchpilot, Orb2 and Rac1) to specific subsets of Kenyon cells. However, it is largely unknown if these proteins physically colocalize at the same KC synapses to form intracellular signaling cascades. Some of 97 these proteins might preferentially localize to specific MB compartments. Alternatively, they may distribute uniformly along the axon of Kenyon cells, but be activated in only specific compartments. Identification of cell-type-specific cotransmitters in DANs enabled a beginning to the exploration of this question (Aso, 2019).

Optogenetic activation of specific DANs was used to induce memory in specific MB compartments, while manipulating genes in specific types of KCs. This approach allowed mapping and characterization of the function of memory-related genes at a subcellular level. For example, the Gycbeta100B gene, which encodes a subunit of sGC, has been identified as 'memory suppressor gene' that enhances memory retention when pan-neuronally knocked down, but the site of its action was unknown. Gycbeta100B appears to be broadly dispersed throughout KC axons, based on the observed distribution of a Gycbeta100B-EGFP fusion protein. The experiments ectopically expressing NOS in PPL1-α3 DANs that do not normally signal with NO is most easily explained if sGC is available for activation in all MB compartments. What are the molecular pathways downstream to cGMP? How do dopamine and NO signaling pathways interact in regulation of KC synapses? Previous studies and RNA-Seq data suggest several points of possible crosstalk. In cultured KCs from cricket brains, cGMP-dependent protein kinase (PKG) mediates NO-induced augmentation of a Ca2+ channel current. However, no expression of either of the genes encoding Drosophila PKGs (foraging and Pkg21D) was detected in KCs in RNA-Seq studies. On the other hand, cyclic nucleotide-gated channels and the cGMP-specific phosphodiesterase Pde9 are expressed in KCs. Biochemical studies have shown that the activity of sGC is calcium dependent and that PKA can enhance the NO-induced activity of sGC by phosphorylating sGC; sGC isolated from flies mutant for adenylate cyclase, rutabaga, show lower activity than sGC from wild-type brains, suggesting crosstalk between the cAMP and cGMP pathways (Aso, 2019).

All memory systems must contend with a tension between the strength and longevity of the memories they form. The formation of a strong immediate memory interferes with and shortens the lifetimes of previously formed memories, and reducing this interference requires a reduction in initial memory strength that can only be overcome through repeated exposure. Theoretical studies have argued that this tension can be resolved by memory systems that exhibit a heterogeneity of timescales, balancing the need for both fast, labile memory and slow consolidation of reliable memories. The mechanisms responsible for this heterogeneity, and whether they arise from complex signaling within synapses themselves), heterogeneity across brain areas, or both, have not been identified (Aso, 2019).

NO acts antagonistically to dopamine and reduces memory retention while facilitating the rapid updating of memory following a new experience. Viewed in isolation, the NO-dependent reduction in memory retention within a single compartment may seem disadvantageous, but in the presence of parallel learning pathways, this shortened retention may represent a key mechanism for the generation of multiple memory timescales that are crucial for effective learning. During shock conditioning, for example, multiple DANs respond to the aversive stimulus, including PPL1-γ1pedc, PPL1-γ2α'1, PPL1-α3. This study has shown that optogenetic activation of these DAN cell types individually induces negative-valence olfactory memory with distinct learning rates. The NOS-expressing PPL1-γ1pedc induces memory with the fastest learning rate in a wild-type background, and this study shows that it induces an NO-dependent memory trace when dopamine synthesis is blocked, with a much slower learning rate and opposite valence (Aso, 2019).

Robust and stable NO-dependent effects were only observed when training was repeated 10 times. Under such repeated training, compartments with slower learning rates, such as α3, form memory traces in parallel to those formed in γ1pedc. Thus, flies may benefit from the fast and labile memory formed in γ1pedc without suffering the potential disadvantages of 58 shortened memory retention, as long-term memories are formed in parallel in other compartments. The Drosophila MB provides a tractable experimental system to study the mechanisms and benefits of diversifying learning rate, retention, and flexibility in parallel memory units, as well as exploring how the outputs from such units are integrated to drive behavior (Aso, 2019).

The alpha'beta' subtype of Drosophila mushroom body neurons (MBn) is required for memory acquisition, consolidation and early memory retrieval after aversive olfactory conditioning. However, in vivo functional imaging studies have failed to detect an early forming memory trace in these neurons as reflected by an enhanced G-CaMP signal in response to presentation of the learned odor. This study shows that aversive olfactory conditioning suppresses the calcium responses to the learned odor in both alpha'3 and alpha'2 axon segments of alpha'beta' MBn and in the dendrites of alpha'3 MBOn immediately after conditioning using female flies. Notably, the cellular memory traces in both alpha'3 MBn and alpha'3 MBOn are short-lived and persist for less than 30 min. The suppressed response in alpha'3 MBn is accompanied by a reduction of acetylcholine (ACh) release, suggesting that the memory trace in postsynaptic alpha'3 MBOn may simply reflect the suppression in presynaptic alpha'3 MBn. Furthermore, this study shows that the alpha'3 MBn memory trace does not occur from the inhibition of GABAergic neurons via GABAA receptor activation. Since activation of the alpha'3 MBOn drives approach behavior of adult flies, the results demonstrate that aversive conditioning promotes avoidance behavior through suppression of the alpha'3 MBn-MBOn circuit (Zhang, 2019).

Animals learn to avoid a neutral stimulus that is repeatedly coupled with an unpleasant one. This type of learning, aversive associative learning, induces cellular memory traces in engram cells in the brain and changes the representation of the neutral stimulus. In Drosophila, several memory traces detected with the calcium indicator G-CaMP have been observed in the mushroom body (MB), a brain region critical for olfactory learning and memory. These memory traces are detectable across discrete time periods extending from 30 min to several days after training. However, memory traces that form immediately in the MB after conditioning have not been detected with in vivo Ca2+ imaging (Zhang, 2019).

The MB is composed of ~2000 intrinsic neurons in each hemisphere that integrates olfactory cues received from antennal lobe projection neurons with aversive or rewarding stimuli from two clusters (PPL1, PAM) of dopamine neurons. MBn are classified into three major subtypes: α'β', αβ, and γ neurons, based on their birth order and projection patterns of their axons in the brain. The axons of α'β' and αβ MBn bifurcate and project within the vertical α'/α lobe and horizontal β'/β lobe neuropil, whereas the axons of γ neurons project only within the horizontal γ lobe neuropil. Although each of these MBn subtypes contributes to aversive olfactory memory, they do so at different times after conditioning, with synaptic transmission from the α'β' and γ MBn required for robust expression of early and intermediate-term memory (immediate to 3 h) and synaptic transmission from the αβ MBn having a more pronounced role for memory expression after 3 h. Importantly, although the α'β' MBn are required for memory acquisition, consolidation and early memory retrieval, no immediate memory trace in α'β' MBns has been detected using in vivo Ca2+ imaging (Zhang, 2019).

Five different types of MB output neurons (MBOns) tile the α'β' lobe with their dendritic trees into five discrete compartments, matching the tiling by axon terminals from presynaptic DAns. Several of these MBOns are required for aversive memory or appetitive memory expression, and intermediate-term memory traces (~1-2 h after conditioning) have been detected in some of these neurons. However, early memory traces have not been documented in these MBOns, and the relationship between such putative traces and those in the presynaptic MBn is unexplored. Connectome studies revealed that DAns make direct connection with MBOns, opening the possibility that MBOns form traces independently of the MBn (Zhang, 2019).

This study shows that a cellular memory trace forms immediately after conditioning in the MBn axons occupying the α'3 compartment and in the downstream α'3 MBOn. Functional Ca2+ imaging reveals that aversive conditioning suppresses subsequent responses to the learned odor in both the presynaptic α'3 compartment and the postsynaptic α'3 MBOn across a similar time period, suggestive of a causal relationship. In vivo ACh imaging revealed that the suppressed Ca2+ responses are accompanied by reduced ACh release in the α'3 compartment, supporting the model that the α'3 MBOn memory trace occurs from suppressed presynaptic activity. This study also shows that the conditioning-induced suppression in the α'3 compartment does not occur from increased inhibition through the Resistance to dieldrin (Rdl) GABAA receptor, indicating that mechanisms other than Rdl receptor activation are responsible for the suppression of activity (Zhang, 2019).

This study provides evidence for the existence of immediate cellular memory traces that form in at least two adjacent segments of the axons in the vertical lobe neuropil of the α'β' MBn and at least one (α'3 MBOn) of the corresponding output neurons. These memory traces, detected as decreased Ca2+ responses to the CS+ odor immediately after conditioning when compared with preconditioning responses, and persisting for >30 min before the response properties return to the naive state, are consistent with the fact that α'β' MBn are required for memory acquisition, consolidation and early memory retrieval. Several other previously characterized early memory traces due to odor conditioning provide an interesting background to these newly discovered traces. The neurites of the DPM neurons innervating the vertical MB lobe neuropil exhibit an increased Ca2+ response to the learned odor from ~30-70 min after conditioning. A memory trace forms in the antennal lobe, registered as the recruitment of new projection neuron activity in response to the learned odor, that lasts <10 min after conditioning. The activity of GABAergic APL neurons that synapse in the vertical lobe neuropil of the MBn is suppressed for a period of a few minutes after conditioning. Further, in vivo functional imaging of the α'β' MBn axons revealed an early memory trace displayed as increased Ca2+ influx by 30 min after conditioning that persists for at least 1 h. The action of these five memory traces, together along with other unknown traces, may provide the cellular modifications required for behavioral performance gains to be made across the first hour after conditioning. Memory traces in compartments other than α'3 and/or their MBOns may underlie the requirement of α'β' MBn for memory retrieval beyond the first hour (Zhang, 2019).

However, the developmental trajectory of the memory traces forming in the α'β' MBn lobe is of additional interest. As indicated above, a cellular memory trace forms in these neurons by 30 min after conditioning that is manifested as an increased Ca2+ response to the conditioned odor. The data presented in this study show that the α'β' MBn axons become suppressed across the first ~15 min after conditioning. The combined studies thus indicate that the CS+ odor response properties in the α'β' MBn axons are initially suppressed after conditioning but then become enhanced at later times. The time courses for the two cellular memory traces do not match exactly (0-15 min for the suppression and ~30-60 min for the increase) given the current data showing no detectable increase at 30 or 45 min, but this is easily explained by variation in the strength of conditioning or minor technical differences between the two studies. Thus, the most parsimonious conclusion is that the vertical axon compartments of the α'β' MBn initially exhibit a suppressed response to the CS+ followed by an increased response with the transition from suppression to enhancement occurring somewhere between ~30-45 min after conditioning. How this evolution in response properties from negative to positive with time translates into behavioral memory expression remains unclear (Zhang, 2019).

The suppressed responses to the CS+ odor were found in both the axon segments of the α'β' MBn and the dendrites of α'3 MBOn. Given that activation of α'3 MBOn drives approach behavior, the results are consistent with the model that aversive conditioning promotes avoidance through suppressing the MBn-MBOn circuits that signal positive valence, at least across the time that the MBOn responses are suppressed. Notably, the memory traces in α'3 MBn and α'3 MBOn persisted for the similar time, raising the question of whether the suppressed responses form independently or whether the α'3 MBOn memory trace simply reflects the presynaptic one. The data support the model in which the suppressed α'β' MBn responses are simply transmitted to the MBOn from reduced synaptic activity: the suppressed Ca2+ response in α'3 MBn axon compartment is correlated with reduced ACh release and the suppressed response in the α'3 MBOn dendrites. Behavioral data suggest that if the early cellular memory traces that form in the α'3 MBn-MBOn circuit cannot be readout precisely, the expression of behavioral memory early after conditioning becomes impaired. However, the possibility that memory traces can be formed independently in α'3 MBOn cannot be excluded (Zhang, 2019).

This study formulated the hypothesis that the immediate suppression in α'3 MBn axons after aversive conditioning might be due to enhanced GABAergic input to the α'3 compartment in an effort to delineate the underlying mechanism. However, attempts to detect any impairment of the immediate suppression in α'3 axonal compartment failed when Rdl GABAA receptor was knocked down in α'β' neurons. Thus, the data argue against attributing the suppression in the α'3 compartment to GABAergic inhibition through GABAA receptor (Zhang, 2019).

Two neuronal populations, c673a and Fru-GAL4, regulate fat storage in fruit flies. Both populations partially overlap with a structure in the insect brain known as the mushroom body (MB), which plays a critical role in memory formation. This overlap prompted an examination of whether the MB is also involved in fat storage homeostasis. Using a variety of transgenic agents, the neural activity of different portions of the MB and associated neurons were selectively manipulated to decipher their roles in fat storage regulation. The data show that silencing of MB neurons that project into the &alpha:'β' lobes decreases de novo fatty acid synthesis and causes leanness, while sustained hyperactivation of the same neurons causes overfeeding and produces obesity. The &alpha:'β' neurons oppose and dominate the fat regulating functions of the c673a and Fru-GAL4 neurons. It was also shown that MB neurons that project into the γ lobe also regulate fat storage, probably because they are a subset of the Fru neurons. It was possible to identify input and output neurons whose activity affects fat storage, feeding, and metabolism. The activity of cholinergic output neurons that innervating the β'2 compartment (MBON-β'2mp and MBON-γ5β'2a) regulates food consumption, while glutamatergic output neurons innervating α' compartments (MBON-γ2α'1 and MBON-α'2) control fat metabolism. This study has identified a new fat storage regulating center, the α'β' lobes of the MB. The study also delineated the neuronal circuits involved in the actions of the α'β' lobes, and showed that food intake and fat metabolism are controlled by separate sets of postsynaptic neurons that are segregated into different output pathways (Al-Anzi, 2018).

Current thought envisions dopamine neurons conveying the reinforcing effect of the unconditioned stimulus during associative learning to the axons of Drosophila mushroom body Kenyon cells for normal olfactory learning. This study shows, using functional GFP reconstitution experiments, that Kenyon cells and dopamine neurons form axoaxonic reciprocal synapses. The dopamine neurons receive cholinergic input via nicotinic acetylcholine receptors from the Kenyon cells; knocking down these receptors impairs olfactory learning revealing the importance of these receptors at the synapse. Blocking the synaptic output of Kenyon cells during olfactory conditioning reduces presynaptic calcium transients in dopamine neurons (DAn), a finding consistent with reciprocal communication. Moreover, silencing Kenyon cells decreases the normal chronic activity of the dopamine neurons. These results reveal a new and critical role for positive feedback onto dopamine neurons through reciprocal connections with Kenyon cells for normal olfactory learning (Cervantes-Sandoval, 2017).

The results demonstrate that DAn are both pre- and post-synaptic to KC through axoaxonic reciprocal connections, in contrast to current models which envision them as providing only pre-synaptic input. The DAn>KC half of the reciprocal synapse employs DA as neurotransmitter, although the possibility cannot be ruled out that other neurotransmitters are co-released with DA. The KC>DAn half of the reciprocal synapse is cholinergic. However, the fact that it was not possible to observe both cAMP and calcium responses in DAn with KC stimulation suggests that there may be other mediators of this reciprocal connection. Blocking the cholinergic input to DAn attenuates aversive olfactory learning, providing evidence that its function is, at least in part, to provide an amplification signal for the initial DA release due to activating the US pathway. Consistent with this role it was found that silencing KC impairs DAn presynaptic calcium responses to conditioning, odor and shock stimuli, presumably influencing dopamine release and explaining the learning phenotype. Overall, results support the existence of a positive feedback loop required for optimal learning. It is envisioned that DAn receive direct input from the US during conditioning which is conveyed to KC. The KCs also receive coincident olfactory input and this coincidence provides positive feedback onto the DAn through cholinergic synapses to further increase DAn activity (Cervantes-Sandoval, 2017).

The results also show that KC input to DAn shapes their ongoing or chronic activity. It is plausible that ongoing activity in the DAn provides a moment-by-moment update of the external environment and internal states and the behavioral status of the fly that appropriately reconfigures the KC>MBOn flow of information. Thus, the DAn/KC/MBOn circuit may form a recurrent network that serves as the insect's brain center for the rapid integration of sensory information and decision-making. Local feedback loops, achieved by reciprocal connectivity like that described in this study, may provide computational benefits to fine tune and optimize the output. Behavioral flexibility may be achieved by passing information through local reentrant loops with constant updating from the external or internal state of the organism (Cervantes-Sandoval, 2017).

A caveat in this and other studies is that it is not possible to exclude that other, non-KC, cholinergic input to DAn contributes to memory acquisition. Massive efforts to generate 'connectomes' in multiple species may offer resolution to this issue at some point in the future. Alternatively, they may continue to reveal additional connections and complexities that defy an immediate understanding. Studies like the present one, that reveal unexpected relationships between synaptic partners in difficult-to-untangle circuits, expose the need to advance beyond 'connectomics' and develop new tools that allow silencing or activation of specific channels and specific synaptic connections between neurons of interest without affecting other functions in the same cells (Cervantes-Sandoval, 2017).

Go to top

Associating stimuli with positive or negative reinforcement is essential for survival, but a complete wiring diagram of a higher-order circuit supporting associative memory has not been previously available. This study reconstructed one such circuit at synaptic resolution, the Drosophila larval mushroom body. Most Kenyon cells were found to integrate random combinations of inputs, but a subset receives stereotyped inputs from single projection neurons. This organization maximizes performance of a model output neuron on a stimulus discrimination task. A novel canonical circuit in each mushroom body compartment with previously unidentified connections is reported: reciprocal Kenyon cell to modulatory neuron connections, modulatory neuron to output neuron connections, and a surprisingly high number of recurrent connections between Kenyon cells. Stereotyped connections found between output neurons could enhance the selection of learned behaviours. The complete circuit map of the mushroom body should guide future functional studies of this learning and memory centre (Eichler, 2017).

Massively parallel, higher-order neuronal circuits such as the cerebellum and insect mushroom body (MB) serve to form and retain associations between stimuli and reinforcement in vertebrates and evolutionarily complex invertebrates. Although these systems provide a biological substrate for adaptive behaviour, no complete synapse-resolution wiring diagram of their connectivity has been available to guide analysis and aid understanding. The MB is a higher-order parallel fibre system in many invertebrate brains, including hemimetabolous as well as holometabolous insects and their larval stages. MB function is essential for associative learning in adult insects and in Drosophila larvae from the earliest larval stages onwards. Indeed, the basic organization of the adult and the larval MB and their afferent circuits is very similar; however, larvae have about an order of magnitude fewer neurons. Thus, to systematically investigate the organizational logic of the MB, this study used serial section electron microscopy to map with synaptic resolution the complete MB connectome in a first-instar Drosophila larva. L1 are foraging animals capable of all behaviours previously described in later larval stages including adaptive behaviours dependent on associative learning. Their smaller neurons enable fast electron microscopy imaging of the entire nervous system and reconstruction of complete circuits (Eichler, 2017).

Models of sensory processing in many neural circuits feature neurons that fire in response to combinations of sensory inputs, generating a high-dimensional representation of the sensory environment. The intrinsic MB neurons, the Kenyon cells (KCs), integrate in their dendrites inputs from combinations of projection neurons (PNs) that encode various stimuli, predominantly olfactory in both adult and larva1, but also thermal, gustatory, and visual in adult and larva. Previous analyses in adults and larvae suggest that the connectivity between olfactory PNs and KCs is random, but they do not eliminate the possibility of some degree of bilateral symmetry, which requires access to the full PN-to-KC wiring diagram in both hemispheres (Eichler, 2017).

The MB contains circuitry capable of associating reward or punishment with the representation of the sensory environment formed by KCs. KCs have long parallel axons that first run together, forming the peduncle, and then bifurcate, forming the so-called lobes, in both larvae and adults. KCs receive localized inputs along their axonal fibres from dopaminergic as well as octopaminergic modulatory neurons (DANs and OANs, respectively) that define separate compartments. DANs and OANs have been shown to convey reinforcement signals in adult insects and larval Drosophila. The dendrites of the mushroom body output neurons (MBONs) respect the DAN compartments in adults and larvae. It has been shown in adult Drosophila that co-activation of KCs and DANs can associatively modulate the KC-MBON synapse. Thus, the compartments represent anatomical and functional MB units where sensory input (KCs) is integrated with internal reinforcement signals (DANs/OANs) to modulate instructive output for behavioural control (MBONs). However, the synaptic connectivity of KCs, DAN/OANs, and MBONs at this crucial point of integration was previously unknown (Eichler, 2017).

Furthermore, studies in adult Drosophila have shown that, despite the compartmental organization of the MB, many MBONs interact with MBONs from other compartments, suggesting that the MBON network functions combinatorially during memory formation and retrieval. However, a comprehensive account of all MB neuron connections is lacking. Thus, to provide a basis for understanding how the MB, a prototypical parallel fibre learning and memory circuit, functions as an integrated whole, this study provides a full, synapse-resolution connectome of all MB neurons of an L1 Drosophila larvaxxWe provide the first complete wiring diagram of a parallel fibre circuit for adaptive behavioural control. Such circuits exist in various forms, for example the cerebellum in vertebrates and the MB in insects. They contribute to multiple aspects of behavioural control including stimulus classification, the formation and retrieval of Pavlovian associations, and memory-based action selection. A comprehensive wiring diagram of such a multi-purpose structure is an essential starting point for functionally testing the observed structural connections and for elucidating the circuit implementation of these fundamental brain functions (Eichler, 2017).

Even though individual neurons may change through metamorphosis, many of the basic aspects of the MB architecture are shared between larval and adult Drosophila stages and with other insects. It is therefore expected that the circuit motifs identified in this study are not unique to the L1 developmental stage, but instead represent a general feature of Drosophila and insect MBs (Eichler, 2017).

Electron microscopy reconstruction revealed a canonical circuit in each MB compartment featuring two unexpected motifs in addition to the previously known MBIN-to-KC and KC-to-MBON connections. First, it was surprising to observe that the number of KC-to-MBIN and KC-to-MBON synapses are comparable. As KCs were shown to be cholinergic in adults, KC-to-MBIN connections could be potentially depolarizing. Untrained, novel odours can activate DANs in adult Drosophila and OANs in bees. Similar brief short-latency activations of dopamine neurons by novel stimuli are observed in monkeys, too, and are interpreted as salience signals. Learning could potentially modulate the strength of the KC-to-MBIN connection, either weakening it or strengthening it. The latter scenario could explain the increase in DAN activation by reinforcement-predicting odours observed in adult Drosophila, bees, and monkeys. In addition, dopamine receptors have been shown to be required in Drosophila KCs for memory formation. Another unexpected finding was that MBINs synapse directly onto MBONs, rather than only onto KCs. Such a motif could provide a substrate for neuromodulator-gated Hebbian spike-timing dependent plasticity, which has been observed in the locust MB (Eichler, 2017).

In addition to random and bilaterally asymmetric olfactory and structured non-olfactory PN-to-KC connectivity, the analysis identified single-claw KCs whose number and lack of redundancy are inconsistent with random wiring. Random wiring has previously been shown to increase the dimension of sensory representations when the number of neurons participating in the representation is large compared with the number of afferent fibres, as in the cerebellum or adult MB. However, the current model shows that when the number of neurons is limited, random wiring alone is inferior to a combination of random and structured connectivity that ensures each input is sampled without redundancy. The presence of single-claw KCs may reflect an implementation of such a strategy. In general, the results are consistent with a developmental program that produces complete and high-dimensional KC sensory representations to support stimulus discrimination at both larval and adult stages (Eichler, 2017).

This study reveals the complete MBON-MBON network at synaptic resolution. Previous studies in the larva have shown that odour paired with activation of medial and vertical lobe DANs leads to learned approach8 and avoidance, respectively. This connectivity analysis reveals that glutamatergic MBONs from the medial lobe laterally connect to MBONs of the vertical lobe. The glutamatergic MBON-MBON connections could be inhibitory, although further studies are needed to confirm this. Furthermore, inhibitory GABAergic MBONs from the vertical lobe laterally connect to MBONs of the medial lobe. An example is the feedforward inhibition of medial lobe MBON-i1 output neuron by the vertical lobe GABAergic MBON-g1, -g2 output neurons. A similar motif has been observed in the Drosophila adult, where aversive learning induces depression of conditioned odour responses in the approach-promoting MBON-MVP2 (MBON-11), which in turn disinhibits conditioned odour responses in the avoidance-promoting MBON-M4/M6 (MBON-03) because of the MBON-MVP2 to MBON-M4/M6 feedforward inhibitory connection (Eichler, 2017).

Combining the present connectivity analysis of the MBON-MBON network in the larva and previous studies in the adult Drosophila, the rule seems to be that MBONs encoding opposite learned valance laterally inhibit each other. Such inhibitory interactions have been proposed as a prototypical circuit motif for memory-based action selection (Eichler, 2017).

Go to top

Animals exhibit a behavioral response to novel sensory stimuli about which they have no prior knowledge. This study has examined the neural and behavioral correlates of novelty and familiarity in the olfactory system of Drosophila. Novel odors elicit strong activity in output neurons (MBONs) of the α'3 compartment of the mushroom body that is rapidly suppressed upon repeated exposure to the same odor. This transition in neural activity upon familiarization requires odor-evoked activity in the dopaminergic neuron innervating this compartment. Moreover, exposure of a fly to novel odors evokes an alerting response that can also be elicited by optogenetic activation of α'3 MBONs. Silencing these MBONs eliminates the alerting behavior. These data suggest that the α'3 compartment plays a causal role in the behavioral response to novel and familiar stimuli as a consequence of dopamine-mediated plasticity at the Kenyon cell-MBONα'3 synapse (Hattori, 2017).

Novel stimuli can elicit a behavioral response that alerts an organism to unexpected, potentially salient events. An alerting response to sensory stimuli not previously encountered by an animal, such as the orienting response described by Pavlov, provides an organism the opportunity to explore the potential significance of the novel stimulus. A behavioral response to novelty is elicited by sensory cues about which an animal has no prior knowledge. Most behaviors, in contrast, are based upon past experience acquired either over long periods of evolutionary time (innate behaviors) or by learning over the life of an animal. The observation that sensory cues can be identified as novel and can evoke a behavioral response in the absence of prior knowledge poses an interesting problem (Hattori, 2017).

A neural circuit encoding novelty should respond to all novel stimuli, but this response should be suppressed upon familiarization. The memory of a familiar sensory cue should be stimulus specific and long-lasting, distinguishing it from sensory adaptation. Neural responses that correlate with novelty and familiarity are seen in a number of mammalian brain regions. The transition from novelty to familiarity is associated with suppression of neural responses in higher brain centers that appears distinct from intrinsic or sensory adaptation. Electrophysiologic recordings along the visual and auditory pathways reveal that neurons exhibit activity in response to novel or unexpected cues that diminish upon repeated exposure. In the auditory pathway, neurons in the inferior colliculus and the auditory cortex exhibit responses to novel or unexpected tones that attenuate upon repetition. Similarly, neurons in the perirhinal and inferior temporal cortices respond to novel visual stimuli, and this response attenuates rapidly upon repetition, a phenomenon known as repetition suppression (Hattori, 2017).

Dopaminergic neurons in the substantia nigra (SN) pars compacta and ventral tegmental area (VTA) also exhibit phasic bursting activity in response to novel or unexpected sensory events. Unexpected flashes of light or auditory tones evoke burst firing in 60%-70% of the dopaminergic neurons that attenuates as the novel stimulus becomes familiar. Related neural events may underlie attenuation in the BOLD signal observed in extra striate cortex as well as SN and VTA in fMRI studies of humans upon repeated exposure to sensory stimuli. Thus, mammals have evolved neural systems that distinguish novel from familiar sensory stimuli that may facilitate the determination of the potential salience of unfamiliar environmental events (Hattori, 2017).

This study has analyzed the behavioral and neural correlates of novelty and familiarity in the olfactory system of Drosophila. Olfactory perception in the fly is initiated by the binding of an odor to an ensemble of olfactory sensory neurons in the antennae that results in the activation of a unique and topographically fixed combination of glomeruli in the antennal lobe. Each antennal lobe projection neuron extends dendrites into one of the 54 glomeruli and extends axons that bifurcate to innervate two distinct brain regions, the lateral horn, and the mushroom body (MB). The invariant circuitry of the lateral horn is thought to mediate innate behaviors, whereas the unstructured projections to the MB translate olfactory sensory information into learned behavioral responses. In the MB, each odor activates a sparse representation (5%-10%) of principal neurons, the Kenyon cells (KCs). KCs extend axons that form en passant synapses in the compartments of the MB lobes. The KCs synapse on the MB output neurons (MBONs), which have distinct spatially stereotyped dendritic arbors within compartments that collectively tile the lobes. MBONs provide the only output of the MB and the activity of different MBON combinations biases behavior. Each of the 15 compartments is also innervated by the axons of one to three of 20 dopaminergic cell types (dopaminergic neurons, or DANs). Distinct DANs respond to different unconditioned stimuli and dopamine release elicits plasticity in the synapses between the KCs and MBONs. The alignment of DAN arbors with compartmentalized KC-MBON synapses creates a unit for learning that transforms the disordered KC representation into ordered MBON output to collectively bias behavioral responses to sensory stimuli (Hattori, 2017).

The transition from novelty to familiarity involves memory formation, and learning and memory in the fly are accomplished by the circuitry of the MB. This study has identified a neural circuit in the MB that appears to encode a representation of novelty and familiarity. MBONs were observed innervating the α'3 compartment respond to novel odors and that their activity is rapidly suppressed upon repeated exposure to the same stimulus. This suppression upon familiarization is observed for all novel odors tested regardless of innate valence, is stimulus specific, lasts for more than 20 min, and recovers in 1 hr. Repetition suppression of MBON-α'3 is distinct from sensory adaptation and requires odor-evoked activity in the DAN innervating the α'3 compartment. These data suggest that repeated exposure to an odor mediates dopamine-dependent plasticity at the KC synapses onto MBON-α'3 that suppresses MBON output. Moreover, behavioral experiments demonstrate that the α'3 MBONs mediate an alerting response to novel odors. Exposure of a fly to novel olfactory stimuli evokes an alerting behavior. This behavioral response can be elicited by optogenetic activation of α'3 MBONs and eliminated by α'3 MBON silencing. These observations suggest that the behavioral response to novelty and the transition to familiarity is mediated by the circuitry of KCs, DANs, and MBONs within the α'3 compartment (Hattori, 2017).

This study has examined the neural and behavioral correlates of novelty and familiarity in the olfactory system of Drosophila. Exposure of flies to a novel odor interrupts grooming. This alerting response is dependent upon the activity of output neurons of the α'3 compartment of the MB. Optogenetic activation of the α'3 MBONs elicits an alerting response, whereas silencing these neurons eliminates the behavioral response to novel odors. A neural correlate of this behavioral response is observed in the activity of MBON-α'3. Novel odors elicit strong activity in these neurons that is rapidly suppressed upon repeated exposure to the same odor. This transition in MBON-α'3 response upon familiarization requires the activity of PPL1-α'3, the DAN innervating the α'3 compartment, and dopamine receptors in the MBONs. These data suggest that the α'3 compartment may play a causal role in the behavioral response to novel and familiar stimuli as a consequence of dopamine-mediated plasticity at the KC-MBONα'3 synapse. Although the circuitry of the α'3 compartment is central to the behavioral response to novel and familiar odors, the data do not exclude a contribution from other compartments (Hattori, 2017).

Plasticity at the KC-MBON synapses has been invoked to explain olfactory learning and memory (Heisenberg, 2003). In associative learning, exposure to a conditioned stimulus (CS), when paired with an unconditioned stimulus (US), imposes an associative memory upon the CS. The identity of the CS is represented by activity in a specific ensemble of KCs, whereas USs of different valence activate distinct DANs. Dopamine input depresses the KC-MBON synapse in specific compartments to transform the unstructured KC representation of an odor into an ordered MBON representation encoding behavioral bias. The neural mechanism governing the novelty response differs from this classical model of associative learning. In the α'3 compartment, a novel odor elicits strong MBON output and also activates the DAN. Dopamine release from PPL1-α'3 depresses the KC-MBONα'3 synapse, suppressing MBON output on further exposure to this odor. In this manner, a novel odor effectively serves as both a CS and a US to drive the transition from novelty to familiarity (Hattori, 2017).

A second distinction between the novelty response and associative learning emerges from the observation that the response to novelty is suppressed by learning whereas the conditioned response depends on learning. The stereotyped alerting behavior in response to novel odor does not require learning and is therefore innate. In associative learning models, exposure to odor prior to learning activates all MBONs examined, but this combinatorial of MBONs does not elicit a behavior. Rather, behavioral bias is imposed by the suppression of specific MBON output after learning. In the novelty response, innate alerting behavior is elicited by strong output from MBON-α'3 in response to novel odors prior to learning. Learning that accompanies the transition to familiarity suppresses MBON-α'3 activity and the behavioral response to novelty. If, however, the novel odor is accompanied by salient events in the environment, this will result in activation of additional DANs, leading to the formation of associations in other compartments. The alerting response evoked by MBON-α'3 may enhance the awareness of these environmental events. Thus, α'3 output may elicit an immediate and stereotyped response to an odor independent of its salience, which is then assessed by the remaining MB compartments to mediate more measured associative responses (Hattori, 2017).

Odor-evoked dopamine release by PPL1-α'3 appears to be essential to modulate MBON output in the transition from novelty to familiarity. Dopamine release also contributes to decay in the memory of familiar odors. MBON-α'3 activity in response to novel odors is suppressed upon repeated exposure, but activity is restored after 1 hr. Dopamine release in the absence of odor, following repetition suppression, accelerates this recovery process. These observations are consistent with recent experiments demonstrating that dopamine release within a compartment in the absence of odor can lead to synaptic facilitation and the restoration of MBON output. In this manner, familiarity in the fly is a transient phenomenon and the restoration of the perception of novelty may be accelerated by dopamine (Hattori, 2017).

It is suggested that the neural events responsible for the transition from novelty to familiarity involve depression of only those KC-MBON synapses activated by the novel odor. In this manner, novelty and familiarity can be both universal and odor specific. A given odor activates about 5%-10% of the KCs in the MB and by inference 5%-10% of the KC-MBON synapses. An organism is likely to encounter multiple odors that may be construed as novel in the course of hours. If a single novel odor suppresses 5%-10% of the KC-MBON synapses and this synaptic depression is long term, all KC-MBON synapses within the α'3 compartment would be depressed after exposure to roughly 20 novel odors. Depression of all the synapses would prevent subsequent response to novel odors. The data suggest this problem may be obviated in two ways. First, the depression is relatively short lived. Second, although novel odors result in the depression of active KC-MBON synapses, they also enhance the recovery of previously suppressed but currently inactive synapse. This implies that the rate of synaptic recovery is proportional to the rate of exposure to novel odors. Thus, the α'3 compartment has evolved a mechanism to assure that novelty responses can be generated in both dense and sparse odor environments without saturation (Hattori, 2017).

The MB is an associative center in invertebrate brains thought to impose valence on sensory representations. The current data suggest that the MB not only functions in classical learning paradigms, but also supports novelty detection and the transition to familiarity. An organism can have no knowledge of a novel stimulus, and hence it exhibits an indiscriminate alerting response. The MB also integrates information about the organism's internal state (hunger, satiety, sleep, wakefulness, roaming, dwelling) allowing the fly to more comprehensively contextualize the diverse sensory experiences it may encounter throughout its life. Thus, the MB may afford the fly an 'individuality' allowing different flies to respond differently to the same stimuli in accord with its unique history and current state (Hattori, 2017).

Go to top

Understanding memory formation, storage and retrieval requires knowledge of the underlying neuronal circuits. In Drosophila, the mushroom body (MB) is the major site of associative learning. This study reconstructed the morphologies and synaptic connections of all 983 neurons within the three functional units, or compartments, that compose the adult MB's alpha lobe, using a dataset of isotropic 8 nm voxels collected by focused ion-beam milling scanning electron microscopy. It was found that Kenyon cells (KCs), whose sparse activity encodes sensory information, each make multiple en passant synapses to MB output neurons (MBONs) in each compartment. Some MBONs have inputs from all KCs, while others differentially sample sensory modalities. Only 6% of KC>MBON synapses receive a direct synapse from a dopaminergic neuron (DAN). Two unanticipated classes of synapses, KC>DAN and DAN>MBON, were identified. DAN activation produces a slow depolarization of the MBON in these DAN>MBON synapses and can weaken memory recall (Takemura, 2017).

Associative memory helps animals adapt their behaviors to a dynamically changing world. The molecular mechanisms of memory formation are thought to involve persistent changes in the efficiency of synaptic transmission between neurons. In associative learning, persistent changes in synaptic efficacy correlated with memory formation have been found at points of convergence between two neuronal representations: one providing information from sensory inputs about the outside world and a second indicating whether the current environment is punitive or rewarding. Such sites of convergence have been identified for multiple forms of associative learning. However, a comprehensive synaptic level description of connectivity at such a site of convergence is not available for an animal as complex as the fruit fly, Drosophila (Takemura, 2017).

The mushroom body (MB) is the center of associative learning in insects. Sensory information enters the MB via the calyx, where the dendritic claws of Kenyon cells (KCs) receive synaptic inputs from projection neurons of olfactory and other modalities including visual, gustatory and thermal. The parallel axonal fibers of the KCs form the MB-lobes, the output region of the MB. A pattern of sparse activity in the KC population represents the identity of the stimulus. This sparseness is maintained through two mechanisms. First, individual KCs generally only spike when they receive simultaneous inputs from multiple projection neurons. Second, overall KC excitability is regulated by feedback inhibition from a GABAergic neuron, MB-APL, that arborizes throughout the MB. Thus, only a small subset of KCs respond to a given sensory stimulus. Upon this representation of the sensory world, dopaminergic or octopaminergic neurons convey information of punishment or reward and induce memories that associate the sensory stimulus with its valence (Takemura, 2017).

The functional architecture of the MB circuit is best understood in adult Drosophila (see Diagram of the α lobe of the mushroom body). In each MB, the parallel axonal fibers of ~2000 KCs can be divided into 16 compartmental units by the dendrites of 21 types of MB output neurons (MBONs) and the axon terminals of 20 types of dopaminergic neurons (DANs). A large body of behavioral and physiological studies suggests that these anatomical compartments are also parallel units of associative learning. In each compartment, the dendrites of a few MBONs overlap with axon bundles of hundreds of KCs. Punishment and reward activate distinct sets of DANs. DAN input to a compartment has been shown to induce enduring changes in efficacy of KC>MBONs synapses in those specific KCs that were active in that compartment at the time of dopamine release. The valence of the memory appears to be determined by which compartment receives dopamine during training, while the sensory specificity of the memory is determined by which KCs were active during training (Takemura, 2017).

Compartments can have distinct rates of memory acquisition and decay, and the 16 compartments together appear to form a set of parallel memory units whose activities are coordinated through both direct and indirect inter-compartmental connections. The DANs which project to the α1 compartment, the ventral-most compartment of the vertical lobe, play a key role in the formation of appetitive long-term memory of nutritional foods. DANs that project to the other α lobe compartments, α2 and α3, play roles in aversive long-term memory. All three of these compartments receive feedforward inputs from GABAergic and glutamatergic MBONs whose dendrites lie in other MB compartments known to be involved in aversive or appetitive memory. In addition, two types of MB-intrinsic neurons send arbors throughout the MB-lobes: a large GABAergic neuron, MB-APL, which provides negative feedback important for sparse coding, and the MB-DPM neuron, which is involved in memory consolidation and sleep regulation (Takemura, 2017).

Previous EM studies in the MB lobes of cockroaches, locusts, crickets, ants, honey bees and Drosophila identified KCs by their abundance, fasciculating axons and small size. Additionally, large GABA immunoreactive neurons that contact KC axons were identified in the locust pedunculus. While these data provided early insights to guide modeling of the MB circuit, the volumes analyzed were limited and most neuronal processes could not be definitively assigned to specific cell types. This paper reports a dense reconstruction of the three compartments that make up the α lobe of an adult Drosophila male. Because a dense reconstruction was performed, with the goal of determining the morphology and connectivity of all cells in the volume, it can be confidently stated that all cell types with processes in the α lobe have been identified (Takemura, 2017).

Comprehensive knowledge of the connectivity in the α lobe has allowed addressing of several outstanding issues. The first concerns the nature of KC>MBON connectivity. Although each KC passes through all three compartments, it is not known if individual KCs have en passant synapses in each compartment. Thus, it remains an open question whether the sensory representation provided to each compartment and each MBON within a compartment is the same or whether different MBONs within a compartment might sample from non-overlapping sets of KCs, and thus use independent sensory representations for learning. It was also not known which, if any, other cell types are direct postsynaptic targets of KCs (Takemura, 2017).

The second concerns dopamine modulation. What are the locations of dopaminergic synapses and what does this distribution imply about the targets of dopaminergic modulation as well as volume versus local transmission? Cell-type-specific rescue of dopamine receptor mutants suggests that dopamine acts presynaptically in the KCs of KC>MBON synapses. However, postsynaptic mechanisms have also been proposed and a recent study detected expression of dopamine receptors in MBONs, raising the possibility that MBONs might also be direct targets of DAN modulation. Behavioral, imaging and electrophysiological data indicate that dopamine modulation respects the borders between compartments, but it is not known whether these borders have a distinct structure, such as a glial sheet (Takemura, 2017).

The third concerns the two MBON types that send feedforward projections into the α lobe. These MBONs have important roles in associative learning as revealed by behavioral assays and have been postulated to integrate memories of opposing valence and different time scales. However, it is not known which cell types these feedforward MBON projections targets within the MB (Takemura, 2017).

The fourth concerns the two neurons, MB-APL and MB-DPM, which arborize throughout the MB and are thought to regulate MB function globally. What is their local synaptic connectivity within the α lobe and what can this inform about how they perform their roles (Takemura, 2017)?

Finally, the three compartments of the α lobe differ in important aspects, including valence of the memory formed, the time course of memory formation and retrieval, and the numerical complexity of their DAN inputs and MBON outputs. Are there obvious differences in the microcircuits of different compartments (Takemura, 2017)?

In this paper reports the answers to these questions. In addition, the utility of detailed anatomy at the electron microscopic level to provide novel insights is demonstrated: It is shown that nearly all cell types in the α lobe contain more than one morphological class of synaptic vesicle, raising the possibility that these cells utilize multiple neurotransmitters. In addition, two prevalent sets of synaptic motifs - from DANs to MBONs and from KCs to DAN s- are described that were unanticipated despite the extensive anatomical, physiological, behavioral and theoretical studies that have been performed on the insect MB. These novel DAN to MBON connections are characterized using behavioral and physiological assays and find that DAN activation produces a slow depolarization of postsynaptic MBONs and can weaken memory recall (Takemura, 2017).

The connections between the neurons observed in this study are are summarized in a Summary diagram of the connectome reconstruction of the α lobe. In each of the lobe's three compartments, parallel axonal fibers of ~1000 KCs project through the dendrites of a few MBONs and the terminal arbors of a few DANs. The results provide support for several aspects of the generally accepted model for MB circuit function. First, it was found that each KC forms en passant synapses with multiple MBONs down the length of its axon, making it possible for parallel processing across the different compartments of the MB lobes. Secondly, with the assumption that released dopamine diffuses locally, KC>MBON synapses would receive dopaminergic input close to the sites of vesicle release, consistent with the prevailing hypothesis that plasticity occurs at the presynaptic terminals of KCs. However, several circuit motifs were found that were not anticipated by previous work. For example, synaptic connections were found from KCs to DANs, indicating that DANs get axo-axonal inputs within the MB lobes themselves. A recent report provides evidence that these KC>DAN synapses are functional (Cervantes-Sandoval, 2017). An even more unexpected motif was the direct synaptic contacts from DAN to MBON found in every compartment. Functional connectivity experiments confirmed that these connections are monosynaptic, and showed that they give rise to a slow depolarization in the MBON. Moreover, stimulating DANs in freely behaving flies yields effects consistent with a net excitatory DAN>MBON connection. Finally, the synaptic connections are described of two feedforward MBONs, which have been proposed to mediate the interaction of the various parallel memories within the MB lobes, as well as two intrinsic MB neurons, APL and DPM (Takemura, 2017).

This work not only provides definitive evidence for, and quantitative detail about, many previously observed circuit motifs, but also reveals several motifs not anticipated by prior anatomical, behavioral or theoretical studies. These additional circuit motifs provide new insights and raise new questions about the computations carried out by the MB. It is noted that these same novel connections were also found in a parallel study of the larval MB. Not only were the same circuit motifs found in the larval MB and adult α lobe, but also the relative prevalence of these connections was strikingly similar: DAN>MBON synapses were 4.5% the number of KC>MBON synapses in the adult α lobe and 3.4% in the larval MB. KC>DAN synapses were 1.5 times as prevalent as DAN>KC synapses in the adult α lobe, as compared with 1.1 in the larval MB. KCs make 48% of their synapses onto other KCs in the adult α lobe and 45% in the larval upper vertical lobe compartments. It is tempting to speculate that the conservation of the relative abundances of these connections across developmental stages reflects important functional constraints on the circuit (Takemura, 2017).

A large body of work supports the idea that individual KC>MBON synapses are the elemental substrates of associative memory storage in the MB. The dominant hypothesis in the field is that coincidence detection occurs within the presynaptic terminals of the KCs. The Conditioned Stimulus (CS, for example an odor) evokes a spiking response in a sparse subset of KCs, which in turn leads to Ca2+ influx. The Unconditioned Stimulus (US, for example electric shock) activates dopaminergic inputs to the MB lobes, where they likely activate G-protein-coupled dopamine receptors on the KC cell membrane. The coincidence of these two events is thought to be detected by the Ca2+ sensitive, calmodulin-dependent adenylate cyclase rutabaga, which initiates a cAMP signaling cascade that leads to the biochemical changes underlying synaptic plasticity (Takemura, 2017).

The tiling of MBON and DAN projections down the length of the KC axons suggests that each of these compartments serves as an independent module, with the association of reinforcement with sensory input taking place in parallel across several different modules. One important assumption in this model is that each KC sends parallel input to each compartment by making synapses all the way down the length of its axon. Light microscopic imaging established that the axons of individual α/β KCs do indeed run through all three compartments of the α lobe. However, they also revealed that the axonal branching patterns differ between KC classes. For example, the axons of α/βp KCs branch in α2, whereas those of α/βc and α/βs KCs do not, raising the question of how extensive KC outputs are across the different compartments. The dense EM reconstruction established that in fact all α/β KCs form en passant synapses on MBONs in each of the three α lobe compartments (Takemura, 2017).

In many cases, these synapses were found at enlarged boutons that contained the presynaptic machinery. However, output sites were also found on the smooth axons of the α/βc KCs, which lack obvious bouton-like swellings. Only occasional, short (generally <5 μm) segments of KC axons where the axon became thinner than 300 nm in diameter lacked presynaptic sites. Of course, it is not known whether all these synapses are functional. EM analysis showed that within each compartment, every KC passing through a layer of the compartment that was extensively innervated by an MBON made at least one synapse with that MBON. Previous electrophysiological measurements of connectivity in the α2 compartment indicated that only about 30% of KCs connect to MBON-α2sc, suggesting the possibility that the majority of KC>MBON synapses are functionally silent, as they are in cerebellar cortex, where 98% of the parallel fiber-to-Purkinje cell synapses are believed to be silent. However, a more trivial explanation cannot be ruled out: These measurements were made in the presence of cholinergic antagonists that could have partially blocked synaptic events and lead to an underestimate of total connectivity levels (Takemura, 2017).

The EM data revealed that the the number of synapses made by individual KCs was well-described by a Poisson distribution, where each synapse connects with a uniform, independent, and random probability to one of the KCs. Although the predicted distributions strongly depend on the number of connections between two cell types, almost all KC connections to other cells obeyed Poisson statistics. This was true of every KC in the α1 and α3 compartments, where each MBON has compartment-filling dendrites. The α2 compartment is somewhat unusual in that its MBONs innervate only subzones of the compartment. While light microscopy showed that MBON-α2sc primarily innervates the surface and core of the compartment, MBON-α2sp was found to project more to the surface and posterior. The connectome results bore out these observations from the light and electron microscopy, although EM reconstructions also showed that these borders were not sharp, and these MBONs receive less extensive and weaker connections outside these subzones. Nevertheless, within the primary area of innervation, it was again the case that every KC made synapses with all MBONs along its axon. Thus each of the 949 α/β KCs can deliver information to the MBONs in each of the three α lobe compartments (Takemura, 2017).

A strictly feed-forward view of the circuit may miss important processing, however, as earlier studies suggested, and the current results re-emphasize. Firstly, gap junctions between KCs have been reported. This opens up the possibility for lateral propagation of signals across KCs, either biochemical or electrical. For example, in mammalian systems, axo-axonal gap junction coupling can synchronize firing between neurons. Secondly, chemical synapses between KCs have been reported in the MB pedunculus in the locust. The reconstructions show that such KC>KC connections are also present in the lobes, where they are surprisingly prevalent. In fact, the most frequent outputs of the α/βs KCs are other α/βs KCs, assuming the morphologically defined KC>KC connections are functional synapses (Takemura, 2017).

A high percentage (55%) of these putative KC>KC synapses occur in rosette-like structures where multiple KCs also converge on a single dendritic process of an MBON . These are relatively unusual structures, not observed in EM reconstructions of the Drosophila visual system and, indeed, there is no direct evidence that they are functional synapses. At present it is only possible to speculate on their role. As points of heavy convergence, they might allow the effects of synapses from different KCs onto the same dendrite to act synergistically. Activity of a single KC may spread to its neighbors within the rosette, potentially generating a large compound synaptic release event onto the MBON in the middle. Such a signal amplification mechanism may be important to ensure that individual KCs can have a significant impact on MBON membrane potential by recruiting their rosette partners. How the specificity of learning could be maintained in this scenario is, however, unclear. Several basic questions will need to be answered before it is possible to begin to understand the functional significance of these rosettes. For example, can a single KC in the rosette indeed activate its neighbors? And how similar are the response properties of the different KCs that contribute to one rosette (Takemura, 2017)?

In conclusion, the connectivity of the KCs that carry olfactory and other sensory representations supports a model where parallel distributed memory processing occurs in each compartment. However, several circuit motifs that seem designed to spread and possibly amplify signals at the sites of KC output indicate that this circuit is likely more complicated than a simple feed-forward view of the system suggests (Takemura, 2017).

Dopamine-induced plasticity of the KC>MBON synapse is thought to be central to associative learning in this system. The reconstructions showed that dopaminergic neurons make well-defined synaptic contacts within the α lobe, with closely apposed post-synaptic membranes. This contrasts somewhat with dopaminergic innervation in the mammalian system, where there is typically not such close contact with a single clear post-synaptic partner, and volume transmission is the predominant model for dopamine release. It is not known whether the direct and indirect dopaminergic release sites have different functional consequences. Nevertheless, it seems likely that some type of volume transmission happens in the mushroom body. First, ~10 times more KC>MBON synapses than presynaptic sites of dopamine release were found in the α lobe, but previous work showed that learning-induced plasticity depresses MBON responses so strongly that most inputs are likely affected. Second, dopamine would need to diffuse only ~2 μm to reach every KC>MBON synapse within a compartment, but would also be sufficiently short range to prevent significant spill-over of dopamine to neighboring compartments, ensuring that the modularity of plasticity is maintained (Takemura, 2017).

Functional connectivity measurements showed that stimulating the DANs elicits large amplitude calcium signals from MBONs, similar to previous results. Intracellular recordings revealed that this was a surprisingly strong connection, sufficient to elicit spikes in the MBON. The response persisted when both spiking and nicotinic transmission was blocked, to limit the possibility that the DANs act through the KCs, which are cholinergic. Conversely, the response was strongly reduced by adding a dopamine receptor antagonist. Taken together, these results indicate that the response is likely a direct action of dopamine released by the DANs on the MBON, although a more complex mechanism or a role for the transmitter contained in the dense core vesicles observed in the DANs cannot be formally rule out. The depolarization exhibited markedly slow dynamics, peaking >2 s after stimulation offset, and then decaying over tens of seconds. Dopaminergic responses of similar amplitude and time course have been reported in both mammalian systems and in Aplysia, where it is mediated by cAMP-driven changes in a non-selective cation conductance (Takemura, 2017).

It is possible to induce memory formation in this circuit by pairing odor delivery with artificial activation of DANs. Targeting this optogenetic training procedure to DANs that innervate different compartments within the α lobe gives rise to memories with different valence, induction threshold and persistence. In the α1 compartment, a single pairing for 1 min induces an appetitive memory that lasts for 1 day. In contrast, optogenetic training focused on the α3 compartment requires multiple 1 min pairings, repeated at spaced intervals, and induces an aversive memory that lasts for 4 days. Although it seems likely that the different valences reflect the different projection sites of the MBONs for each of these compartments, where the differences in induction threshold and memory persistence might arise is less clear. There is no simple explanation for these differences from the EM-level circuit structure, as the basic wiring motifs were very similar in each compartment. Moreover, any explanation that invokes biochemical differences in KC>MBON synapses would require crisp spatial localization of the signaling pathway machinery that triggers plasticity, as exactly the same KCs participate in memory formation in different compartments. However, the observation that there are DAN>MBON synapses raises the possibility that biochemical differences in the MBONs might contribute to these differences in plasticity induction and maintenance. Indeed, RNAseq data from a set of four different MBONs showed expression of dopamine receptors. An alternative possibility, suggested by the findings in this study, is that the cotransmitter found in the dense core vesicles in the DANs is responsible for these differences. The size of these vesicles differs between DANs innervating the different compartments. Thus, these cells might release distinct co-transmitters, as has been observed in mammalian brain, which could trigger different signaling cascades in either the KCs or the MBONs to differentially modulate the induction and expression of plasticity across compartments (Takemura, 2017).

Models of MB function have generally considered the role for DANs to be confined to relaying signals about punishment or reward to the MB. However, in the mammalian brain, DANs can dynamically change their responses to both US and CS. In this study, it was found that the axonal terminals of the DANs receive many inputs from KCs within the lobes. In other words, both MBONs, DANs and even KCs receive extensive synaptic input from KCs in each compartment. If the current model that plasticity is pre-synaptic proves to be correct, this suggests that the responses of the DANs themselves would be subject to plasticity. If the synaptic depression observed at KC>MBON synapses also acts at KC>DAN connections, odor-evoked DAN responses would be diminished as a result of learning. This would serve as a negative feedback loop, reducing the strength of plasticity on successive training cycles with the same odor. Indeed, a gradually plateauing of the learning curve is a common feature of memory formation in different systems, including olfactory conditioning in Drosophila (Takemura, 2017).

One of the more surprising findings of this study was the observation that there are many direct DAN>MBON synaptic connections. Moreover, the functional connectivity measures indicate that these were relatively strong excitatory inputs. The excitatory sign of the DAN>MBON connection is also consistent with the behavioral effects of DAN activation that was observed. What role these DAN>MBON connections play in overall circuit function is an important question for future work. There are two general possibilities that are felt to be interesting to consider. Dopaminergic modulation has been proposed to play a general role in routing of information through the MB to different downstream neurons. Although changes in KC>MBON strength contribute to this process, the current results suggest that such state changes could also potentially be conveyed to the MBONs directly from the DANs. State-dependent changes in DAN activity have indeed been observed with calcium imaging. The slow synaptic dynamics observed in the DAN>MBON connection in MBON-α1 suggest the possibility that small changes in DAN firing might be capable of producing sustained changes in MBON membrane potential reflecting the current internal state of the animal (Takemura, 2017).

A second possibility, suggested from the framework of reinforcement learning established in vertebrates, is related to motivation and the comparison of expected versus actual reward. In Drosophila, prior work on odor-sugar conditioning in larvae provided evidence that flies form a comparison between the current state of reward and the reward expected from the conditioned cue. This work showed that animals behaviorally express memories only when the expected reward intensity is higher than the currently available reward. This is similar to the results presented in this study; just as the presence of reward diminished memory expression in the larvae, stimulating the DANs suppressed performance of animals trained by the optogenetic conditioning. The need to compare current and expected reward could potentially explain why there is an opponent relationship between the depression of KC>MBON synapses that drives associative learning and the excitatory effects of the DAN>MBON connection. If depression dominates, the association drives behavior, but this can be overridden by sufficient levels of DAN activity. In this respect, it is noteworthy that DANs appear to be able to act directly on the MBON, without participation of the KCs. Overall, this comparison could ensure that learned behavior is motivated not strictly by the expectation of reward, but rather the expected increase in reward, assessed at the moment of testing (Takemura, 2017).

The organization of the MB into a set of compartments arranged in series along the KC axons is well suited for simultaneously storing multiple independent memories of a given sensory stimulus. However, there must be some means by which these modules interact with one another to ensure coordinated, coherent expression of memory. Feedforward connections that link different compartments, first discovered by light microscopic anatomy, have recently been shown to be important for mediating such interactions. In particular, MBON-γ1pedc>α/β is an inhibitory neuron that connects aversive and appetitive learning compartments; it ensures that the circuit can readily toggle between different behavioral outputs (Takemura, 2017).

The EM reconstructions included both MBON-γ1pedc>α/β and MBON-β1>α, two feedforward neurons which project from their respective compartments to widely innervate other parts of the MB. Memories stored in the α lobe compartments are long-term and relatively inflexible, whereas the short-term memories formed in β1 and γ1pedc are readily updated by recent experiences. The feedforward connections are thought to enable the short-term memories in β1 and γ1pedc to temporarily mask expression of the stable memories stored in the α lobe. Indeed training an animal with either a multi-component aversive/appetitive food stimulus, or by simultaneous optogenetic activation of a composite set of DANs covering both appetitive and aversive compartments results in a compound memory that is initially aversive and later transitions to appetitive. The connectome results show that the primary synaptic targets of these feedforward neurons are the MBONs in the downstream compartment. By contrast, relatively few connections onto KCs were observed. Overall, this suggests that the feedforward connections can strongly influence the output from a compartment, but likely have little impact on the sensory information delivered to each compartment from the KCs. This is consistent with observations that MBON-γ1pedc>α/β strongly modulates activity of glutamatergic neurons at the tip of the horizontal lobe, but not their dendritic responses. Targeting these feedforward connections to the MBON may ensure that conflicting memories can form simultaneously in response to a complex sensory input, but with the behavioral manifestation of those memories capable of undergoing a crisp switch (Takemura, 2017).

This study has provided synapse level anatomical information on neuronal circuits involved in learning and memory in Drosophila. The comprehensive nature of this dataset should enable modeling studies not previously possible and suggests many experiments to explore the physiological and behavioral significance of the circuit motifs that were observed. That many of these motifs were not anticipated by over 30 years of extensive anatomical, experimental and theoretical studies on the role of the insect MB argues strongly for the value of electron microscopic connectomic studies (Takemura, 2017).

A dense (complete) reconstruction of neurons and synapses is resource intensive, so it is reasonable to ask if tracing a subset of cells or synapses could have yielded similar results with less effort. This is hard to answer in general, since there are many sparse tracing strategies, and each can be pursued to differing degrees of completeness. It is likely that most sparse tracing strategies would have discovered the new pathways reported in this, as the connections are numerous and connect well known cell types. Conversely, the conclusions that all cell types in this circuit had been identified would have been more difficult to make with confidence and a rare cell type, such as the SIFamide neuron, might have been missed. Perhaps, most importantly, statistical arguments, particularly those that require an accurate assessment of which cells are not connected, such as the absence of network structures such as rings or chains, would have been hard to make from sparse tracing. More generally, the model independent nature of dense tracing helps to discover any 'unknown unknowns', provides the strongest constraints on how neural circuits are constructed, and allows retrospective analysis of network properties not targeted during reconstruction (Takemura, 2017).

Dopaminergic neurons (DANs) signal punishment and reward during associative learning. In mammals, DANs show associative plasticity that correlates with the discrepancy between predicted and actual reinforcement (prediction error) during classical conditioning. Also in insects, such as Drosophila, DANs show associative plasticity that is, however, less understood. This study examined ssociative plasticity in DANs and their synaptic partners, the Kenyon cells (KCs) in the mushroom bodies (MBs), while training Drosophila to associate an odorant with a temporally separated electric shock (trace conditioning). In most MB compartments DANs strengthened their responses to the conditioned odorant relative to untrained animals. This response plasticity preserved the initial degree of similarity between the odorant- and the shock-induced spatial response patterns, which decreased in untrained animals. Contrary to DANs, KCs (α'/β'-type) decreased their responses to the conditioned odorant relative to untrained animals. No evidence was found for prediction error coding by DANs during conditioning. Rather, the data supports the hypothesis that DAN plasticity encodes conditioning-induced changes in the odorant's predictive power (Dylla, 2017).

Associative learning enables animals to anticipate negative or positive events. The neural mechanisms of associative learning are commonly studied in classical conditioning paradigms, in which animals are trained to associate a cue (conditioned stimulus; CS) with a punishment or reward (unconditioned stimulus; US). In the standard conditioning paradigm CS and US overlap in time, while in the trace conditioning paradigm there is a temporal gap between the CS and US. During both standard conditioning and trace conditioning, the US is mediated by dopaminergic neurons (DANs), in animals as diverse as monkeys and fruit flies (Dylla, 2017).

Genetic tools for monitoring and manipulating neuronal activity in the fruit fly Drosophila melanogaster promoted the understanding of the neural mechanisms of dopamine-mediated learning. Those mechanisms are well-described for standard 'odor-shock conditioning' in Drosophila, in which an olfactory CS is paired with a temporally overlapping electric shock US. During conditioning, an odor-shock association is formed in the mushroom body (MB) neuropil. The intrinsic neurons of the MB, the Kenyon cells (KCs), receive olfactory input in the MB-calyx and project to the vertical (α and α'), and the medial (β, β', and γ) MB-lobes. During odor-shock conditioning, the olfactory CS activates an odorant-specific KC population, and the electric shock US activates DANs that innervate the MB-lobes. In KCs, the CS-induced increase in intracellular calcium and the US-(dopamine)-induced second messengers synergistically activate an adenylyl cyclase, which alters the synaptic strength between KCs and MB output neurons (MBONs). This change in KC-to-MBON synapses is thought to encode the associative odor memory (Dylla, 2017).

The MB-lobes are divided into 15 compartments (α1-3, β1-2, α'1-3, β'1-2, and γ1-5), each of which is innervated by a distinct population of DANs and MBONs. These compartments constitute functional units, which are involved in different forms of associative learning. In compartments such as γ1, γ2, and β2, DANs mediate electric shock reinforcement. Besides mediating reinforcement during classical conditioning, Drosophila DANs are involved in long-term memory formation, forgetting, extinction learning and memory reconsolidation, and in integrating internal states with memory and sensory processing. A single DAN can even serve different functions, for example, PPL1-γ1pedc (also referred to as MB-MP1) signals reinforcement, and controls state-dependent memory retrieval (Dylla, 2017).

The functional complexity of Drosophila DANs is further increased by the fact that DANs show learning-induced associative plasticity: they increase their response to the CS during classical conditioning. Mammalian DANs also increase their CS-induced responses during classical conditioning. In addition, they decrease their response to the US, and when a predicted US does not occur, they decrease their activity below baseline level. This pattern of response plasticity in mammalian DANs is compatible with the hypothesis that animals only learn to associate a CS with a US, when the US occurs unpredictably. Thus, mammalian DANs appear to encode this prediction error. In Drosophila, however, DANs do not change their response to the US. Therefore, Drosophila DANs appear to encode the US prediction by the CS rather than encoding the US prediction error during classical conditioning. It is not clear, whether classical conditioning in insects is driven by US prediction error. There is evidence for prediction error-driven conditioning in crickets, but there is also a controversy about whether or not blocking (a failure to learn, when the US is already predicted by another CS) occurs (Dylla, 2017).

This study reassessed the hypothesis that Drosophila DANs do not encode the prediction error during classical conditioning. Different from Riemensperger (2005) who pooled DAN activity across the mushroom body lobes, this study differentiated between DAN types that innervate different compartments of the MB lobes. Moreover, instead of using standard conditioning, trace conditioning with a 5 s gap between the CS and the US was used, allowing for precise distinguishing between responses to either the CS or the US (Dylla, 2017).

This study investigated associative plasticity in the responses of DANs and their synaptic partners, the KCs, across the compartments of the Drosophila MB. Using calcium imaging, CS- and US-induced responses of a subpopulation of DANs (labeled by TH-GAL4) and of KCs (labeled by OK107-GAL4) were recorded during odor-shock trace conditioning. Note, that most compartments are innervated by multiple TH-GAL4-labeled DANs. Therefore, the average activity that was recorded in most of the compartments might mask possible differences in the response properties and plasticity between individual DANs and KCs. Only DAN responses in the compartments γ2 and α'1 reflect the responses of a single neuron (Dylla, 2017).

Across MB compartments, DANs and KCs differed in their response strength to odorants and electric shock, and they differed in CS-US pairing-induced plasticity. Compared to the unpaired control groups, KCs decreased their responses to the CS in all compartments of the β'-lobe and in the junction, while DANs increased their responses to the CS in all compartments of the γ- and β'-lobe, and in the junction. The occurrence of associative plasticity in DANs in the compartments γ3-5 and β'1 is surprising, given that these DANs are not known to be involved in odor-shock conditioning, after training there was neither an associative change in US-induced DAN responses nor a change of activity during US-omission after CS presentation. It is therefore concluded that Drosophila DANs do not encode the US-prediction error during classical conditioning (Dylla, 2017).

Previous studies suggested that DANs in the MB lobes respond strongly to electric shock and weakly to odorants. The compartment-resolved analysis of the calcium imaging data refines this picture: It is confirmed that DANs of all imaged compartments respond to both electric shock and odorants, and it was shown that their relative response strength to odorants and electric shock differs across compartments. For example, DANs innervating γ1 responded stronger to electric shock than to odorants, while DANs innervating β'2 responded equally strong to odorants and electric shock. The strongest DAN responses to electric shock were shown to be in the compartments γ1 and γ2. These compartments receive input from PPL1-γ1pedc and PPL1-γ2α'1 DANs that mediate electric shock reinforcement. In all compartments, except in α1/α'1, the DAN response strength correlated positively with the current strength encountered by individual flies. Thus, DANs are capable of encoding the strength of the electric shock US, and this property may account for the positive dependence between electric shock strength and learning performance in flies (Dylla, 2017).

Calcium responses in KCs differ between MB lobes, and they differ between the compartments of a given lobe, possibly due to compartment-specific modulation by DANs and MBONs. KCs in γ2 and γ3 responded strongest to odorants, confirming the results of Cohn (2015). KCs generally responded only weakly to electric shocks. Previously published strong KC responses to electric shock may be because electric shocks were applied to the flies' abdomen rather than to their legs, which might have resulted in a stronger stimulation (Dylla, 2017).

The associative strengthening of DAN responses to the olfactory CS (as compared to the unpaired control group), confirms the previous report by Riemensperger (2005). Associative plasticity occurred in those DANs that innervate the MBs (PPL1 and PAM cluster DANs; note that the used TH-GAL4 driver line covers only a small subpopulation of PAM neurons but not in DANs that innervate the central complex (PPL1 and PPM3 cluster DANs). This is in line with the established role of MB innervating-DANs in associative memory formation, while central complex-innervating DANs are involved in behaviors such as locomotion, wakefulness, arousal, and aggression, and are therefore not expected to show odor-shock conditioning-induced plasticity (Dylla, 2017).

In contrast to previous studies, this study did not find an associative increase in KC calcium responses to the CS in the MB-lobes after odor-shock conditioning. This may indicate either a difference between trace conditioning and standard conditioning, or a difference in other experimental parameters that may also account for inconsistencies in the published effects of odor-shock conditioning (Dylla, 2017).

The associative decrease in KC responses in the β'-lobe compartments is in line with previous studies that showed conditioning-induced depression of KC-to-MBON synapses (Cohn, 2015; Hige, 2015). Therefore, it propose that the associative decrease in KC responses to the CS reflects a presynaptic depression at KC-to-MBON synapses in β'-lobe compartments (Dylla, 2017).

What is the site of neuronal plasticity that underlies the relative increase in DANs' responses to the olfactory CS? Riemensperger (2005) proposed that DANs get odorant-driven excitatory input via a MBON feedback loop that is strengthened during odor-shock conditioning. However, the DAN population is composed of different neuron types that do not share a common input either from MBONs or from other neurons that could explain the global associative plasticity across MB compartments. Because KCs presumably provide the only common odor-driven input to all MB-innervating DANs, it is suggested that the site of associative plasticity is located in a KC-to-DAN synapse. Indeed, KC-to-DAN synapses have recently been reported in Drosophila (Cervantes-Sandoval, 2017). Associative increase in CS-induced DAN responses occurred despite unaltered or decreased KC responses in the same compartment. This suggests that the associative plasticity occurs post-synaptic in DANs and is not inherited from KCs (Dylla, 2017).

What is the neuronal substrate of CS-US coincidence detection in DANs and KCs? Drosophila trace conditioning depends on dopamine receptor-triggered signaling in KCs, as is the case for standard conditioning. However, the CS-US coincidence detection mechanism in trace conditioning is unknown. In standard conditioning the CS-induced increase in KCs' calcium concentration coincides with the US-(dopamine)-induced second messengers, which is thought to synergistically activate the rutabaga adenylyl cyclase, and ultimately alters the strength of KC-to-MBON synapses. This mechanism would not work for trace conditioning, because (1) at the time the US occurs, CS-induced increase in KCs' calcium concentration is back to baseline levels, and (2) trace conditioning does not involve the rutabaga adenylyl cyclase. It is therefore hypothesized that a non-rutabaga adenylyl cyclase or a protein kinase C could serve as a molecular coincidence detector for the CS trace and the US. For example, the CS-induced calcium and dopamine signaling could lead to a sustained activation of an adenylyl cyclase or protein kinase C in KCs, which then would increase synergistically and drive synaptic plasticity during the US-induced dopamine signaling (Dylla, 2017).

DAN responses to odorants and associative strengthening of DAN responses to the CS-odorant are not included in current models of associative learning in the MB. However, associative plasticity is a common feature of US-mediating neurons, which occurs in mammalian and Drosophila DANs, and in an octopaminergic neuron in honey bees (Dylla, 2017).

What could be the function of odorant-induced responses and odor-shock conditioning-induced plasticity in DANs? MB-innervating DANs strengthened their response to the CS (as compared to the unpaired group) during odor-shock conditioning, in line with Riemensperger (2005). However, other than in monkey DANs, this study did not observe associative plasticity in DANs' response to the US. The data therefore support the idea that Drosophila DANs encode predictive power of the CS, e.g., US-prediction, but not the US-prediction error during classical conditioning (Dylla, 2017).

This study found shock-induced responses and associative plasticity in DANs that are not involved in odor-shock conditioning, for example in DANs innervating β'1, γ3, γ4, and γ5. This suggests that those DANs serve a function in aversive odor learning which is not captured by the commonly applied conditioning paradigms. For example, the relative strengthening of CS-induced responses could mediate reinforcement during second-order conditioning, in which a previously reinforced CS1 can act as US in subsequent conditioning of a second CS2. As Drosophila is capable of second-order learning, this theory can be tested in behavioral experiments: if associative strengthening of DAN responses to the CS underlies CS1-induced reinforcement in second-order conditioning, then preventing associative plasticity in DANs, or blocking their output during CS2-CS1 pairing should abolish second-order conditioning (Dylla, 2017).

The occurrence of CS-induced responses and associative plasticity in most of the MB-innervating DANs suggests that the separation between the CS- and US-pathway and between different US-pathways is less strict than suggested in current models of associative learning in the MB. Associative plasticity in the spatial pattern of CS-induced DAN responses makes them a potential neuronal substrate for encoding the US identity in CS-US memories and the predictive power of a CS (Dylla, 2017).

These data revealed similar response properties and plasticity rules across Drosophila DANs in the γ- and β'-lobe. This contrasts with their anatomical and functional heterogeneity, which indicates yet undiscovered mechanisms and functions of DAN plasticity. Note, that this study could not test whether the flies learned in the imaging setup, as currently no behavioral readout exists for odor-shock conditioning during physiological experiments. Nevertheless, since a conditioning protocol and stimulus application comparable to an established behavioral paradigm was used, it is believed that the associative plasticity in neuronal responses that was found underlies behavioral associative plasticity. Therewith the data lay the foundations for causal studies on the function of associative plasticity in DANs (Dylla, 2017).

In Drosophila, negatively reinforcing dopaminergic neurons also provide the inhibitory control of satiety over appetitive memory expression. This study shows that aversive learning causes a persistent depression of the conditioned odor drive to two downstream feed-forward inhibitory GABAergic interneurons of the mushroom body, called MVP2, or mushroom body output neuron (MBON)-γ1pedc>α/β. However, MVP2 neuron output is only essential for expression of short-term aversive memory. Stimulating MVP2 neurons preferentially inhibits the odor-evoked activity of avoidance-directing MBONs and odor-driven avoidance behavior, whereas their inhibition enhances odor avoidance. In contrast, odor-evoked activity of MVP2 neurons is elevated in hungry flies, and their feed-forward inhibition is required for expression of appetitive memory at all times. Moreover, imposing MVP2 activity promotes inappropriate appetitive memory expression in food-satiated flies. Aversive learning and appetitive motivation therefore toggle alternate modes of a common feed-forward inhibitory MVP2 pathway to promote conditioned odor avoidance or approach (Perisse, 2016).

Prior work in Drosophila indicated that negative reinforcement and hunger-state-dependent motivational control of appetitive memory performance might be controlled by the same dopaminergic neurons (DANs). The presynaptic field of the MP1/PPL1-γ1pedc DANs occupies a defined region of the MB that also contains the MVP2/MBON-γ1pedc >αβ dendrites, suggesting that these DANs modulate the efficacy of this specific KC-MBON connection. The results of the current study demonstrate that the MVP2 MBONs also play a critical role in the expression of short-term aversive memory and the state-dependence of appetitive memory expression. Since these findings directly mirror the described roles for the MP1 DANs (Krashes, 2009), it is concluded that DAN modulation of the KC-MVP2 junction is critical for both negative reinforcement during olfactory learning and the motivational salience of appetitive odor cues (Perisse, 2016).

The GABA-ergic MVP2 neurons have postsynaptic and presynaptic processes in the MB, suggesting that they are interneurons of the MB and feed-forward inhibit other MBON compartments. Dendrites of MVP2 neurons (and the presynaptic terminals of the MP1 DANs) innervate the γ1 region and more densely innervate the αβs than the αβ core (αβc) region of the αβ ensemble (Krashes, 2009). MVP2 are therefore likely to be primarily driven by αβs KCs. Since αβs neurons contribute to conditioned approach and avoidance, whereas αβc are particularly important for conditioned approach (Perisse, 2013), there is an imbalance in the drive to approach and avoidance behaviors at this level of the MBON network (Perisse, 2016).

Artificial activation of MVP2 neurons in naive flies drives approach behavior, consistent with them preferentially inhibiting MBON compartments that direct avoidance -- as opposed to those that drive approach. Anatomical and functional connectivity and odor-directed behavioral data are consistent with such a model. MVP2 stimulation inhibits odor-evoked activity in M4/6 but not in V2αV2α' MBONs. MVP2 stimulation also promotes expression of approach memory in food-satiated flies, yet it inhibits naive odor avoidance behavior. It is concluded that MVP2 directly inhibit the M4/6 class of horizontal lobe MBONs through synapses made on the primary axonal segment as it exits the MB lobes. Inhibition exerted in this area might be expected to control the gain of the MBON responses following integration of KC inputs in the MBON dendrite in a manner similar to perisomatic inhibition in mammals. Consistent with this anatomy and idea, no obvious changes were found in the odor drive to the dendritic region of M4/6 neurons between hungry and satiated flies, but a hunger-dependent decrease was apparent when odor-evoked responses were measured in the efferent neurites. In contrast, MVP2 neurons do not functionally inhibit or densely innervate the neurites of V2αV2α' MBONs, nor does hunger reduce odor-evoked responses in V2αV2α' MBONs. It therefore seems likely that MVP2 neurons contact DANs or other neurons that occupy the α2 compartment of the MB lobe (Perisse, 2016).

The data also demonstrate that aversive learning reduces the relative conditioned odor drive to MVP2 neurons, which would presumably decrease feed-forward inhibition onto the relevant MBON compartments and thereby render them more responsive to odors. Output from the glutamatergic M4/6 neurons, which are postsynaptic to the KCs in the horizontal tip regions, is required for expression of aversive and appetitive memory. Furthermore, the relative odor-drive to M4/6 neurons was shown to be depressed by reward learning and potentiated by aversive learning. Since aversive learning reduces the conditioned odor drive of the MVP2 neuron, it is proposed that the observed increase in odor-drive to M4/6 after aversive learning results from reduced feed-forward inhibition from MVP2. This would mean that bi-directional output plasticity could emerge via a direct junctional plasticity following reward conditioning, but a network property of reduced MVP2 feed-forward inhibition after aversive conditioning. Such a layered feed-forward network architecture linking one site of DAN-driven KC-MBON plasticity to another KC-MBON connection would provide a means to achieve odor-specific bi-directional plasticity at a particular synaptic junction using dopamine-driven synaptic depression in two different places. It is proposed that this circuit design principle in which plasticity at one site of a neuron can, via feed-forward inhibition, indirectly alter the efficacy of output elsewhere in the same neuron, could be a general feature in the brain of the fly and other animals. It is possible that the KC-MVP2 junction also exhibits bi-directional plasticity, notably with inverted polarity relative to M4/6 plasticity traces (Perisse, 2016).

The layered network architecture places the aversive memory relevant MVP2 plasticity on top of the M4/6 plasticity that is relevant for appetitive memory (Owald, 2015). This organization could accommodate the co-existence of aversive and appetitive olfactory memories following conditioning reinforced by sugar laced with bitter taste. Immediately after such training flies avoid the conditioned odor because the aversive taste memory relieves feed-forward inhibition onto the sites that are depressed by appetitive sugar plasticity and therefore over-rides the expression of approach memory. However, as the aversive memory decays, feed-forward inhibition returns and appetitive memory is then expressed. A similar mechanism might account for the time-dependent switch from conditioned aversion to approach following odor conditioning reinforced by alcohol (Kau, 2011). It is notable that learning-induced plasticity of relative odor-drive to MVP2 persists for at least 3 hr after training whereas output from MVP2 is dispensable for the expression of aversive memory at that time. Since expression of different phases of aversive memory requires distinct combinations of MBON pathways (Bouzaiane, 2015), it is proposed that more persistent MVP2 plasticity might provide a permissive gate for both the formation of aversive memory in, and the expression from, other parts of the MBON network. This would be reminiscent of fear conditioning in the neural circuitry of the mouse amygdala, where dopamine suppresses feed-forward GABA-ergic inhibition from local interneurons to facilitate the induction of long-term potentiation (Perisse, 2016).

MVP2 neuron output is required for the expression of sugar-reinforced approach memory at all times. Moreover, odors evoked larger MVP2 responses in hungry than in food-satiated flies, and elevating MVP2 activity in satiated flies promoted inappropriate expression of appetitive memory. These results are consistent with the model that hunger generally increases feed-forward inhibition through MVP2 to support appetitive memory expression). This result is also the mirror-image of that with MP1 DANs whose activity increases when the flies are satiated and whose inhibition leads to the expression of appetitive memory in satiated flies. Taken with prior work, it is therefore proposed that hunger increases dNPF, which releases MP1 inhibition over the KC-MVP2 connection. This results in an increase of odor-evoked MVP2 feed-forward inhibition onto the MBON compartments such as M4/6 that contain the KC-MBON synapses that are directly modified by appetitive conditioning. The increase of MVP2 inhibition into these, and other, compartments allows more efficient expression of the appetitive memory-directed approach behavior by effectively raising the motivational salience of learned food-related odors. Appetitive conditioning may also increase odor-specific recruitment of MVP2 feed-forward inhibition (Perisse, 2016).

The current findings therefore suggest that the MVP2 neuron pathway functions in at least three modes that are presumably selected by the aversively reinforcing MP1 DANs. If the flies are aversively conditioned, phasic MP1 specifically depresses conditioned-odor drive to MVP2 neurons. In a food-satiated fly, tonic MP1 limits general odor-driven MVP2 activity. Lastly, in the hungry fly, lower MP1 activity generally enhances odor-drive to MVP2. In the OFF modes, low-level MVP2 feed-forward inhibition skews the MBON network toward behavioral avoidance, whereas in the ON mode the increased feed-forward inhibition from MVP2 skews the MBON network toward favoring conditioned approach. The MP1 DANs signal the aversive reinforcing properties of electric shock, heat, and bitter taste, suggesting they provide general aversive influence. The satiated state presumably uses a tonic version of this aversive signal to limit the fly approaching an appetitive odor cue (Perisse, 2016).

The parallels between the fly and mammalian dopaminergic systems appear striking. DANs in the basal ganglia of the mammalian brain also support reinforcement learning and the prediction of stimuli that potentially lead to rewarding outcomes. Furthermore, like the fly DANs, mammalian DANs can be anatomically divided into those that generate aversion and different types of reward. GABA-ergic neurons in the mouse ventral tegmental area, whose cell bodies are interspersed with the DANs, have been proposed to signal the value of expected reward and provide a source of subtraction to DANs that calculate a reward prediction error. Negatively reinforcing MP1 DANs in the fly modulate odor-drive to the MVP2 neurons to provide the motivational control over actions to gain reward. Therefore, MVP2 neurons may provide an inhibitory bridge between MBON domains that are controlled by aversive and rewarding DANs (Perisse, 2016).

Massive activation of dopamine neurons is critical for natural reward and drug abuse. In contrast, the significance of their spontaneous activity remains elusive. In Drosophila melanogaster, depolarization of the protocerebral anterior medial (PAM) cluster dopamine neurons en masse signals reward to the mushroom body (MB) and drives appetitive memory. Focusing on the functional heterogeneity of PAM cluster neurons, a single class of PAM neurons, PAM-γ3, mediates sugar reward by suppressing their own activity. PAM-γ3 is selectively required for appetitive olfactory learning, while activation of these neurons in turn induces aversive memory. Ongoing activity of PAM-γ3 gets suppressed upon sugar ingestion. Strikingly, transient inactivation of basal PAM-γ3 activity can substitute for reward and induces appetitive memory. Furthermore, the satiety-signaling neuropeptide Allatostatin A (AstA) was identified as a key mediator that conveys inhibitory input onto PAM-γ3. These results suggest the significance of basal dopamine release in reward signaling and reveal a circuit mechanism for negative regulation (Yamagata, 2016).

Sugar ingestion triggers multiple reward signals in the fly brain. This study has provided lines of evidence that part of the reward is signaled by inactivating dopamine neurons. The role of PAM-γ3 highlights the striking functional heterogeneity of PAM cluster dopamine neurons. The decrease and increase of dopamine can convey reward to the adjacent compartments of the same MB lobe-γ3 and γ4-. The reward signal by the transient decrease of dopamine is in stark contrast to the widely acknowledged role of dopamine. Midbrain dopamine neurons in mammals were shown to be suppressed upon the presentation of aversive stimuli or the omission of an expected reward, implying valence coding by the bidirectional activity. As depolarization of PAM-γ3 can signal aversive reinforcement, these neurons convey the opposite modulatory signals to the specific MB domain by the sign of their activity. Intriguingly, the presentation and cessation of electric shock act as punishment and reward, respectively. Such bidirectional activity of PAM-γ3 may represent the presentation and omission of reward. (Yamagata, 2016).

While thermoactivation of PAM-γ3 induced robust aversive memory, blocking their synaptic transmission did not affect shock learning, leaving a question regarding their role in endogenous aversive memory process. PAM-γ3 may only be involved in processing aversive reinforcement different from electric shock-like heat. However, two studies show that dopamine neurons mediating aversive reinforcement of high temperature and bitter N,N-Diethyl-3-methylbenzamide (DEET) are part of those for electric shock. Identification of such aversive stimuli that are signaled by PAM-γ3 activation is certainly interesting, as it is perceived as the opposite of sugar reward and thus provides the whole picture of the valence spectrum. Another scenario where sufficiency and necessity do not match is the compensation of the reinforcing effect by other dopamine cell types (e.g. MB-M3). The lack of PAM-γ3 requirements for electric shock memory may be explained by a similar mechanism. (Yamagata, 2016).

How can the suppression of PAM-γ3 modulate the downstream cell and drive appetitive memory? Optogenetic activation of the MB output neurons from the γ3 compartment induces approach behavior. This suggests that the suppression of the PAM-γ3 neurons upon reward leads to local potentiation of Kenyon cell output. This model is supported by recent studies showing the depression of MB output synapses during associative learning. A likely molecular mechanism is the de-repression of inhibitory D2-like dopamine receptors, DD2R. As D2R signaling is a widely conserved mechanism, it may be one of the most ancestral modes of neuromodulation. (Yamagata, 2016).

Furthermore, recent anatomical and physiological studies demonstrated that different MB-projecting dopamine neurons are connected to each other and act in coordination to respond to sugar or shock. Therefore, memories induced by activation or inhibition of PAM-γ3 may well involve the activity of other dopamine cell types (Yamagata, 2016).

The finding that appetitive reinforcement is encoded by both activation and suppression of dopamine neurons raises the question as to the complexity of reward processing circuits (see Reward signals by excitation and inhibition of dopamine neurons). It is, however, reasonable to implement a component like PAM-γ3 as a target of the satiety-signaling inhibitory neuropeptide AstA. Intriguingly, the visualization of AstA receptor distribution by DAR-1-GAL4 revealed expression in two types of MB-projecting dopamine neurons: PAM-γ3 and MB-MV1 (also named as PPL1- γ2α'1). Given the roles of MB-MV1 in aversive reinforcement and locomotion arrest, AstA/DAR-1 signaling may also inhibit a punishment pathway upon feeding. It is thus speculated that this complex dopamine reward circuit may be configured to make use of bidirectional appetitive signals in the brain (Yamagata, 2016).

Dopaminergic neurons provide reward learning signals in mammals and insects. Recent work in Drosophila has demonstrated that water-reinforcing dopaminergic neurons are different to those for nutritious sugars. This study tested whether the sweet taste and nutrient properties of sugar reinforcement further subdivide the fly reward system. They found that dopaminergic neurons expressing the OAMB octopamine receptor specifically conveyed the short-term reinforcing effects of sweet taste. These dopaminergic neurons projected to the β'2 and γ4 regions of the mushroom body lobes. In contrast, nutrient-dependent long-term memory required different dopaminergic neurons that project to the γ5b regions, and it could be artificially reinforced by those projecting to the β lobe and adjacent α1 region. Surprisingly, whereas artificial implantation and expression of short-term memory occurred in satiated flies, formation and expression of artificial long-term memory required flies to be hungry. These studies suggest that short-term and long-term sugar memories have different physiological constraints. They also demonstrate further functional heterogeneity within the rewarding dopaminergic neuron population (Huetteroth, 2015).

These results demonstrate that the sweet taste and nutrient properties of sugars are independently processed and reinforce memories of different duration. Sweet taste is transduced through octopaminergic neurons whose released octopamine, via the OAMB receptor, activates dopaminergic neurons that project to the β'2am and γ4 regions of the mushroom body. Octopaminergic reinforcement also modulates the state dependence of STM via the OCTβ2R receptor that is required in the dopaminergic MB-MP1 neurons (Huetteroth, 2015).

Nutrient-dependent LTM does not involve octopamine or sweet-taste-reinforcing dopaminergic neurons. Nutrient reinforcement instead requires dopaminergic neurons innervating γ5b of the mushroom body, whereas those going to β1, β2, and the adjacent α1 region are sufficient. More work will be required to understand this distributed process, which apparently has an immediate and delayed dynamic (Huetteroth, 2015).

Whereas formation and expression of sweet-taste-reinforced STM is insensitive to satiety state, artificial formation and expression of nutrient-relevant memory require flies to be hungry. Even direct stimulation of the relevant rewarding dopaminergic neurons cannot implant appetitive LTM in food-satiated flies. These experiments suggest that hunger establishes an internal state that permits the nutrient-reinforcing signals to be effective. It will be interesting to understand what the permissive state involves and where it is required. Others have previously described a role for CREB-regulated transcription co-activator 1 (CRTC) in enabling hunger-dependent LTM in the fly and promoting persistent memory in the mouse. It therefore seems plausible that such a mechanism might be required in the mushroom body neurons to permit nutrient-dependent reinforcement (Huetteroth, 2015).

The detection of environmental temperature and regulation of body temperature are integral determinants of behaviour for all animals. These functions become less efficient in aged animals, particularly during exposure to cold environments, yet the cellular and molecular mechanisms are not well understood. This study identifies an age-related change in the temperature preference of adult fruit flies that results from a shift in the relative contributions of two parallel mushroom body (MB) circuits-the β'- and β-systems. The β'-circuit primarily controls cold avoidance through dopamine signalling in young flies, whereas the β-circuit increasingly contributes to cold avoidance as adult flies age. Elevating dopamine levels in β'-afferent neurons of aged flies restores cold sensitivity, suggesting that the alteration of cold avoidance behaviour with ageing is functionally reversible. These results provide a framework for investigating how molecules and individual neural circuits modulate homeostatic alterations during the course of senescence (Shih, 2015).

Learned and adaptive behaviors rely on neural circuits that flexibly couple the same sensory input to alternative output pathways. This study shows that the Drosophila mushroom body functions like a switchboard in which neuromodulation reroutes the same odor signal to different behavioral circuits, depending on the state and experience of the fly (see Compartmentalized Architecture of the Mushroom Body). Using functional synaptic imaging and electrophysiology, it was shown that dopaminergic inputs to the mushroom body modulate synaptic transmission with exquisite spatial specificity, allowing individual neurons to differentially convey olfactory signals to each of their postsynaptic targets. Moreover, the dopaminergic neurons function as an interconnected network, encoding information about both an animal's external context and internal state to coordinate synaptic plasticity throughout the mushroom body. These data suggest a general circuit mechanism for behavioral flexibility in which neuromodulatory networks act with synaptic precision to transform a single sensory input into different patterns of output activity (Cohn, 2015).

This study took advantage of the mushroom body's orderly architecture to gain insight into the circuit mechanisms through which neuromodulation mediates flexible sensory processing. Compartmentalized dopaminergic signaling permits independent tuning of synaptic transmission between an individual KC and its repertoire of postsynaptic MBON targets. As a consequence, the same KC odor representation can evoke different patterns of output activity, depending on the state of the animal and the dopaminergic network. Recent data indicate that the ensemble of MBONs acts in concert to bias an animal's behavioral response to an odor such that altering the balance of their activity can modify the olfactory preferences of both naive and trained animals. In accord with such a model, this study revealed how a distributed neuromodulatory network is poised to orchestrate plasticity across all 15 compartments of the mushroom body and reweight the net output of the MBONs, allowing for adaptive behavioral responses based on the immediate needs or past experience of the animal (Cohn, 2015).

Distinct subsets of DANs are sufficient to drive learned olfactory associations, leading to the suggestion they may act autonomously to encode the rewarding or punishing contextual stimuli that assign meaning to an odor. The current data, however, suggest a more complex circuit architecture, in which rich functional interconnectivity between compartments contributes to coordinated and bidirectional patterns of activity across the DAN population. This raises the possibility that reinforcement experiences may be represented by combinatorial patterns of DAN excitation and inhibition in different compartments, endowing the dopaminergic population with a greater capacity to instruct behavior via the limited repertoire of mushroom body outputs. Intriguingly, midbrain dopaminergic neurons responsive to punishment and reward also project to distinct targets in the mammalian brain and display a similar functional opponency as a consequence of reciprocal network interactions. Thus, the concerted and partially antagonistic action of neuromodulatory pathways may represent a general and conserved circuit principle for generating adaptive behavioral responses (Cohn, 2015).

Distinct DAN network activity states are evoked by electric shock and sugar ingestion, reinforcers classically used in associative olfactory conditioning paradigms because of their strong inherent valence. However, similarly distributed patterns of DAN activity are correlated with the fly's motor activity, implying that an animal's behavioral state might serve as a reinforcement stimulus that itself drives synaptic plasticity to shape odor processing. Metabolic states, such as thirst and hunger, have been shown to gate appetitive reinforcement by water and sugar rewards, permitting state-dependent formation of olfactory associations only in motivated animals. The current data highlight an additional facet of how an animal's internal state can regulate dopamine release to adjust the salience of contextual cues. Together, these observations indicate that the distributed DAN network integrates information about external context and internal state with MBON feedback to represent the moment-by-moment experience of an animal and dynamically regulate the flow of olfactory signals through the mushroom body (Cohn, 2015).

The independent regulation of synapses along an axon is thought to permit a single neuron to convey specialized information to different downstream targets, providing additional flexibility and computational power to neural circuits. In the mushroom body, synapse-specific plasticity is achieved through spatially restricted patterns of dopaminergic modulation that divide a KC axon into functionally distinct segments. Thus, the ensemble of synapses within a compartment, as the site of convergence for sensory and contextual signals, represents the elementary functional unit that underlies experience-dependent mushroom body output (Cohn, 2015).

Within a compartment, multiple neuromodulatory mechanisms appear to shape synaptic signaling. Broad potentiation of KC-MBON synapses is seen after DAN activation, but odor-specific depression is seen if DANs were coincidently activated with KCs, consistent with the synaptic changes previously proposed to occur after learning. Taken together, these findings indicate that neuromodulation in the mushroom body instructs opposing forms of synaptic plasticity, analogous to the bidirectional tuning of synaptic strength by dopamine in mammalian brain centers. The molecular mechanisms through which dopamine can direct diverse synaptic changes within a compartment remain to be elucidated, but they may depend on signaling through different dopamine receptors or downstream signaling cascades that function as coincidence detectors. Indeed, while DopR1 in KCs is essential to the formation of learned olfactory associations, this receptor was found to play only a subtle role in the context-dependent patterning of Ca2+ along their axons. Conversely, DopR2 strongly influences the topography of presynaptic Ca2+ along KC axons, in accord with evidence that tonic release of dopamine during ongoing behavior acts through this receptor to interfere with the maintenance of specific learned olfactory associations. Thus, distinct molecular pathways may transform the same dopaminergic reinforcement signals into synaptic changes of opposite polarity to shape olfactory processing based on both the present context and prior experiences of an individual (Cohn, 2015).

The mushroom body has been most extensively studied as a site for associative learning in which the temporal pairing of an odor with a reinforcement experience selectively alters subsequent behavioral responses to that odor. The current data suggest that the convergence of DAN network activity and KC olfactory representations within the mushroom body lobes may drive associative plasticity in each compartment, allowing the odor tuning of the MBON repertoire to reflect the unique experiences of an individual. However, these observations also provide insight into the mushroom body's broader role in the context-dependent regulation of innate behaviors. The ongoing activity of the distributed DAN network, encoding information about an animal's current environmental context and behavioral state, is poised to continuously reconfigure the activity patterns of the MBON population to allow for adaptive responses based on the acute needs of the animal. This context-dependent synaptic modulation could potentially erode odor-specific learned associations within the mushroom body, permitting the immediate circumstances of an animal to dominate over previously learned olfactory associations that may no longer be predictive or relevant. The axons of MBONs ultimately converge with output pathways from the lateral horn, a Drosophila brain center thought to mediate stereotyped responses to odors, providing a potential substrate for learned and context-dependent output from the mushroom body to influence inherent olfactory preferences (Cohn, 2015).

Thus, the dual role of neuromodulation in the mushroom body-to select among alternative circuit states that regulate both innate and learned behaviors-is reminiscent of its function in other higher integrative brain centers. In the basal ganglia, for example, different temporal patterns of dopamine release are thought to select the relevant circuit configurations that control inherently motivated behaviors as well as reinforcement learning. The generation of flexible behavioral responses based on experience, whether past or present, may therefore rely on common integrative brain structures in which neuromodulatory networks act with exquisite spatial precision to shape sensory processing (Cohn, 2015).

Go to top

Animals discriminate stimuli, learn their predictive value and use this knowledge to modify their behavior. In Drosophila, the mushroom body (MB) plays a key role in these processes. Sensory stimuli are sparsely represented by approximately 2000 Kenyon cells, which converge onto 34 output neurons (MBONs) of 21 types (see Circuit diagrams of the mushroom body). The role of MBONs were studied in several associative learning tasks and in sleep regulation, revealing the extent to which information flow is segregated into distinct channels and suggesting possible roles for the multi-layered MBON network. Optogenetic activation of MBONs can, depending on cell type, induce repulsion or attraction in flies. The behavioral effects of MBON perturbation are combinatorial, suggesting that the MBON ensemble collectively provides an adaptive mechanism to assign valence-positive or negative survival value-to a sensory stimulus then store that information, and recall it when that same stimulus is encountered again. It is proposed that local, stimulus-specific dopaminergic modulation selectively alters the balance within the MBON network for those stimuli. These results suggest that valence encoded by the MBON ensemble biases memory-based action selection (Aso, 2014: PubMed).

To survive in a dynamic environment, an animal must discover and remember the outcomes associated with the stimuli it encounters. It then needs to choose adaptive behaviors, such as approaching cues that predict food and avoiding cues that predict danger. The neural computations involved in using such memory-based valuation of sensory cues to guide action selection require at least three processes: (1) sensory processing to represent the identity of environmental stimuli and distinguish among them; (2) an adaptive mechanism to assign valence-positive or negative survival value-to a sensory stimulus, store that information, and recall it when that same stimulus is encountered again; and (3) decision mechanisms that receive and integrate information about the valence of learned stimuli and then bias behavioral output. To understand such decision-making processes, one approach is to locate the sites of synaptic plasticity underlying memory formation, identify the postsynaptic neurons that transmit stored information to the downstream circuit and discover how their altered activities bias behavior (Aso, 2014).

The mushroom body (MB) is the main center of associative memory in insect brains. While the MB processes several modalities of sensory information and regulates locomotion and sleep, MB function has been most extensively studied in the context of olfactory memory-specifically, associating olfactory stimuli with environmental conditions in order to guide behavior. In Drosophila, olfactory information is delivered to the MB by projection neurons from each of ~50 antennal lobe glomeruli. Connections between the projection neurons and the ~2000 Kenyon cells (KCs), the neurons whose parallel axonal fibers form the MB lobes, are not stereotyped; that is, individual flies show distinct wiring patterns between projection neurons and KCs. Sparse activity of the KCs represents the identity of odors. The output of the MB is conveyed to the rest of the brain by a remarkably small number of neurons-34 cells of 21 cell types per brain hemisphere (Aso, 2014).

The information flow from the KCs to the MB output neurons (MBONs) has been proposed to transform the representation of odor identity to more abstract information, such as the valence of an odor based on prior experience (see discussion in Aso, 2014a). In contrast to KCs, MBONs have broadly tuned odor responses; any given odor results in a response in most MBONs, although the magnitude of the response varies among MBON cell types. Unlike the stereotyped response to odors of the olfactory projection neurons that deliver odor information to the MB, the odor tuning of the MBONs is modified by plasticity and varies significantly between individual flies, suggesting that MBONs change their response to odors based on experience (Aso, 2014).

For olfactory associative memory in Drosophila, multiple lines of evidence are consistent with a model in which dopamine-dependent plasticity in the presynaptic terminals of KCs alters the strength of synapses onto MBON dendrites. This is thought to provide a mechanism by which the response of MBONs to a specific odor could represent that odor's predictive value. D1-like dopamine receptors and components of the cAMP signaling pathway, such as the Ca2+/Calmodulin-responsive adenylate cyclase encoded by the rutabaga gene, are required specifically in the KCs for memory formatio and rutabaga was shown to be required for the establishment of the differences in MBON odor tuning between individuals. Reward and punishment recruit distinct sets of dopaminergic neurons (DANs) that project to specific regions in the MB lobes. Moreover, exogenous activation of these DANs can substitute for reinforcing stimuli to induce either appetitive or aversive memory, depending on DAN cell type. In sum, while the identity of the learned odor is likely encoded by the small subset of KCs activated by that odor, whether dopamine-mediated modulation assigns positive or negative valence to that odor would be determined by where in the MB lobes KC-MBON synapses are modulated and thus which MBON cell types alter their response to the learned odor (Aso, 2014ab).

Combining the above observations with the comprehensive anatomical characterization of MB inputs and outputs lays the groundwork for testing models of how the MB functions as a whole. It is suggested that each of the 15 MB compartments-regions along the MB lobes defined by the arborization patterns of MBONs and DANs (see Circuit diagrams of the mushroom body)-functions as an elemental valuation system that receives reward or punishment signals and translates the pattern of KC activity to a MBON output that serves to bias behavior by altering either attraction or aversion. This view implies that multiple independent valuation modules for positive or negative experiences coexist in the MB lobes, raising the question of how the outputs across all the modules are integrated to result in a coherent, adaptive biasing of behavior (Aso, 2014).

Although several MBON cell types have been shown to play a role in associative odor memory, the functions of most MBONs have not been studied. Based on anatomical analyses (Aso, 2014a), it is believed that just 34 MBONs of 21 types provide the sole output pathways from the MB lobes. To gain mechanistic insight into how the ensemble of MBONs biases behavior, it would be first important to know the nature of the information conveyed by individual MBONs and the extent to which their functions are specialized or segregated into different information channels. Then it would be necessary to discover how the activities of individual MBONs contribute to influence the behavior exerted by the complete population of MBONs. Thus, in order to understand how memory is translated into changes in behavior, experimental access would be needed to a comprehensive set of MBONs and investigate how the outputs from different MBONs bias behavior, singly and in combination (Aso, 2014).

In the accompanying paper (Aso, 2014a), the detailed anatomy is described of the DANs and MBONs (see Circuit diagrams of the mushroom body and the generation of intersectional split-GAL4 driver lines to facilitate their study. All but one of the 21 MBON cell types consists of only one or two cells per hemisphere. Dendrites of MBONs that use the same neurotransmitter-GABA, glutamate or acetylcholine-are spatially clustered in the MB lobes. Intriguingly, this spatial clustering resembles the innervation patterns of modulatory input by two clusters of dopaminergic neurons, PPL1 and PAM. MBONs have their axonal terminals in a small number of brain regions, but their projection patterns also suggest pathways for relaying signals between compartments of the MB lobes; three MBONs send direct projections to the MB lobes and several other MBONs appear to target the dendrites of specific DANs (Aso, 2014).

Split-GAL4 drivers give the capability to express genetically encoded effectors in identified MBONs to modify their function. This study examine the roles of specific MBONs in various learning and memory tasks as well as in the regulation of locomotion and sleep. Whether direct activation of specific MBONs are sufficient to elicit approach or avoidance was also studied. The results indicate that the ensemble of MBONs does not directly specify particular motor patterns. Instead, MBONs collectively bias behavior by conveying the valence of learned sensory stimuli, irrespective of the modality of the stimulus or the specific reward or punishment used during conditioning (Aso, 2014).

In the insect brain, a sparse and non-stereotyped ensemble of Kenyon cells represents environmental cues such as odors. The behavioral response to these cues can be neutral, repulsive or attractive, influenced by the prior experiences of that individual and how dopaminergic and other modulatory inputs have changed the weight of its KC-MBON synapses. MBONs are thought to encode the predictive value associated with a stimulus. A fly would then use that information to bias its selection of behavioral responses in an ever-changing environment. The accompanying paper (Aso, 2014), describes in detail the projection patterns of the MBON and DAN cell types that comprise the MB lobes in Drosophila. This report begins the process of determining the nature of the information conveyed by MBONs. Specific MBON cell types were correlated with roles in associative memory and sleep regulation (see A map of MBON functions). These anatomical and behavioral results lay the groundwork for understanding the circuit principles for a system of memory-based valuation and action selection (Aso, 2014b).

Optogenetic activation of MBONs in untrained flies can induce approach or avoidance. The ability of the MBONs to induce changes in behavior in the absence of odors suggests that MBONs can bias behavior directly. This observation is consistent with a recent study showing that flies are able to associate artificial activation of a random set of KCs-instead of an odor stimulus-with electric shock, and avoid reactivation of the same set of KCs in the absence of odors (Vasmer, 2014), a result that recapitulates a finding in the potentially analogous piriform cortex of rodents. This study found that the sign of the response to MBON activation was highly correlated with neurotransmitter type; all the MBONs whose activation resulted in avoidance were glutamatergic, whereas all the attractive MBONs were cholinergic or GABAergic (Aso, 2014).

By tracking flies as they encounter a border between darkness and CsChrimson-activating light, activation of an MBON was shown to bias walking direction. Although activation of glutamatergic MBONs repelled flies, the avoidance behaviors were not stereotyped; flies showed a variety of motor patterns when avoiding the red light. This observation implies that MBONs are unlikely to function as command neurons to drive a specific motor pattern, as has been observed, for example, in recently identified descending neurons that induce only stereotyped backward walking (Aso, 2014).

Rather, fly locomotion can be considered as a goal-directed system that uses changes in MBON activity as an internal guide for taxis. For example, walking in a direction that increases the relative activity of aversive-encoding MBONs, which would occur as a fly approaches an odorant it had previously learned to associate with punishment (or when a fly expressing CsChrimson in an avoidance-inducing MBON approaches the CsChrimson activating light), signals the locomotive system to turn around and walk the other direction. Detailed studies of locomotor circuitry will be required to determine the mechanisms of executing such taxic behaviors and should help elucidate how MBON inputs guide this system (Aso, 2014b).

In this view, the MBON population functions as neither a purely motor nor a purely sensory signal. From the motor perspective, as described above, MBONs bias locomotive outcomes rather than dictate a stereotyped low-level motor program. From the sensory perspective, this study has shown that the same MBONs can be required for experience-dependent behavioral plasticity irrespective of whether a conditioned stimulus is a color or an odor, and irrespective of the specific identity of the odor. Taken together with the fact that MBONs lie immediately downstream of the sites of memory formation, these observations support the proposal that MBONs convey that a stimulus has a particular value-but not the identity of the stimulus itself. This contrasts with sensory neurons whose activity can also induce approach or avoidance, but which do convey the stimulus per se. In mammals, neural representations of abstract variables such as 'value', 'risk' and 'confidence' are thought to participate in cognition leading to action selection. From the point of view of this framework, the MBON population representing the value of a learned stimulus and informing locomotion might be operationally viewed as a cognitive primitive (Aso, 2014).

Co-activating multiple MBON cell types revealed that the effects of activating different MBONs appear to be additive; that is, activating MBONs with the same sign of action increases the strength of the behavioral response, whereas activating MBONs of opposite sign reduces the behavioral response. Thus, groups of MBONs, rather than individual MBONs, likely act collectively to bias behavioral responses. Consistent with the idea of a distributed MBON population code, all 19 MBON cell types imaged so far show a calcium response to any given odor (Aso, 2014).

If it is the ensemble activity of a large number of MBONs that determines memory-guided behavior, how can local modulation of only one or a few MB compartments by dopamine lead to a strong behavioral response? Activation of a single DAN such as PPL1-γ1pedc that innervates a highly localized region of the MB can induce robust aversive memory, yet the odor associated with the punishment will activate MBONs from all compartments, including MBONs that can drive approach as well as those that drive avoidance. It is proposed that, in response to a novel odor stimulus, the activities of MBONs representing opposing valences may initially be 'balanced', so that they do not impose a significant bias. Behavior would then be governed simply by any innate preference a fly might have to that odor, using neuronal circuits not involving the MB. Now suppose an outcome associated with that stimulus is learned. Such learning involves compartment-specific, dopamine-dependent plasticity of the KC-MBON synapses activated by that stimulus. If that occurs, the subsequent ensemble response of the MBONs to that stimulus would no longer be in balance and an attraction to, or avoidance of, that stimulus would result. Consistent with this idea, eliminating MB function by disrupting KCs, which are nearly 10% of neurons in the central brain, had surprisingly minor effects on odor preference (Aso, 2014).

Recent studies of dopamine signaling have implicated distinct sites of memory formation within the MB lobes. Consistent with this large body of work, this study found that one type of PPL1 cluster DAN, PPL1-γ1pedc, played a central role in formation of aversive memory in both olfactory and visual learning paradigms. This DAN also mediates aversive reinforcement of bitter taste. For appetitive memory, PAM cluster DANs that innervate other regions of the MB lobes, in particular the compartments of glutamatergic MBONs, are sufficient to induce appetitive memory. These results strongly suggest that the synaptic plasticity underlying appetitive and aversive memory generally occurs in different compartments of the MB lobes (Aso, 2014).

The sign of preference observed in response to CsChrimson activation of particular MBONs was, in general, opposite to that of the memory induced by dopaminergic input to the corresponding MB compartments. For example, activation of MBON-γ1pedc>α/β and MBON-γ2α'1 attracted flies, whereas DAN input to these regions induced aversive memory. Conversely, activation of glutamatergic MBONs repelled flies, while DAN input to the corresponding regions is known to induce appetitive memory. These results are most easily explained if dopamine modulation led to synaptic depression of the outputs of the KCs representing the CS + stimulus. Consistent with this mechanism, the PE1 MBONs in honeybees as well as the V2 cluster MBONs in Drosophila reduce their response to a learned odor and depression of KC-MBON synapses has been shown for octopamine modulation in the locust MB. Moreover, long-term synaptic depression is known to occur in the granular cell synapses to Purkinje cells in the vertebrate cerebellum, a local neuronal circuit with many analogies to the MB. Other mechanisms are also possible and multiple mechanisms are likely to be used. For example, dopamine may modulate terminals of KCs to potentiate release of an inhibitory cotransmitter such as short neuropeptide F, which has been demonstrated to be functional in KCs and hyperpolarizes cells expressing the sNPF receptor. RNA profiling of MBONs should provide insights into the molecular composition of synapses between KCs and MBONs. It is also noteworthy that the effect of dopamine can be dependent on the activity status of Kenyon cells; activation of PPL1-γ1pedc together with odor presentation induces memory, while its activation without an odor has been reported to erase memory. In the vertebrate basal ganglia, dopamine dependent synaptic plasticity important for aversive and appetitive learning is known to result in both synaptic potentiation and synaptic depression (Aso, 2014).

This study sought the effects of selectively and specifically manipulating the activities of a comprehensive set of MBONs on several behaviors. As a consequence, some insights were gained into the extent to which the relative importance of particular MBONs differed between behaviors. Most obvious was the segregation between appetitive and aversive behaviors. For example, this study found that blocking MBON-γ1pedc>α/β impaired both short-term aversive odor and visual memory, suggesting a general role in aversive memory independent of modality. Conversely, a subset of glutamatergic MBONs was required in all three appetitive memory assays. It still remains to be demonstrated that the outputs of these MBONs are required transiently during memory retrieval. Nevertheless, CsChrimson activation experiments demonstrate that activation of these MBONs can directly and transiently induce attraction and avoidance behaviors (Aso, 2014).

In the cases described above, the DANs and MBONs mediating a particular behavior innervate the same regions of MB lobes. Cases were found where the DANs and MBONs required for a behavior do not innervate the same compartments of the MB lobes. For example, even though several cholinergic MBONs are required for appetitive memory, the compartments with cholinergic MBONs do not receive inputs from reward-mediating PAM cluster DANs, but instead from PPL1 cluster DANs that have been shown to be dispensable for odor-sugar memory. What accounts for this mismatch? Perhaps these cholinergic MBONs' primarily function is in memory consolidation rather than retrieval. But the fact that CsChrimson activation of the cholinergic MBON-γ2α'1 and V2 cluster MBONs resulted in attraction, strongly suggests that at least some of the cholinergic MBONs have a role in directly mediating the conditioned response. Indeed, previous studies found a requirement for cholinergic MBONs (the V2 cluster and MBON-α3) during memory retrieva. One attractive model is that requirement of cholinergic MBONs originates from the transfer of information between disparate regions of the MB lobes through the inter-compartmental MBONs connections within the lobes or by way of connections outside the MB, like those described in the next two sections (Aso, 2014).

The multilayered arrangement of MBONs provides a circuit mechanism that enables local modulation in one compartment to affect the response of MBONs in other compartments. Once local modulation breaks the balance between MBONs, these inter-compartmental connections could amplify the differential level of activity of MBONs for opposing effects. For example, the avoidance-mediating MBON-γ4>γ1γ2 targets the compartments of attraction-mediating MBON-γ2α'1 and MBON-γ1pedc>α/β (Aso, 2014).

This network topology might also provide a fly with the ability to modify its sensory associations in response to a changing environment. Consider the α lobe. Previous studies and the current results indicate that circuits in the α lobe play key roles in long-term aversive and appetitive memory. The α lobe is targeted by MBONs from other compartments and comprises the last layer in the layered output model of the MB. The GABAergic MBON-γ1pedc>α/β and the glutamatergic MBON-β1>α both project to α2 and α3, where their axonal termini lie in close apposition. DAN input to the compartments housing the dendrites of these feedforward MBONs induces aversive and appetitive memory, respectively. This circuit structure is well-suited to deal with conflicts between long-lasting memory traces and the need to adapt to survive in a dynamic environment where the meaning of a given sensory input may change. To test the proposed role of the layered arrangement of MBONs in resolving conflicts between old memories and new sensory inputs, behavioral paradigms will be needed that, unlike the simple associative learning tasks used in the current study, assess the neuronal requirements for memory extinction and reversal learning (Aso, 2014).

The neuronal circuits that are downstream of the MBONs and that might read the ensemble of MBON activity remain to be discovered. However, the anatomy of the MBONs suggests that, at least in some cases, summation and canceling effects may result from convergence of MBON terminals on common targets. For example, the terminals of the sleep-promoting cholinergic MBON-γ2α'1 overlap with terminals of wake-promoting glutamatergic MBONs (γ5β'2a, β'2mp and β'2mp bilateral) in a confined area in CRE and SMP. In addition, some MBONs appear to terminate on the dendrites of DANs innervating other compartments, forming feedback loops. Using these mechanisms, local modulation in a specific compartment could broadly impact the ensemble of MBON activity and how it is interpreted (Aso, 2014).

Testing these and other models for the roles of the MBON network, both within the MB lobes and in the surrounding neuropils, will be facilitated by an EM-level connectome to confirm the synaptic connections that were inferred based on light microscopy. Physiological assays will be needed to confirm the sign of synaptic connections and to measure plasticity. For example, the sign of action of glutamate in the targets of glutamatergic MBONs are not known, as this depends on the receptor expressed by the target cells. In this regard, it is noted that previous studies demonstrated a role for NMDA receptors in olfactory memory (Aso, 2014b).

Neurons that are thought to mediate innate response to odors (a subset of projection neurons from the antennal lobes and output neurons from the lateral horn) also project to these same convergence zones (see Convergence zone of MBON terminals as network nodes to integrate innate and learned valences). It is proposed that these convergence zones serve as network nodes where behavioral output is selected in the light of both the innate and learned valences of stimuli. What are the neurons downstream to these convergence zones? One obvious possibility is neurons that project to the fan-shaped body of the central complex whose dendrites are known to widely arborize in these same areas. It would make sense for the MB to provide input to the central complex, a brain region involved in coordinating motor patterns (Aso, 2014).

Using inactivation to uncover the roles of specific cell types is inherently limited by redundancy and resiliency within the underlying neural circuits. For example, consider the MBONs from the α/β lobes. The output of the α/β Kenyon cells is known to be required for retrieval of aversive memory. The current anatomical and behavioral results show that MBON-γ1pedc>α/β, a cell type that was found to be critical for aversive memory, has terminals largely confined inside the α/β lobes, well-positioned to regulate a total 6 types of MBONs from the α/β lobes. Yet no requirement was detected for any of these MBONs in short term aversive memory when tested individually. The ability to detect phenotypes also depends on the strength of the effector; for example, four glutamatergic MBON drivers showed aversive memory impairment in initial screening assays with a strong inhibitor of synaptic function, but these effects could not be confirmed using a weaker effector (Aso, 2014).

The failure to see effects when inactivating individual cell types is most easily explained by combinatorial roles and redundancy between MBONs. It is noted that this high level of resiliency is very reminiscent of observations made with genetic networks, where less than half of gene knockouts of evolutionarily conserved Drosophila genes result in a detectable phenotype. Whether or not a requirement is detected for a particular MBON in a particular learning paradigm is likely to depend on which DANs are recruited by the unconditioned stimulus used in that paradigm as well as the degree of redundancy in the MBON representation of valence. It will be informative to test systematically whether blocking combinations of MBONs, which did not show significant behavioral effect when blocked separately, results in significant memory impairment. It will also be important in future experiments to employ imaging and electrophysiological methods, in which the activities of individual neurons, and the consequences of plasticity, can be observed without being obscured by redundancy (Aso, 2014).

The MBs are implicated in functions beyond processing of associative memory. MBONs that influence approach to, or avoidance of, a learned stimulus may also have roles in innate preference behaviors for temperature and hunger-dependent CO2 avoidance. Moreover, the behavioral repertoire that MBONs govern are expected to go beyond simple approach and avoidance; the MB is known to play a role in experience-dependent regulation of proboscis extension as well as regulation of sleep and post-mating behaviors such as oviposition. Intriguingly, this study found that MBONs whose activation was repulsive promoted wakefulness, whereas MBONs whose activation was attractive promoted sleep; it would make sense for flies to be awake and attentive in an adverse environment. Other internal states, in addition to sleep, are likely to affect the decision to carry out a particular memory-guided behavior; for example, the state of satiety has been shown to regulate memory expression. The diverse influences of MBONs on behavior can be most easily explained if it is assumed that the activity of the ensemble of MBON conveys an abstract representation of both valence and internal state. In this view, the ensemble of MBONs may represent internal states along axes such as pleasant-unpleasant or aroused-not aroused. It is upon these axes that primitive forms of emotion are thought to have evolved (Aso, 2014).

Drosophila melanogaster can acquire a stable appetitive olfactory memory when the presentation of a sugar reward and an odor are paired. However, the neuronal mechanisms by which a single training induces long-term memory are poorly understood. This study shows that two distinct subsets of dopamine neurons in the fly brain signal reward for short-term (STM) and long-term memories (LTM). One subset induces memory that decays within several hours, whereas the other induces memory that gradually develops after training. They convey reward signals to spatially segregated synaptic domains of the mushroom body (MB), a potential site for convergence. Furthermore, a single type of dopamine neuron was identified that conveys the reward signal to restricted subdomains of the mushroom body lobes and induces long-term memory. Constant appetitive memory retention after a single training session thus comprises two memory components triggered by distinct dopamine neurons (Yamagata, 2014).

In Drosophila, sugar ingestion in a single training session induces stable appetitive odor memory. The results showed that observed memory represents the composite of STM and LTM that are induced by the distinct and complementary reward signals of dopamine. These separate reward signals very well corroborate parallel processing of STM and LTM in the MB. The stm-PAM neurons (a specific cluster of dopamine neurons) induce appetitive memory that decays within hours, whereas memory by ltm-PAM gradually develops after training. These dopamine neurons convey reward signals to spatially segregated synaptic domains of the MB, whereas the other PAM cluster neurons may also contribute to LTM reward. Furthermore, this study identified a single type of dopamine neurons (PAM-α1) encoding a reward signal for LTM. The PAM-α1 targets a spatially restricted subdomain in the α-lobe of the MB, suggesting local associative modulation in the α-lobe. These results indicate that sugar ingestion activates multiple reward signals of different qualities to form complementary memories, rather than a single reward system forming STM that later transforms into LTM (Yamagata, 2014).

In the defensive siphon and tail withdrawal reflex of Aplysia, different modes of 5-HT application to the sensory neurons in the pedal and pleural ganglion differentially induce short-term and long-term sensitization memories. As both ganglia are innervated by a single identified serotonergic neuron, it is likely that the tail shock activates the same cell to induce both forms of sensitization independently. This cellular configuration is in strong contrast to the reward system in Drosophila appetitive memory, despite parallel formation of STM and LTM in both systems. The sugar reward may be more intricately encoded in the fly, given the importance of long-lasting food-related memory in survival (Yamagata, 2014).

Representations of different reinforcing stimuli by the same transmitter seems to be variable. In Drosophila aversive memory, reinforcing signals of electric shock and heat punishment converge to the same dopamine neurons, whereas distinct dopamine neurons are recruited for different aversive stimuli in mammals. Appetitive memory of Drosophila is closer to the latter case. However, a critical difference is that stm- and ltm-PAM neurons seem to signal different properties of the same reward, again pointing to more complex representation of the sugar reward. It would be interesting to compare neuronal representations of different rewarding stimuli, such as ethanol and water. An important future question would be to understand physiological mechanisms by which the MB computes the distinct dopamine inputs to control approach behavior (Yamagata, 2014).

Memory induced by ltm-PAM develops gradually after training. This gradual increase may reflect the time to implement learning-dependent molecular changes. Many molecules that are specifically required for LTM have been identified, and some of these molecules are involved in learning-dependent gene transcription/translation. For instance, fasting-dependent LTM that is formed without training repetition requires CREB-regulated transcription coactivator (CRTC)-mediated cAMP response element binding protein (CREB) transcription in the MB. Such a transcription-dependent mechanism takes time and may thus underlie gradual formation of memory induced by the ltm-PAM (Yamagata, 2014).

The complementary memory dynamics as a consequence of the distinct reward signals suggest that the 'intent' of appetitive memory undergoes a transition from palatability to caloric content. Similar temporal transitions have been found in feeding choice, where flies initially choose sugars according to sweet taste, but later prioritize caloric contents. Similarly, ethanol exposure initially acts as an aversive reinforcer, but eventually turns into reward and induces LTM. The sequential regulation of appetitive behavior by the same stimulus may be conserved across relevant appetitive stimuli. As palatability is not always a faithful predictor of its nutritional value, it may be a general design of reward systems to balance short-term benefit and long-term fitness (Yamagata, 2014).

Go to top

Activity of dopaminergic neurons is necessary and sufficient to evoke learning-related plasticity in neuronal networks that modulate learning. During olfactory classical conditioning, large subsets of dopaminergic neurons are activated, releasing dopamine across broad sets of postsynaptic neurons. It is unclear how such diffuse dopamine release generates the highly localized patterns of plasticity required for memory formation. This study has mapped spatial patterns of dopaminergic modulation of intracellular signaling and plasticity in Drosophila mushroom body (MB) neurons, combining presynaptic thermogenetic stimulation of dopaminergic neurons with postsynaptic functional imaging in vivo. Stimulation of dopaminergic neurons generated increases in cyclic AMP (cAMP) across multiple spatial regions in the MB. However, odor presentation paired with stimulation of dopaminergic neurons evoked plasticity in Ca2+ responses in discrete spatial patterns. These patterns of plasticity correlated with behavioral requirements for each set of MB neurons in aversive and appetitive conditioning. Finally, broad elevation of cAMP differentially facilitated responses in the gamma lobe, suggesting that it is more sensitive to elevations of cAMP and that it is recruited first into dopamine-dependent memory traces. These data suggest that the spatial pattern of learning-related plasticity is dependent on the postsynaptic neurons' sensitivity to cAMP signaling. This may represent a mechanism through which single-cycle conditioning allocates short-term memory to a specific subset of eligible neurons (gamma neurons) (Boto, 2014).

Dopaminergic neurons are involved in modulating diverse behaviors, including learning, motor control, motivation, arousal, addiction and obesity, and salience-based decision making. In Drosophila, dopaminergic neurons innervate multiple brain regions, including the mushroom body (MB), where they modulate aversive learning, forgetting, state-dependent modulation of appetitive memory retrieval, expression of ethanol-induced reward memory, and temperature-preference behavior (Boto, 2014).

Dopaminergic circuits play a particularly critical role in memory acquisition. During olfactory classical conditioning, where an odor (conditioned stimulus [CS]) is paired with an aversive event (e.g., electric shock; the unconditioned stimulus [US]), dopaminergic neurons respond strongly to the aversive US (Mao, 2009). Dopamine functions in concert with activity-dependent Ca2+ influx to synergistically elevate cyclic AMP (cAMP) (Tomchik, 2009) and PKA (Gervasi, 2010), suggesting that dopamine is one component of a molecular coincidence detector underlying learning. Proper dopamine signaling is necessary for aversive and appetitive memory. Moreover, driving activity of a subset of TH-GAL4+ dopaminergic neurons that differentially innervates the vertical α/α' MB lobes (with less dense innervation of the horizontal β/β'/γ lobes, peduncle, and calyx), is sufficient to induce behavioral aversion to a paired odorant in larvae and adult flies. Conversely, stimulation of a different set of Ddc-GAL4+ dopaminergic neurons, the PAM cluster that innervates mainly the horizontal β/β'/γ lobes, is sufficient to induce behavioral attraction to a paired odorant. Thus, dopaminergic neurons comprise multiple circuits with distinct roles in memory acquisition (Boto, 2014).

Multiple subsets of MB neurons receive CS and US information and express molecules associated with the coincidence detection, making them theoretically eligible to generate dopamine/cAMP-dependent plasticity. Yet only some subsets are required to support memory at any given time following conditioning, leaving open the question of how spatial patterns of plasticity are generated during conditioning. This question has been approached, by using a technique to probe the postsynaptic effects of neuronal pathway activation. Odor presentation was paired with stimulation of presynaptic dopaminergic neurons via ectopic expression of the heat-sensitive channel TRPA1, while monitoring postsynaptic effects with genetically encoded optical reporters for Ca2+, cAMP, and PKA in vivo (Boto, 2014).

The present data demonstrate four major points about how dopaminergic circuits function in neuronal plasticity underlying olfactory classical conditioning. (1) Stimulation of small subsets of dopaminergic neurons evokes consistent, compartmentalized elevations of cAMP across the MB lobes. (2) Broad stimulation of dopaminergic neurons generates broad postsynaptic elevation of cAMP, but Ca2+ response plasticity occurs in discrete spatial regions. (3) Stimulation of TH-GAL4+ neurons and Ddc/R58E02-GAL4+ neurons, which mediate opposing behavioral responses to conditioned stimuli, generates an overlapping pattern of Ca2+ response plasticity in the γ lobe, with additional regions recruited by Ddc/R58E02-GAL4+ stimulation. Finally, (4) the spatial pattern of plasticity coincides with differential sensitivity to cAMP in the γ lobe. Collectively, these data suggest that different subsets of neurons exhibit heterogeneous sensitivity to activation of second messenger signaling cascades, which might shape their responses to neuromodulatory network activity and modulate their propensity for recruitment into memory traces (Boto, 2014).

The data suggest that dopaminergic neurons mediate Ca2+ response plasticity largely in the γ lobe and suggest a potential mechanism for localization of short-term, learning-related plasticity. These data coincide with multiple previous studies that have demonstrated a critical role of γ neurons in short-term memory. Rescue of Rutabaga (Rut) in the γ lobe of rut mutants is sufficient to restore performance in short-term memory, whereas rescue in α/β lobes supports long-term memory. Rescue of the D1-like DopR receptor in the γ lobe is sufficient to rescue both short- and long-term memory in a mutant background, suggesting that the γ neurons mediate the dopaminergic input during conditioning. In addition, stimulating MP1 dopaminergic neurons innervating the heel of the γ lobe is sufficient as an aversive reinforcer. Finally, learning induces plasticity in synaptic vesicle release from MB γ lobes, which depends in part on G(o) signaling (Zhang, 2013). The data support a critical role for the γ lobe in short-term memory. Furthermore, the observation of differential sensitivity of the γ lobe to cAMP might provide an elegant explanation for why it is specifically recruited into short-term memory traces (Boto, 2014).

Direct elevation of cAMP was sufficient to generate localized, concentration-dependent Ca2+ response plasticity in the MB γ lobe in these experiments. Because applying forskolin in the bath is expected to elevate cAMP across the brain, the spatial specificity of the effect is remarkable. This was not an acute effect, because the forskolin was washed out before imaging the first postconditioning odor response. At the concentrations that were tested, only the γ lobe was facilitated. Therefore, it is concluded that the γ lobe is most sensitive to elevation of cAMP, which has the effect of differentially recruiting γ neurons into the representation of short-term memory via dopamine-mediated neuronal plasticity. It is possible that additional signaling cascades are involved in generating learning-related plasticity in α/β and α'/β' neurons, given that no Ca2+ response plasticity was observed in those neurons following forskolin application (Boto, 2014).

The dominant model for cellular mechanisms of olfactory associative learning is that integration of information about the conditioned and unconditioned stimuli are integrated by Rut, which functions as a molecular coincidence detector. This would suggest that MB neurons, which receive CS and US information, would exhibit at least somewhat uniform Ca2+ response plasticity. From this molecular and cellular perspective, the finding that the α/β and α'/β' neurons did not exhibit Ca2+ response plasticity when an odor was paired with stimulation of dopaminergic neurons is surprising. These neurons are theoretically eligible to encode memory, because they receive information about the CS and US. However, the finding that γ neurons differentially exhibit dopamine-dependent plasticity following single-cycle conditioning is consistent with the data from the behavioral experiments. In summary, the present results suggest that differential cAMP sensitivity provides a potential mechanism allowing specific subsets of eligible neurons in an array (γ neurons) to differentially encode CS-US coincidence relative to other subsets (α/β neurons) that also receive CS/US information (Boto, 2014).

Locomotion involves rhythmic limb movement patterns that originate in circuits outside the brain. Purposeful locomotion requires descending commands from the brain, but it is not understood how these commands are structured. Focusing on the control of steering in walking Drosophila. First, different limb "gestures" associated with different steering maneuvers are described. Next, this study identified a set of descending neurons whose activity predicts steering. Focusing on two descending cell types downstream from distinct brain networks, this study shows that they evoke specific limb gestures: one lengthens strides on the outside of a turn, while the other attenuates strides on the inside of a turn. Notably, a single descending neuron can have opposite effects during different locomotor rhythm phases, and networks positioned to implement this phase-specific gating were identified. Together, these results show how purposeful locomotion emerges from brain cells that drive specific, coordinated modulations of low-level patterns (Yang, 2023).

Insect asynchronous flight is one of the most prevalent forms of animal locomotion used by more than 600,000 species. Despite profound insights into the motor patterns, biomechanics and aerodynamics underlying asynchronous flight, the architecture and function of the central-pattern-generating (CPG) neural network remain unclear. On the basis of an experiment-theory approach including electrophysiology, optophysiology, Drosophila genetics and mathematical modelling, this study identified a miniaturized circuit solution with unexpected properties. The CPG network consists of motoneurons interconnected by electrical synapses that, in contrast to doctrine, produce network activity splayed out in time instead of synchronized across neurons. Experimental and mathematical evidence support a generic mechanism for network desynchronization that relies on weak electrical synapses and specific excitability dynamics of the coupled neurons. In small networks, electrical synapses can synchronize or desynchronize network activity, depending on the neuron-intrinsic dynamics and ion channel composition. In the asynchronous flight CPG, this mechanism translates unpatterned premotor input into stereotyped neuronal firing with fixed sequences of cell activation that ensure stable wingbeat power and, as we show, is conserved across multiple species. These findings prove a wider functional versatility of electrical synapses in the dynamic control of neural circuits and highlight the relevance of detecting electrical synapses in connectomics (Hurkey, 2023).

Relative motion breaks a camouflaged target from a same-textured background, thus eliciting discrimination of a motion-defined object. Ring (R) neurons are critical components in the Drosophila central complex, which has been implicated in multiple visually guided behaviors. Using two-photon calcium imaging with female flies, this study demonstrated that a specific population of R neurons that innervate the superior domain of bulb neuropil, termed superior R neurons, encoded a motion-defined bar with high spatial frequency contents. Upstream superior tuberculo-bulbar (TuBu) neurons transmitted visual signals by releasing acetylcholine within synapses connected with superior R neurons. Blocking TuBu or R neurons impaired tracking performance of the bar, which reveals their importance in motion-defined feature encoding. Additionally, the presentation of a low spatial frequency luminance-defined bar evoked consistent excitation in R neurons of the superior bulb, whereas either excited or inhibited responses were evoked in the inferior bulb. The distinct properties of the responses to the two bar stimuli indicate there is a functional division between the bulb subdomains. Moreover, physiological and behavioral tests with restricted lines suggest that R4d neurons play a vital role in tracking motion-defined bars. It is concluded that the central complex receives the motion-defined features via a visual pathway from superior TuBu to R neurons and might encode different visual features via distinct response patterns at the population level, thereby driving visually guided behaviors (Duan, 2023).

The recent discovery of the head-direction (HD) system in fruit flies has provided unprecedented insights into the neural mechanisms of spatial orientation. Despite the progress, the neural substance of global inhibition, an essential component of the HD circuits, remains controversial. Some studies suggested that the ring neurons provide global inhibition, while others suggested the Δ7 neurons. The present study provides evaluations from the theoretical perspective by performing systematic analyses on the computational models based on the ring-neuron (R models) and Δ7-neurons (Delta models) hypotheses with modifications according to the latest connectomic data. Four tests were performed: robustness, persistency, speed, and dynamical characteristics. It was discovered that the two models led to a comparable performance in general, but each excelled in different tests. The R Models were more robust, while the Delta models were better in the persistency test. A hybrid model was tested that combines both inhibitory mechanisms. While the performances of the R and Delta models in each test are highly parameter-dependent, the Hybrid model performed well in all tests with the same set of parameters. These results suggest the possibility of combined inhibitory mechanisms in the HD circuits of fruit flies (Chang, 2023).

How different sensory stimuli are collected, processed, and further transformed into a coordinated motor response is a fundamental question in neuroscience. In particular, the internal and external conditions that drive animals to switch to backward walking and the mechanisms by which the nervous system supports such behavior are still unknown. In fruit flies, moonwalker descending neurons (MDNs) are considered command-type neurons for backward locomotion as they receive visual and mechanosensory inputs and transmit motor-related signals to downstream neurons to elicit backward locomotion. Whether other modalities converge onto MDNs, which central brain neurons activate MDNs, and whether other retreat-driving pathways exist is currently unknown. This study shows that olfactory stimulation can elicit MDN-mediated backward locomotion. Moreover, the moonwalker subesophageal zone neurons (MooSEZs), a pair of bilateral neurons, which can trigger straight and rotational backward locomotion. MooSEZs act via postsynaptic MDNs and via other descending neurons. Although they respond to olfactory input, they are not required for odor-induced backward walking. Thus, this work reveals an important modality input to MDNs, a novel set of neurons presynaptic to MDNs driving backward locomotion and an MDN-independent backward locomotion pathway (Israel, 2022).

This study showed that olfactory stimulus can trigger backward locomotion, which is mediated by MDNs. The MooSEZs, which trigger backward locomotion were unequivocally identified in the fly brain. MooSEZs synapse onto MDNs and act both upstream and in parallel to them. Although located in the SEZ, MooSEZs do not seem to respond to gustatory stimuli. Rather, MooSEZs respond to olfactory input. It was further shown that MooSEZs are not necessary for olfactory-driven backward locomotion. However, it was demonstrated that MooSEZs contribute to strong rotational backward locomotion, presumably via DNa01 and DNa02, which are downstream of MooSEZs (Israel, 2022).

The anatomy of the MooSEZs suggests mild polarity, with dendritic regions mostly in the ventral SEZ, and presynaptic projections predominantly in the inferior protocerebrum. The protocerebral innervation is found in the inferior posterior slope (IPS) and the crepine, but also reaching into wedge, vest, and the lower lateral accessory lobe (LAL), posterior to the AL. In the SEZ, the MooSEZs stay in the gnathal ganglia, sparing the prow. The SEZ predominantly receives mechanosensory and gustatory sensory input and SEZ output controls movements involved in feeding behavior. Although the main dendritic region of MooSEZs is in the ventral SEZ, they do not seem to respond to bitter taste. Rather, MooSEZs respond to olfactory input (Israel, 2022).

Interestingly, a significant portion of MooSEZ axonal projections terminate within the LAL, a major premotor region. The LAL receives inputs from the central complex (CX), the navigation hub of the insect brain, as well as from a variety of sensory processing regions, and projects output signals mainly through descending tracts to the VNC. The LAL was shown to be involved in a wide range of orientation-related behaviors across different insect species such as pheromone orientation in moths, flight in locusts, turning in cockroaches, and phonotaxis in crickets. Given the exceptional functional similarities of the LAL across insect species, it will be interesting to find whether MooSEZs are part of a canonical locomotor circuit and thus conserved across insect species (Israel, 2022).

MDNs were demonstrated to receive sensory information from the fly's visual system and mechanosensory input from the VNC. Whether other sensory inputs converge onto MDNs was not resolved yet. The current results add olfaction to the multi-modality input of MDNs. In addition, this study identified the SEZ as participating in MDN input. A recent study in larvae reported the existence of two SEZ neurons, AMB neurons, which mediate backward crawling in response to aversive blue light via activation of MDNs. The current results combined with the results from larvae, suggest that SEZ neurons may also act as modality integrators. The participation of the SEZ as an input source to MDNs also raises the option that SEZ gustatory or tactile input can activate MDNs (Israel, 2022).

These results indicate that MooSEZs add an additional MDN-independent component to the backward walking motor response. Indeed, some experimental evidence from previously published work in both adult flies and larvae are in line with this conclusion. For instance, in adult flies, optogenetic activation of TLAs while blocking MDNs resulted in residual backward walking. In addition, backward movement was correlated with a large membrane voltage difference in DNa01 and DNa02, two non-MDN descending neurons, which are downstream of MooSEZs. In larvae, it was recently demonstrated that MDNs were not necessary for dead end evoked backward crawling. Taken together, it seems that rather than a single command circuit originating in MDNs, additional neural circuits are recruited to allow for an adaptive backward locomotion response (Israel, 2022).

This study shows that unilateral activation of MooSEZs elicits ipsilateral backward turning. This result is analogous to recent findings demonstrating that unilateral activation of specific motor-related neurons can induce walking with a significant ipsilateral turning component. In addition, it was previously demonstrated that unilateral activation of MDNs also led to a steering bias.20 Thus, as MDNs are postsynaptic to MooSEZs, it is possible that the rotational response elicited by MooSEZ unilateral activation might be mediated to some extent by them. However, the data show that blocking MDNs had no effect on MooSEZ-induced backward turning suggesting that the dominant pathway driving the observed rotational backward walking is MDN independent. This conclusion is further supported by the anatomical data demonstrating that both DNa01 and DNa02, which participate in steering maneuvers and descend to the VNC,83,73 are downstream targets of MooSEZs. Flies usually encounter in their natural habitat complex odor plumes, which are constantly changing as a function of the flies' spatial locations relative to the positions of the varied odor sources around them, and the wind structure in their surroundings. Thus, flies are frequently exposed to asymmetrical olfactory cues in their natural environment. The ability of MooSEZs to recruit downstream motor circuits to perform a directional retreat response upon asymmetrical activation may enable flies to effectively guide their behavior under natural conditions (Israel, 2022).

This study demonstrates that an olfactory cue can trigger backward locomotion via MDNs. It was also shown that olfactory input activates MooSEZs that can trigger backward locomotion by themselves independent of MDNs. However, silencing MooSEZs had no effect on odor-evoked backward locomotion. These observations suggest that MooSEZs do not function as command-type neurons for odor-evoked backward locomotion, but rather seem to be part of a distributed neural circuit architecture controlling backward locomotion. Thus, odors may activate multiple parallel neural pathways driving backward locomotion. Indeed, functional redundant signaling also exists in other neuronal circuits in the fly brain. For example, optogenetic activation of LC16 VPNs, which were shown to respond to a visual looming stimulus,46 elicited backward locomotion via MDNs. However, silencing LC16 neurons had no effect on backward locomotion in response to a visual looming stimulus.46 Thus, although optogenetic activation of LC16 neurons drives backward locomotion, silencing these neurons has no effect on the behavioral output to the visual stimulus that activates them. Another example are neurons expressing SIFamide (SIFa). These neurons have elevated activity in starved flies and are involved in hunger-mediated behavior. Acute activation of SIFa neurons was sufficient to increase the response of fed flies to a food-related odor and to enhance their food consumption. However, silencing SIFa neurons did not decrease food intake in starved flies. Taken together, contrary to optogenetic activation of MooSEZs, which generates a robust and constant activation that further activates MDN-independent motor circuits, it seems that ecological odor stimulation elicits a physiologically adequate response in MooSEZs, which is presumably used to support odor-evoked backward locomotion but cannot sustain odor-evoked backward locomotion by itself (Israel, 2022).

Visual motion provides rich geometrical cues about the three-dimensional configuration of the world. However, how brains decode the spatial information carried by motion signals remains poorly understood. This paper studied a collision-avoidance behavior in Drosophila as a simple model of motion-based spatial vision. With simulations and psychophysics, it was demonstrated that walking Drosophila exhibit a pattern of slowing to avoid collisions by exploiting the geometry of positional changes of objects on near-collision courses. This behavior requires the visual neuron LPLC1, whose tuning mirrors the behavior and whose activity drives slowing. LPLC1 pools inputs from object and motion detectors, and spatially biased inhibition tunes it to the geometry of collisions. Connectomic analyses identified circuitry downstream of LPLC1 that faithfully inherits its response properties. Overall, these results reveal how a small neural circuit solves a specific spatial vision task by combining distinct visual features to exploit universal geometrical constraints of the visual world (Tanaka, 2022).

This study explored a collision avoidance behavior in walking Drosophila and its underlying circuit mechanisms as a simple model of motion-based spatial vision. Using high-throughput psychophysics experiments, it was demonstrated that back-to-front motion in the frontolateral visual field—a geometrical cue for near collision—causes slowing in walking flies. Using genetic silencing and activation experiments, the visual projection neuron LPLC1 was shown to be necessary for this putative collision avoidance behavior and its activity is sufficient to cause slowing in walking flies. Physiological response properties of LPLC mirrored the visual tuning of the slowing behavior, most notably in its spatial bias in direction selectivity, which was also consistent with the geometry of near collisions. Using connectomic analyses, optogenetics, and neurochemical imaging and manipulation, it was shown that object-selective T2 and T3 inputs are pooled with direction-selective T4/T5 inputs, likely establishing the object- and direction-selectivity of LPLC1, while spatially biased glutamatergic inhibition creates its position-dependent tuning. Lastly, a downstream neuron of LPLC1 called PLP219 was found to be sufficient to cause slowing, and to inherit the response property of LPLC1 faithfully (Tanaka, 2022).

As objects move relative to an observer, the apparent size and position of the object systematically change as dictated by geometry. How animals detect change in object size and use it to avoid predation has been well studied in various vertebrate species ranging from primates, rodents, birds, and fish , as well as in insects. In contrast, less is known about how and when animals use positional changes or directional motion to detect and avoid collision with moving objects. In general, positional changes of moving objects are more salient than their changes in apparent size: One can show that the maximum apparent expansion rate of an object with radius R moving at a given speed is always less than its maximum apparent translational velocity when the object is more than R away from the observer. Moreover, the ratio between the maximal translation rate and the maximal expansion rate can become arbitrarily large as the object is further and further from the observer. Intuitively, these results correspond to the fact that one can easily tell whether someone 100 meters away is running to the right or left, while it is difficult to tell if that same person is running towards or away from you, based solely on visual motion. This saliency of translation rates is likely one reason that aerial predators employ interception strategies that minimize their apparent positional shifts on their prey's retinae. Less sophisticated pursuit strategies, often used in non-predatory chasing among conspecifics, generate positional changes that can be used by pursuees to detect pursuers. Note that even predators that employ sophisticated strategies will suffer from positional changes after sudden turns of the prey until they settle into a new interception course (Tanaka, 2022).

Positional changes are therefore a useful cue to simply detect objects such as conspecifics and predators, but back-to-front motion in particular can be predictive of future collisions. This is because approaching objects appear to be moving back-to-front only when they will cross the path of the observer in front, which would then pose collision risks if the object slows or stops. This conjecture was by running a simple simulation with randomized trajectories. Based on this geometrical argument, the direction selective slowing behavior of the flies examined in this study is interpreted as a collision avoidance behavior. This is in contrast to other object motion-triggered freezing behaviors in both flies and mice, which are not selective for stimulus direction and thus are unlikely to be a specific response to predicted collision (Tanaka, 2022).

This study found retinotopic biases in the direction selectivity of both behaviors and neural processing. First, the direction selectivity of the collision avoidance slowing to the back-to-front direction was more pronounced in the frontolateral visual field. In addition, direction selectivity of LPLC1 neurons also strongly correlated with the azimuthal location of their receptive field (Tanaka, 2022).

Since the frontolateral visual field is where back-to-front motion is most predictive of immediate collision, the spatial bias in the LPLC1 circuitry can be seen as an adaptation to this geometry (Tanaka, 2022).

Retinotopic biases in visual processing have been found in diverse species. For example, in vertebrate retinae, circuit features such as opsin expression, dendritic morphology, and synaptic strengths can all vary systematically across visual space, depending on species. It is also well established that features such as receptive field sizes and orientation selectivity exhibit retinotopic biases in primate visual cortices. Although these biases have been variously speculated to be adaptations to unique sensory ecology of different species, few were connected to strong geometrical explanations or to behavior (Tanaka, 2022).

Importantly, the geometrical justification provided in this study for the spatial bias in direction selectivity for collision detection is not specific to flies. Thus, it is likely that similar biases exist in other sighted species, arrived at through convergent evolution (Tanaka, 2022).

Indeed, rodent superior colliculus—a center of visual threat detection—has been reported to exhibit a similar retinotopic bias where back-to-front and upward motion is overrepresented in the upper lateral visual field, likely mirroring the geometry of approaching overhead predators (Tanaka, 2022).

Although this study focused on LPLC1's involvement in collision avoidant slowing behavior in walking flies, this does not preclude the possibility that LPLC1 is involved in different behavioral programs in other sensory and behavioral contexts. Supporting this idea, multiple downstream neurons of LPLC1 were found whose activation did not result in slowing and also had divergent visual response properties. Indeed, a previous study reported that strong optogenetic activation of LPLC1 can lead to behavioral phenotypes other than slowing, such as jumping (Tanaka, 2022).

Descending neurons DNp03 and DNp06, which receive inputs from other loom-sensitive, jump-inducing VPNs (LC4, LC6), make good candidates for the neural basis of such jumping phenotypes (Tanaka, 2022).

An interesting question is how the activation of LPLC1 neurons by different stimuli (e. g., small objects moving back-to-front vs. looming objects) results in different behavioral responses. For example, one can imagine that the activation of LPLC1 without activation of other loom sensitive cells (e.g., LC4, LC6) is decoded as the presence of a conspecific in a collision course to initiate slowing, whereas simultaneous activation of LPLC1 alongside other loom detectors strongly implies predators and thus triggers rapid escape. How such population-level decoding and behavioral decision-making is implemented through the network of interglomerular local neurons is of particular interest for future studies (Tanaka, 2022).

In flies, the lobula complex consists of the lobula and lobula plate, which are the highest order brain neuropils that remain specialized for visual processing. Among these two neuropils, lobula plate has been historically under intensive study as the neural basis of visual motion detection and stabilization reflexes, while the functions of the lobula neuropil have remained less clear. The recent series of studies on lobula output neurons have started to show that these neurons detect ethologically relevant objects, like mates and predators, to drive specific behavioral programs. Visual projection neurons innervating both lobula and lobula plate, including LPLC1, are uniquely situated to integrate these object and motion signals. This study showed that LPLC1 likely pools inputs from motion- and object-detecting interneurons (T4/T5 and T2/T3 neurons) to construct a more complex visual feature. While there are other visual projection neuron types spanning lobula plate and lobula whose physiology have been studied, lobula inputs to those neurons remain to be explored (Tanaka, 2022).

Interestingly, a similar computational motif of convergence between motion- and object-detecting pathways seems to be present in the early visual systems of vertebrates as well. Vertebrate retinae are equipped with retinal ganglion cells selective for motion directions as well as small objects. The axon terminals of motion- and object-selective ganglion cells innervate shallowest layers of optic tectum in zebrafish as well as of superior colliculus in mice. While the internal circuitry of the optic tectum / superior colliculus is still not well understood, physiological studies on the neural bases of prey capture in larval zebrafish have identified tectal neurons that show direction selective responses to small objects similar to LPLC1. Similarly, narrow field neurons in mouse superior colliculus, which are also necessary for prey capture behavior, exhibit direction selectivity as well as tight tuning to small object sizes. These results suggest that integration of motion- and object-detector outputs similar to LPLC1 indeed takes place in the optic tectum / superior colliculus. Parallels between vertebrates and invertebrates in the early layers of visual processing and motion detection have been noted. The findings reported in this study extend the computational analogies between insect and vertebrate visual systems to the motif of initial segregation and subsequent convergence of motion and object detecting pathways to drive specialized object-detection behaviors (Tanaka, 2022).

Flexible mapping between activity in sensory systems and movement parameters is a hallmark of motor control. This flexibility depends on the continuous comparison of short-term postural dynamics and the longer-term goals of an animal, thereby necessitating neural mechanisms that can operate across multiple timescales. To understand how such body-brain interactions emerge across timescales to control movement, whole-cell patch recordings were performed from visual neurons involved in course control in Drosophila. The activity of leg mechanosensory cells, propagating via specific ascending neurons, is critical for stride-by-stride steering adjustments driven by the visual circuit, and, at longer timescales, it provides information about the moving body's state to flexibly recruit the visual circuit for course control. Thus, these findings demonstrate the presence of an elegant stride-based mechanism operating at multiple timescales for context-dependent course control. It is proposed that this mechanism functions as a general basis for the adaptive control of locomotion (Fujiwara, 2022).

Neural circuits must both execute the behavioral repertoire of individuals and account for behavioral variation across species. Understanding how this variation emerges over evolutionary time requires large-scale phylogenetic comparisons of behavioral repertoires. This study describes the evolution of walking in fruit flies by capturing high-resolution, unconstrained movement from 13 species and 15 strains of drosophilids. Walking can be captured in a universal behavior space, the structure of which is evolutionarily conserved. However, the occurrence of and transitions between specific movements have evolved rapidly, resulting in repeated convergent evolution in the temporal structure of locomotion. Moreover, a meta-analysis demonstrates that many behaviors evolve more rapidly than other traits. Thus, the architecture and physiology of locomotor circuits can execute precise individual movements in one species and simultaneously support rapid evolutionary changes in the temporal ordering of these modular elements across clades (York, 2022).

Neurons integrate excitatory and inhibitory signals to produce their outputs, but the role of input timing in this integration remains poorly understood. Motion detection is a paradigmatic example of this integration, since theories of motion detection rely on different delays in visual signals. These delays allow circuits to compare scenes at different times to calculate the direction and speed of motion. Different motion detection circuits have different velocity sensitivity, but it remains untested how the response dynamics of individual cell types drive this tuning. This study sped up or slowed down specific neuron types in Drosophila's motion detection circuit by manipulating ion channel expression. Altering the dynamics of individual neuron types upstream of motion detectors increased their sensitivity to fast or slow visual motion, exposing distinct roles for excitatory and inhibitory dynamics in tuning directional signals, including a role for the amacrine cell CT1. A circuit model constrained by functional data and anatomy qualitatively reproduced the observed tuning changes. Overall, these results reveal how excitatory and inhibitory dynamics together tune a canonical circuit computation (Gonzalez-Suarez, 2022).

To navigate towards a food source, animals frequently combine odor cues about source identity with wind direction cues about source location. Where and how these two cues are integrated to support navigation is unclear. This study describes a pathway to the Drosophila fan-shaped body that encodes attractive odor and promotes upwind navigation. This study showed that neurons throughout this pathway encode odor, but not wind direction. Using connectomics, fan-shaped body local neurons called hΔC that receive input from this odor pathway and a previously described wind pathway. hΔC neurons exhibit odor-gated, wind direction-tuned activity, that sparse activation of h∆C neurons promotes navigation in a reproducible direction, and that hΔC activity is required for persistent upwind orientation during odor. Based on connectome data, a computational model was developed showing how hΔC activity can promote navigation towards a goal such as an upwind odor source. The results suggest that odor and wind cues are processed by separate pathways and integrated within the fan-shaped body to support goal-directed navigation (Matheson, 2022).

Many behavioural tasks require the manipulation of mathematical vectors, but, outside of computational models, it is not known how brains perform vector operations. This study shows how the Drosophila central complex, a region implicated in goal-directed navigation, performs vector arithmetic. First, a neural signal in the fan-shaped body is described that explicitly tracks the allocentric travelling angle of a fly, that is, the travelling angle in reference to external cues. Past work has identified neurons in Drosophila and mammals that track the heading angle of an animal referenced to external cues (for example, head direction cells), but this new signal illuminates how the sense of space is properly updated when travelling and heading angles differ (for example, when walking sideways). A neuronal circuit was characterized that performs an egocentric-to-allocentric (that is, body-centred to world-centred) coordinate transformation and vector addition to compute the allocentric travelling direction. This circuit operates by mapping two-dimensional vectors onto sinusoidal patterns of activity across distinct neuronal populations, with the amplitude of the sinusoid representing the length of the vector and its phase representing the angle of the vector. The principles of this circuit may generalize to other brains and to domains beyond navigation where vector operations or reference-frame transformations are required (Lyu 2022).

Whether mammalian brains have neurons that are tuned to the allocentric travelling direction of an animal as in Drosophila is still unknown. Although a defined population of neurons tuned to travelling direction has yet to be highlighted in mammals, such cells could have been missed because their activity would loosely resemble that of the head-direction cells outside a task in which the animal is required to sidestep or walk backwards (Lyu 2022).

Neurons are often modelled as summing their synaptic inputs, but the heading inputs that PFN cells receive from the EPG system appear to be multiplied by the self-motion (for example, optic flow) input, resulting in an amplitude or gain modulation. Multiplicative or gain-modulated responses appear in classic computational models for how neurons in area 7a of the primate parietal cortex might implement a coordinate transformation, alongside similar proposals in mammalian navigation. The Drosophila circuit described in this study strongly resembles aspects of the classic models of the parietal cortex. Units that multiply their inputs are also at the core of the 'attention' mechanism used, for example, in machine-based language processing. The experimental evidence for input multiplication in a biological network may indicate that real neural circuits have greater potential for computation than is generally appreciated (Lyu 2022).

This study describes a travelling direction signal and how it is built; related results and conclusions appear in a parallel study. The mechanisms described for calculating the travelling direction are robust to left-right rotations of the head and to the possibility of the allocentric projection vectors being non-orthogonal. It is possible that the travelling signal of hΔB cells is compared with a goal-travelling direction to drive turns that keep a fly along a desired trajectory. Augmented with an appropriate speed signal (or if the fly generally travels forward relative to its body), the hΔB signal could also be integrated over time to form a spatial-vector memory via path integration. There are hundreds more PFN cells beyond the 40 PFNd and 20 PFNv cells examined in this study, and thus the central complex could readily convert other angular variables from egocentric to allocentric coordinates via the algorithm described in this study. Because many sensory, motor and cognitive processes can be formalized in the language of linear algebra and vector spaces, defining a neuronal circuit for vector computation may open the door to better understanding of several previously enigmatic circuits and neuronal activity patterns across multiple nervous systems (Lyu 2022).

Ephaptic transmission of light information from photoreceptor cells in the retina mediates the increase in firing rate in the OSNs during odor stimulations. This study has not revealed whether the ephaptic transmission directly changes the firing rate of the OSNs. Amputation of the antennal nerve abolished the firing rate increases during sustained light, suggesting that once the light information might be received by neurons in the brain, the information would be relayed by the neurons through the antennal nerve to the antenna, resulting in the firing rate increases in the OSNs (Ikeda, 2022).

While ephaptic coupling has been reported earlier, such as between neighboring neurons within the same sensillum, or between Purkinje cells, which is at a distance of <100 μm , this study shows that ephaptic transmission reaches >200 μm in vivo, equivalent to the depth of the entire fly brain, beyond the distance between neighboring neurons. Light stimulations cause ~10 mV field potential deflections in a retina. If endogenous fields are neglected in the brain, light stimulations may induce ~33.3 mV/mm electric field between the retina and center of the brain (0 mV), since the distance between them is ~300 μm. This electric field is strong enough to modulate neural activities, as even weaker electric fields (<0.5 mV/mm) changed the firing patterns of neurons in vitro (Ikeda, 2022).

In rodents, the firing rate of cerebellar Purkinje cells either decreased or increased when a current was injected into the extracellular field around their axons, causing field potential changes of 0.2 mV. In insects, odor-evoked field potential oscillations whose amplitude is comparable with that caused by the current injection in the rodents, are induced by synchronous firing of olfactory neurons in the antennal lobe which are mediated by GABAergic neurons forming reciprocal synapses with excitatory projection neurons. Changes in the extracellular field potential are commonly observed in many nervous systems. While such extracellular field potential activities have been considered as a side effect of synchronized spiking of neurons, this study suggests that such field potential changes evoked by a sensory stimulus can control the excitability of distant neurons, in addition to adjacent neurons. As ephaptic transmission is more effective at a short distance, the ephaptic transmission from the retinae may contribute significantly to firing rate changes in downstream neurons of the photoreceptor cells in the optic lobe (Ikeda, 2022).

This study also revealed that odor responses of OSNs were clearly modulated when light conditions changed transiently. This mechanism may help flies switch attention to newly presented sensory cues or maintain attention toward those remaining after the change. Turning the light on, for example, reduces the firing rates of the OSNs, which may enable the flies to pay more attention to visual information, whereas turning the light off increases the firing rates of the OSNs, which may help them attend to olfactory sensory cues (Ikeda, 2022).

Recent connectome analyses have revealed the entire synaptic network in the CNS in Drosophila and provides insight into how neural information is subject to synaptic relays to determine the behavioral output. This study has show that ephaptic relays also contribute to modulating the firing rate of distant neurons and modify the sensory responses that can change momentarily in a context-dependent manner. To build an integrated model of the fly brain, we should also consider ephaptic relay of neural information (Scheffer and Meinertzhagen, 2021). The compound eye-antenna model would be a suitable model to determine the role of ephaptic transmission in neural processing (Ikeda, 2022).

Memory consolidation is augmented by repeated learning following rest intervals, which is known as the spacing effect. Although the spacing effect has been associated with cumulative cellular responses in the neurons engaged in memory, this study reports the neural circuit-based mechanism for generating the spacing effect in the memory-related mushroom body (MB) parallel circuits in Drosophila. To investigate the neurons activated during the training, expression was monitored of phosphorylation of mitogen-activated protein kinase (MAPK), ERK [phosphorylation of extracellular signal-related kinase (pERK)]. In an olfactory spaced training paradigm, pERK expression in one of the parallel circuits, consisting of gammam neurons, was progressively inhibited via dopamine. This inhibition resulted in reduced pERK expression in a postsynaptic GABAergic neuron that, in turn, led to an increase in pERK expression in a dopaminergic neuron specifically in the later session during spaced training, suggesting that disinhibition of the dopaminergic neuron occurs during spaced training. The dopaminergic neuron was significant for gene expression in the different MB parallel circuits consisting of alpha/betas neurons for memory consolidation. These results suggest that the spacing effect-generating neurons and the neurons engaged in memory reside in the distinct MB parallel circuits and that the spacing effect can be a consequence of evolved neural circuit architecture (Awata, 2019).

Spaced learning, which consists of repeated learning with appropriate rest intervals, facilitates memory consolidation to a greater extent than repeated learning without rest. This augmentation of memory, known as the spacing effect, has been demonstrated in the animal kingdom. The central issue of this type of memory consolidation is how the neural circuit recognizes the temporally distributed same learning experience as spaced learning without recognizing each learning session as a novel experience and induce memory consolidation. Numerous studies have aimed to elucidate the mechanism by which the neurons recognize spaced learning through the cumulative cellular responses, such as the oscillatory activation of PKA and mitogen-activated protein kinase (MAPK). However, animals encounter various sensory stimuli in the natural environment, and it remains unclear how repeated experiences among intermingled stimuli are specifically subjected to memory consolidation. A recent study has identified the neural correlates of novelty and familiarity in the olfactory system of Drosophila, raising another possibility that the spacing effect may be produced by distinguishing the initial novel training experience from subsequent training experiences at the neural circuit level (Awata, 2019).

The spacing effect in Drosophila has been demonstrated using an aversive training paradigm in which an odor [the conditioned stimulus (CS)] is associated with electric shocks (the unconditioned stimulus). When flies are repeatedly subjected to aversive training with rest intervals, LTM formation occurs, depending on de novo gene expression. In contrast, single aversive training or repeated aversive training without rest intervals (massed training) does not induce LTM formation. Olfactory memory in flies is mediated by parallel circuits in the MB, each of which circuit consists of different types of neurons, including ~ 500 α/β surface (α/βs) neurons, 600 γmain (γm) neurons, and others (see The making of the Drosophila mushroom body and The neuronal architecture of the mushroom body provides a logic for associative learning). Given that retrieval of aversive LTM requires α/βs neurons, the spacing effect may target α/βs neurons for LTM formation. Importantly, MB axons are compartmentalized, and each compartment projects to a different single MB output neuron (MBON). Each MBON exhibits projections to different brain areas, some of which are known to innervate dopamine neurons (DANs) and form feedback loops with MB neurons. This layered structure linking the MB parallel circuits may be important for producing the spacing effect (Awata, 2019).

The present study explored the neural mechanisms underlying the spacing effect by focusing on the MB parallel circuits. The findings suggested that the reduced activity of the MB parallel circuit consisting of γm neurons is important for LTM formation, which affects the activity of the downstream MBON-DAN network. The results suggest that the spacing effect does not only solely depend on the cumulative cellular responses, but also relies on the neural circuit-based computation via the MB parallel circuits (Awata, 2019).

This study adopted an olfactory spaced training paradigm in Drosophila to investigate the neural circuit underlying the spacing effect. Advantage of immunohistochemistry by monitoring phosphorylation of MAPK (ERK), which allowed mapping the neurons activated in the normal training paradigm. Although an increase or decrease in pERK expression may result from either the change in the neural activation or of the ERK-signaling pathway, the optogenetic manipulation in this study suggested that the neural activity change in the MB-MBON-DAN network is significant in LTM formation. While previous studies have demonstrated that γm neurons are actively involved in memory formation, the present study suggests that a decrease in γm activation is also required for LTM formation. As a result, a single GABAergic neuron (MBON-γ1pedc) postsynaptic to γm neurons became inactivated, which, in turn, led to activation of a dopamine neuron (PPL1-α'2α2). The findings further revealed that the PPL1-α'2α2 neuron innervates another MB parallel circuit consisting of α/βs neurons to induce gene expression required for LTM. This study suggests the model in which the multistep linear circuit in the MB would be significant to index spaced learning of the environment. This neural circuit may act in concert with the cumulative cellular responses, such as the previously proposed oscillatory kinase activity during spaced learning. Dopamine-dependent synaptic suppression between MB neurons and MBON as previously demonstrated may also affect the MBON-DAN network (Awata, 2019).

PPL1-α'2α2 activation in the latter sessions of spaced training was required for gene expression in LTM formation. PPL1-α'2α2 activation was observed via calcium imaging during single training. However, increases in PPL1-α'2α2 activation during spaced training via MBON-γ1pedc inactivation may be necessary to provide sufficient signaling for inducing gene expression. Backward spaced training significantly increased pERK expression in the PPL1-α'2α2 neuron, although Arc2 mRNA was not induced, suggesting that association of an odor and electric shocks is also required for Arc2 expression. Consistently, although dTRPA1-dependent activation induced pERK in all α/βs neurons, artificial activation of the PPL1-α'2α2 neuron, and α/βs neurons induced Arc2 protein expression in only a few α/βs neurons, which would be the result of bypassing the requirement of the association due to the artificial activation. Thus, the multiple mechanisms for gene expression should be converged during spaced training, which include activation of the PPL1-α'2α2 neuron (spacing effect information), α/βs neurons (odor information), and other dopamine neurons (electric shock information). A previous study demonstrated that the cfos-expressing neurons show pERK expression upon memory retrieval. In contrast, this study never found pERK expression in the Arc2-expressing neurons upon retraining, memory retrieval, or reverse training. Accordingly, it was found that the pERK-expressing α/βs neurons were slightly reduced following spaced training, compared to single training. There are 2 possibilities. First, the neural activity of the Arc2-expressing neurons could be suppressed by spaced training. Given that synaptic depression between MBs and MBONs has been proposed as the neural correlates of memory, the decreased activity of the Arc2-expressing neurons may play an important role in LTM. Second, the Arc2-expressing neurons could undergo down-regulation in the ERK signaling, although the neurons are activated during memory retrieval. These should be examined in the future study to understand the physiological role of gene expression involved in LTM (Awata, 2019).

Previous studies have suggested that olfactory information relies on sparse coding in the parallel circuits of the MB, although the plasticity of these sparse codings has yet to be explored. In the present study, it was demonstrated that spaced learning preferentially targets sparse coding in the MB parallel circuit consisting of γm neurons via dopamine signaling, leading to memory consolidation in another MB parallel circuit consisting of α/βs neurons. Thus, the neurons responsible for generating the spacing effect and the neurons engaged in memory reside in the different MB parallel circuits. This neural circuit-based computation is accomplished by the MBON-DAN network linking these parallel circuits. This may be generalized to other types of sensory input in Drosophila and may provide insight into the neural representations within parallel neural circuits in other animals (Awata, 2019).

One major question in neuroscience is how to relate connectomes to neural activity, circuit function, and learning. This study offers an answer in the peripheral olfactory circuit of the Drosophila larva, composed of olfactory receptor neurons (ORNs) connected through feedback loops with interconnected inhibitory local neurons (LNs). This study combined structural and activity data and, using a holistic normative framework based on similarity-matching, biologically plausible mechanistic models of the circuit were formulated. In particular, a linear circuit model is considered, for which an exact theoretical solution and a nonnegative circuit model were derived, which was examined through simulations. The latter largely predicts the ORN LN synaptic weights found in the connectome and demonstrates that they reflect correlations in ORN activity patterns. Furthermore, this model accounts for the relationship between ORN LN and LN-LN synaptic counts and the emergence of different LN types. Functionally, it is proposed that LNs encode soft cluster memberships of ORN activity, and partially whiten and normalize the stimulus representations in ORNs through inhibitory feedback. Such a synaptic organization could, in principle, autonomously arise through Hebbian plasticity and would allow the circuit to adapt to different environments in an unsupervised manner. This study thus uncovered a general and potent circuit motif that can learn and extract significant input features and render stimulus representations more efficient. Finally, this study provides a unified framework for relating structure, activity, function, and learning in neural circuits and supports the conjecture that similarity-matching shapes the transformation of neural representations (Chapochnikov, 2023).

Combining the Drosophila larva olfactory circuit connectome, ORN activity data, and a normative model, this study has advanced understanding of sensory computation and adaptation, quantitatively link ORN activity statistics, functional data, and connectome, and makes testable predictions. A canonical circuit model capable of autonomously adapting to different environments is revealed, while maintaining the critical computations of partial whitening (decorrelation) and feature extraction. Such a circuit architecture may arise in other brain areas and may be applicable in machine learning and signal processing. Using ORN activity patterns as input, this normative framework accounts for the biological circuit structural organization and identifies in the connectome signatures of circuit function and adaptation to ORN activity. Such an approach offers a general framework to understand circuit computation and could provide valuable insights into more neural circuits, whose structural and activity data become available (Chapochnikov, 2023).

This paper compares the structural predictions of the normative approach to the connectome. The Neural Network Collection (NNC) model, when adapted to the ORN activity dataset, accounts for key structural characteristics, for example, the ORNs -> LN connection weight vectors. This study asked two questions: 1) Why does the strong resemblance between model and data arise, when the available odor dataset most probably imperfectly matches the true larva odor environment? 2) Why isn't the resemblance even greater, and could the imperfect fit suggest that the model inadequately explains the biological circuit (Chapochnikov, 2023)?

For 1), a possibility is that generic correlations between ORNs arise in large enough ORN activity datasets, causing robust features in the model connectivity. These correlations could result from the intrinsic chemical properties of ORN receptors. Odor statistics would also influence the connection weights, but to a lesser degree. Thus, a more naturalist activity dataset could further improve model predictions (Chapochnikov, 2023).

For 2), first, due to intrinsic noise and variability, no model could be 100% accurate in predicting connectivity. In fact, variability in synaptic count and innervation arises for Drosophilas raised in similar environments, indicating potential 'imprecision' of development and/or learning. Variability was also observed in the left vs. right side connectivity. Second, incomplete ORN activity statistics may decrease prediction accuracy. Third, synaptic count might not exactly reflect synaptic strength. Finally, this model being a simplification of reality misses additional factors shaping circuit connectivity (Chapochnikov, 2023).

This analysis indicates that the matches between model and data likely do not result from chance only, suggesting that the similarity-matching principle influences circuit organization. However, the unsupervised approach assumes that no odor is 'special' for the animal, and thus LNs in the circuit model cluster odors solely based on their representations in the ORN activity space. This contrasts with the biological ORN-LN circuit, where LNs such as Keystone and Picky 0 have specific downstream connections likely related to survival needs and different hardwired animal behaviors, requiring them to detect particular odors. Consequently, the connectivity of such LNs might contribute to the imperfect one-to-one correspondence between the model and the connectome (Chapochnikov, 2023).

The circuit model can learn the optimal connection weights autonomously via Hebbian learning, offering the capacity to adapt to different environments. Studies in adult Drosophila reveal that glomeruli sizes (and thus ORN-LN or ORN-PN synaptic weights) or activity depend on the environment in which the Drosophila grew up. It is, however, unknown if activity-dependent plasticity also occurs in the larval ORN-LN circuit and whether the observed synaptic counts are a result of such plasticity. If present, it is unclear whether the short 6-h life of the larva from which the connectome was reconstructed allows substantial learning to occur and whether changes in synaptic weights would translate to different synaptic counts (Chapochnikov, 2023).

Resolving connectomes of larvae raised in different odor environments and at different times of their life, probing synaptic plasticity, and recording ORN responses to the full odor ensemble present in its environment would help clarify the influence of noise, plasticity, and genetics in circuit shaping (Chapochnikov, 2023).

LNs form a significant part of the neural populations in the brain, perform diverse computational functions, and exhibit extremely varied morphologies and excitabilities. This study proposes a dual role for LNs in this olfactory circuit: altering the odor representation in ORNs and extracting ORN activity features, available for downstream use. In the olfactory system of Drosophila and zebrafish, LNs perform multiple computations, such as gain control, normalization of odor representations, and pattern and channel decorrelation, which is consistent with the current results. Also, in Drosophila the LN population expands the temporal bandwidth of synaptic transmission and temporally tunes PN responses, which was not addressed in this study (Chapochnikov, 2023).

In topographically organized circuits, such as in the visual periphery or in the auditory cortex, distinct LN types uniformly tile the topographic space, and each LN type extracts a specific feature of the input, e.g., in the retina. In nontopographically organized networks, however, the organization and role of LNs remains a matter of research and controversy. This study examined a subcircuit with four LN types, and most types contain several similarly connected LNs. What is the function of multiple similar LNs in the ORN-LN circuit, as also observed in the NNC? First, LNs might differentiate further as the larva grows. Second, several LNs might help expand the dynamic range of a single LN. What are the features extracted by LNs in the Drosophila larva? The NNC model and the distinct connectivity patterns of LN types in the connectome, suggest that different LN types are activated in response to different sets of odors. The extracted features might relate to clusters in ORN activity and to prewired, animal-relevant odors. Since several ORNs --> LN connection weight vectors {𝐰𝑘} in the NNC model resemble those in the biological circuit, the odor clusters identified by the model likely correspond to the set of odors that activate LNs in the biological circuit. The feedforward synaptic count vector from ORNs to the Broad Trio 𝐰BT, which aligns with the first PCA direction of ORN activity and with an ORNs --> LN connection weight vector {𝐰k}in the NNC model could potentially encode the mean ORN activity and thus be related to the global odor concentration. Other LNs might encode features of odors, such as aromatic vs. long-chain alcohols, or specific information influencing larva behavior, but more experiments are required to definitely resolve the features. While the current conclusions differ from a study that found that LN activation is invariant to odor identity), that study imaged several LNs simultaneously and might thus have missed the selectivity of individual LNs (Chapochnikov, 2023).

The connectome reveals LN-LN connections, which are propose play a key role in clustering and shaping the odor representation, and are co-organized with thed ORN-LN connections. A role of LN-LN connections and their relationship to ORN-LN connections is relatively unexplored (Chapochnikov, 2023).

In summary, this study emphasizes the importance of the different ORN-LN and LN-LN connection strengths and argues that LNs are minutely selective and organized to extract features and render the representation of odors more efficient. It is proposed that the circuit's effect on the neural representation of odors in ORNs corresponds to partial ZCA-whitening and divisive normalization. Such computations, which reduce correlations originating from the sensory system and the environment, have appeared in efficient coding and redundancy reduction theories. Partial whitening is in fact a solution to mutual information maximization in the presence of input noise. In this circuit too, complete whitening might also not be desirable due to potential noise amplification. Thus, keeping low-variance signal directions of the input unchanged and dampening larger ones is consistent with mutual information maximization. These conclusions are in line with reports of pattern decorrelation and/or whitening in the olfactory system in zebrafish and mice (Chapochnikov, 2023 and references therein).

The computation in this model also resembles divisive normalization, an ubiquitous computation in the brain, proposed for the analogous circuit in the adult Drosophila. Divisive normalization captures two effects of neuronal and circuit computation: 1) neural response saturation with increasing input up to a maximum spiking rate σ, arising from the neuron's biophysical properties; 2) dampening of the response of a given neuron when other neurons also receive input, often due to lateral inhibition. Aspect (1) is absent in the curren model but could be implemented with a saturating nonlinearity. Depending on the biological value of the maximum output, this model might not accurately capture responses for high-magnitude inputs. However, signatures of (2) are evident in the saturation of the activity pattern magnitudes in ORN axons for increasing ORN soma activity pattern magnitudes. Activity patterns of large magnitude correspond to activity at higher odor concentrations and with a high number of active ORNs. Because such input directions are more statistically significant in the dataset, these stimuli are more strongly dampened by LNs (which encode such directions) than those with few ORNs active. Thus, the model presents a possible linear implementation of a crucial aspect of divisive normalization, which in itself is a nonlinear operation (Chapochnikov, 2023).

Although the basic form of divisive normalization performs channel decorrelation, and not activity pattern decorrelation, models in this paper perform both channel and pattern decorrelation. Nevertheless, a modified version of divisive normalization, which includes different coefficients for the driving inputs in the denominator, performs pattern decorrelation too, as the circuit model. The proposed neural implementations of divisive normalization usually require multiplication by the feedback, which might not be as biologically realistic as circuit implementation described in this study. Several neural architectures similar to the current have been proposed to learn to decorrelate channels, perform normalization, or learn sparse representations in an unsupervised manne. However, these studies either lack a normative/optimization approach or have a different circuit architecture or synaptic learning rules. Using a normative approach has the advantage of directly investigating the underlying principles of neural functioning and also potentially providing a mathematically tractable understanding of the circuit structure and function (Chapochnikov, 2023).

This study complements machine learning approaches to understand neural circuit organization. These approaches use supervised learning and backpropagation to train an artificial neural network to perform tasks such as odor or visual classification. In the olfactory system, circuit configurations arising from this optimization, which could mimic the evolutionary process, display many connectivity features found in biology. Unlike these approaches, a general principle is proposed governing the transformation of neural representations, similarity-matching, and also a mechanism to learn autonomously during the animal's lifetime (Chapochnikov, 2023).

One major question in neuroscience is how to relate connectomes to neural activity, circuit function, and learning. This study offers an answer in the peripheral olfactory circuit of the Drosophila larva, composed of olfactory receptor neurons (ORNs) connected through feedback loops with interconnected inhibitory local neurons (LNs). This study combined structural and activity data and, using a holistic normative framework based on similarity-matching, biologically plausible mechanistic models of the circuit were formulated. In particular, a linear circuit model is considered, for which an exact theoretical solution and a nonnegative circuit model were derived, which was examined through simulations. The latter largely predicts the ORN LN synaptic weights found in the connectome and demonstrates that they reflect correlations in ORN activity patterns. Furthermore, this model accounts for the relationship between ORN LN and LN-LN synaptic counts and the emergence of different LN types. Functionally, it is proposed that LNs encode soft cluster memberships of ORN activity, and partially whiten and normalize the stimulus representations in ORNs through inhibitory feedback. Such a synaptic organization could, in principle, autonomously arise through Hebbian plasticity and would allow the circuit to adapt to different environments in an unsupervised manner. This study thus uncovered a general and potent circuit motif that can learn and extract significant input features and render stimulus representations more efficient. Finally, this study provides a unified framework for relating structure, activity, function, and learning in neural circuits and supports the conjecture that similarity-matching shapes the transformation of neural representations (Chapochnikov, 2023).

Complex behaviors depend on the precise developmental specification of neuronal circuits, but the relationship between genetic programs for neural development, circuit structure, and behavioral output is often unclear. The central complex (CX) is a conserved sensory-motor integration center in insects that governs many higher order behaviors and largely derives from a small number of Type II neural stem cells. This study shows that Imp, a conserved IGF-II mRNA-binding protein expressed in Type II neural stem cells, specifies components of CX olfactory navigation circuitry. This study shows: (1) that multiple components of olfactory navigation circuitry arise from Type II neural stem cells and manipulating Imp expression in Type II neural stem cells alters the number and morphology of many of these circuit elements, with the most potent effects on neurons targeting the ventral layers of the fan-shaped body. (2) Imp regulates the specification of Tachykinin expressing ventral fan-shaped body input neurons. (3) Imp in Type II neural stem cells alters the morphology of the CX neuropil structures. (4) Loss of Imp in Type II neural stem cells abolishes upwind orientation to attractive odor while leaving locomotion and odor-evoked regulation of movement intact. Taken together, this work establishes that a single temporally expressed gene can regulate the expression of a complex behavior through the developmental specification of multiple circuit components and provides a first step towards a developmental dissection of the CX and its roles in behavior (Hamid, 2023).

Neural circuit assembly features simultaneous targeting of numerous neuronal processes from constituent neuron types, yet the dynamics is poorly understood. This study used the Drosophila olfactory circuit to investigate dynamic cellular processes by which olfactory receptor neurons (ORNs) target axons precisely to specific glomeruli in the ipsi- and contralateral antennal lobes. Time-lapse imaging of individual axons from 30 ORN types revealed a rich diversity in extension speed, innervation timing, and ipsilateral branch locations and identified that ipsilateral targeting occurs via stabilization of transient interstitial branches. Fast imaging using adaptive optics-corrected lattice light-sheet microscopy showed that upon approaching target, many ORN types exhibiting 'exploring branches' consisted of parallel microtubule-based terminal branches emanating from an F-actin-rich hub. Antennal nerve ablations uncovered essential roles for bilateral axons in contralateral target selection and for ORN axons to facilitate dendritic refinement of postsynaptic partner neurons. Altogether, these observations provide cellular bases for wiring specificity establishment (Li, 2021).

Prior to this study, it was unclear what cellular mechanism is used for ipsilateral target selection. The data support the following model: ORN axons send out transient interstitial branches at multiple locations along the main axon; the branch that reaches the target region becomes stabilized, and further interstitial branches are suppressed. Stabilization of appropriately positioned branches and elimination of ectopic branches are also used for topographic retinotopic targeting, suggesting that the mechanism of transient interstitial branching followed by stabilization applies to the formation of both continuous and discrete neural maps (Li, 2021).

The exploring branches discovered using AO-LLSM imaging suggest a means by which a growing ORN axon may increase the chance of identifying its target. These exploring branches consist of long, microtubule-based parallel branches that extend and retract rapidly and independently, allowing them to sample a relatively large region for possible targets. The transient occurrence of exploring branches when ORN axons approach their target region suggests that they are induced by local cues near target regions to facilitate target selection. In the ipsilateral antennal lobe, exploring branches were found in ORN types that form ipsilateral branches shortly after the main axon passes by, consistent with them serving as the precursor to the eventual ipsilateral branch. Exploring branches are also found in axon terminals in the contralateral antennal lobe in ORN types, suggesting a general role in facilitating contralateral target identification (Li, 2021).

For ORN types that have a long delay in extending the ipsilateral branch, exploring branches were not observed, suggesting a distinct mechanism for consolidating the ipsilateral branch. Nevertheless, dynamic interstitial branches occur over a prolong period of time until the formation of the ipsilateral branch, suggesting that these ORN types also use stabilization of transient interstitial branches as a means to consolidate the ipsilateral branch (Li, 2021).

In summary, after the initial trajectory choice such that ORN axons navigate in the half of the antennal lobe where their eventual targets are, it is proposed that the next critical step in ORN axon development is the stabilization of transient interstitial branches by target-derived cues, aided at least in part by the exploring branches. Together, these cellular mechanisms begin to explain how each ORN chooses one of 50 glomerular targets precisely (Li, 2021).

A surprising finding is that the cytoskeletal organization of ORN terminals differs substantially from that of classic growth cones, comprising F-actin-based filopodia and lamellipodia at the periphery and a microtubule-enriched central hub. The terminal branches of ORN axons, in particular the exploring branches, are filled with microtubules, whereas F-actin is concentrated at the central hub. Similar cytoskeletal organization were also found in photoreceptor axon terminals. These differences are unlikely due to species difference, as the classic growth cone cytoskeletal organization is found in neurons (mostly dissociated in culture) from Aplysia, Drosophila, and mammals. It cannot be ruled out that F-actin is present in low amount at the terminal of each exploring or post-innervation branch but is beyond the detection limit of utrophin-based F-actin labeling; if so, each terminal branch would have its own growth cone at its tip, resembling classic growth cones. Even if that is the case, ORN axon terminals still differ from classic growth cones by having multiple microtubule-based parallel branches emanating from an F-actin rich central hub. Indeed, EB1-GFP puncta can be found at the tip of the branch, suggesting that microtubules can fill the entire branch. Microtubule polymerization has been shown to mediate membrane extension directly in lipid vesicle (Li, 2021).

It is suspected that the deviation of cytoskeletal organization in ORN axon terminals from the classic growth cone is likely due to the more complex environments axon terminals need to explore in the brain compared with the primary culture. Indeed, a recent study showed that neurons cultured in three-dimensional environments have microtubules extending to the edge of growth cones unconstrained by F-actin. The current findings have important implications for mechanisms that convert cell-surface recognition of extracellular cues into cytoskeletal-based structural changes in axon terminals during axon targeting. Specifically, it is suggested that signaling to microtubule is particularly important at initial stages of target selection (Li, 2021).

Bilaterally symmetric organization of the nervous system is a cardinal feature of all bilaterians. Unilateral antennal nerve severing indicates the requirement of bilateral axons in target selection. The simplest cellular mechanism is direct interactions between ipsilateral and contralateral ORN axons. These interactions may facilitate midline crossing by creating a critical mass of midline-penetrating axons, disruption of which may cause some axons to leave the antennal lobe instead. Later, bilateral axon-axon interactions between the same ORN type may facilitate target selection of contralateral ORNs. The data does not rule out the possibility that bilateral interactions may be indirect; for example, ipsilateral ORNs may change the properties of their partners PNs, which in turn regulate target selection of contralateral ORNs. Indeed, upon unilateral antennal nerve severing, targeting defects was mostly found in ORN types that sequentially innervate ipsilateral and contralateral glomeruli. The ease of severing antennal nerve in explant cultures provides a means to further investigate cellular and molecular mechanisms of bilateral interactions (Li, 2021).

In conclusion, time-lapse imaging has greatly enriched understanding of the cellular events that enable the step-wise assembly of the fly olfactory circuit, and highlight the precise genetic control of multiple steps during ORN axon targeting. These include the choice of a trajectory along which an ORN axon navigates the ipsilateral antennal lobe, the timing and location of stabilizing its ipsilateral branch, and the interactions with contralateral ORN axons to cross the midline and innervate its contralateral target. Finally, ORN axons also help refine dendrites of their partner PNs, which pattern the antennal lobe first. The stage is set to combine live imaging and the cellular insights it has brought with genetic manipulations of key wiring molecules identified by genetic, transcriptomic, and proteomic approaches to reach a deeper level of mechanistic understanding of the circuit assembly process (Li, 2021).

While the targeting precision in explant culture mimics closely in vivo, it takes ORN axons longer to reach the same developmental stage in culture than in vivo. Thus, measurements involving time in explants may be protractions of equivalent events in vivo. The small number of single axons from specific ORN types, due to limited drivers that label specific ORN types strongly in early development, did not allow assessment of the variation of targeting behavior among ORNs of the same type. While this study sampled axon targeting of a large fraction of antennal ORN types, axons from 6 maxillary palp ORN types were not sampled as the explant did not include maxillary palp. It is unclear whether maxillary palp ORN axons follow similar rules as antennal ORN axons. However, as maxillary palp ORN axons reach the antennal lobe substantially later than antennal ORN axons, the lack of maxillary palp ORN axons in explants should not affect the early stages of antennal ORN axon targeting (Li, 2021).

Understanding neuronal representations of odor-evoked activities and their progressive transformation from the sensory level to higher brain centers features one of the major aims in olfactory neuroscience. This study investigated how odor information is transformed and represented in higher-order neurons of the lateral horn, one of the higher olfactory centers implicated in determining innate behavior, using Drosophila melanogaster. This study focused on a subset of third-order glutamatergic lateral horn neurons (LHNs) and characterized their odor coding properties in relation to their presynaptic partner neurons, the projection neurons (PNs), by two-photon functional imaging. Odors evoke reproducible, stereotypic, and odor-specific response patterns in LHNs. Notably, odor-evoked responses in these neurons are valence-specific in a way that their response amplitude is positively correlated with innate odor preferences. It is postulated that this valence-specific activity is the result of integrating inputs from multiple olfactory channels through second-order neurons. GRASP and micro-lesioning experiments provide evidence that glutamatergic LHNs obtain their major excitatory input from uniglomerular PNs, while they receive an odor-specific inhibition through inhibitory multiglomerular PNs. In summary, this study indicates that odor representations in glutamatergic LHNs encode hedonic valence and odor identity and primarily retain the odor coding properties of second-order neurons (Das Chakraborty, 2022).

Insects are the most successful taxon among the whole animal kingdom in terms of their distribution and ability to survive in a multitude of environmental conditions. Largely they rely on their olfactory sense to carry out their fundamental goal-directed behaviors, such as food navigation, mating, ovipositing, or escape from predators. The powerful ability to detect odor cues, to evaluate the information efficiently with a relatively small number of neurons and to transform the neuronal signal into an appropriate behavioral output, makes the insect olfactory system a premier model system for olfactory research. Numerous studies have investigated the neuronal representation of odors at successive neuronal layers from the periphery to higher brain levels using Drosophila melanogaster as a model organism. Although much progress has been made in understanding odor coding at the antennal lobe (AL) level, the coding strategies and processing mechanisms of higher brain centers still remain largely elusive. In this regard, the lateral horn (LH) has recently gained attention as a crucial signal processing center integrating both innate and learned behavioral information. Several studies during recent years have advanced understanding of the anatomical and functional properties of higher-order lateral horn neurons (LHNs) regarding odor processing. The generation of several LH cell-type-specific lines, characterization of polarity and neurotransmitter identity of LHNs, as well as the establishment of detailed EM connectomic datasets have led to a significant progress in the field to study the function of specific LHN classes. The LH is comprised of three categories of neurons, which include LH input neurons (LHINs, which are mainly olfactory projection neurons [PNs] along with mechanosensory, thermosensory, and gustatory neurons), LH local neurons (LHLNs), and LH output neurons (LHONs). In terms of PN-LHN connectivity, the olfactory PNs deriving from individual glomeruli of the AL form stereotyped and conserved connections with certain LHNs. Although all kinds of connections are possible, PNs having similar odor-tuning patterns are prone to target similar LHN types. Certain pairs of narrowly tuned glomeruli encoding ecologically relevant odors and eliciting specific kinds of behavior, such as courtship, aggregation, or food seeking, converge onto the same LHN types and have been shown to be overrepresented in the LH in terms of synaptic densities. Furthermore, a high amount of divergence has also been described to occur at the level of PN to LHN connectivity. Altogether, these complex connectivity patterns, in addition to direct pooling of feedforward inputs from PNs innervating different glomeruli, result in broader tuning patterns of LHNs compared to their presynaptic PNs. In addition to the observed broadly tuned LHNs, narrowly tuned LHNs also exist, which receive input from a single type of PN and which are assumed to be further modulated by odorant-selective inhibition through inhibitory neurons. Although several studies agree that odors are compartmentalized in the LH based on either their chemical identity, behavioral significance, or hedonic valence, it still remains controversial how the odor information is transformed from the PN to the LHN level and which odor features are coded by subtypes of LHNs (Das Chakraborty, 2022).

This study aimed to elucidate the odor coding and processing strategies of LHNs by investigating how a neuronal subset of particular neurotransmitter identity encodes different odor features in the LH and how this representation is correlated to their presynaptic partner neurons in the AL, the uniglomerular PNs (uPNs) and multiglomerular PNs (mPNs). Using photoactivatable GFP, diverse clusters of LHNs were first identified based on their different neurotransmitter identities, and the detailed analysis further focused exclusively on glutamatergic LHNs. Using in vivo two-photon functional imaging, several aspects of odor-evoked activity were characterized in these neurons, such as odor-specific response patterns, reproducibility of repeated stimulations, as well as stereotypy across different individuals. It was possible to successfully demonstrate that attractive and aversive odors are clearly segregated and that the response amplitudes of glutamatergic LHNs are positively correlated with the innate behavioral preference to an odor. How the excitatory input from uPNs and odor-specific inhibition from mPNs contribute to the fine-tuning of odor-specific response patterns of LHNs was further dissected. Altogether, this study demonstrates a significant role of glutamatergic LHNs regarding olfactory processing and extends knowledge about the transformation processes of neuronal information taking place from the periphery to higher brain levels, such as the LH (Das Chakraborty, 2022).

This study functionally characterizes a subset of glutamatergic higher-order neurons in the LH regarding odor coding and processing. It was demonstrated that glutamatergic LHNs respond in a reproducible, stereotypic, and odor-specific manner and these response properties emerge at the level of presynaptic uPNs. Notably, the differential activity levels of glutamatergic LHNs to attractive and aversive odors are positively correlated to the olfactory behavioral preference, indicating that these neurons are mainly tuned to attractive odors. The response features do not arise from the OSN level, but rather derive from local processing within the LH by integrating inputs from multiple olfactory channels through uPNs, which also show a valence-specific odor representation. Furthermore, laser transection experiments demonstrate that these higher-order neurons receive their major excitatory input from uPNs and an odor-specific inhibitory input from mPNs. Lastly, the data show that the observed mPN-mediated inhibition seems to be required for generating an odor-specific response map in the LH (Das Chakraborty, 2022).

A growing body of evidence suggests the existence of an anatomical and functional stereotypy in early processing centers of the insect olfactory pathway. This stereotypy becomes obvious first at the sensory neuron (OSN) level, where OSNs expressing a certain OR target and converge on a stereotypic glomerulus, resulting in a conserved spatial map in the AL between different individuals. This anatomical stereotypy was shown to be retained at the postsynaptic PN level (Das Chakraborty, 2022).

Several studies support the notion that an anatomical stereotypy might also be present at the level of the LH, particularly shown for the PN to LHN connectivity. Along this line, functional studies have demonstrated that LHNs respond in a reproducible and stereotyped manner to odors and this stereotypy is a general feature of the LH. However, how an ensemble of LHNs integrates inputs from several olfactory channels, that is, the presynaptic excitatory and inhibitory PNs, and whether each odor induces a specific and stereotyped response pattern in the LH was not clearly addressed before. In this study, it was demonstrated that each odor is represented by an odor-specific activity pattern in the LH, while the examined glutamatergic LHNs display broader tuning patterns than their presynaptic partner neurons. Although it has been assumed previously that odor specificity may not be encoded in higher-order brain centers, the findings are in accordance with a a recent study, that demonstrated the existence of 33 different LH cell types exhibiting stereotypic odor response properties with increased tuning breadth than PNs. The observed odor-evoked response features of LHNs that are odor-specific but with a broader tuning breadth could be due to several reasons. First, they receive input from similarly as well as differently tuned uPNs; therefore, the topographic map of uPN axonal terminals is not clearly retained at the level of LHNs. Second, LHNs integrate inputs from multiple odor channels, for example, one LHN receives on average excitatory input from ~5.2-6.2 glomeruli. Third, both uPNs and mPNs provide input to glutamatergic LHNs and those LHNs in turn feedback onto those second-order neurons and provide feedforward information to other LHNs. This study observed that uPNs are more efficient in encoding odor identity than the glutamatergic LHNs in the LH, whereas LHNs reveal an improved categorization of odors either based on behavioral significance, 'odor scene' or chemical group. However, the distinct odor-specific response map by glutamatergic LHNs observed in this study suggests that the dimensionality of odor features might not get reduced but still retains information about the odor specificity at the third-order processing stage (Das Chakraborty, 2022).

The functional imaging recordings revealed that odor valence is encoded by glutamatergic LHNs, leading to different activation strengths and patterns for attractive and aversive odors in the LH. Although the valence code is already present at the AL level in uPNs, it could be assumed that the same valence code might be translated to the next level of higher-order neurons. However, evidence of high convergence and divergence of neurons from different sensory modalities in the LH argues against a simple translation of the valence code or odor identity to the LHN level. This observation is well in line with previous studies that have revealed that odor-evoked responses in higher brain centers are generally categorized according to certain odor features as already mentioned above. For example, using functional imaging or patch-clamp recordings of second-order olfactory neurons revealed the existence of distinct attractive and aversive odor response domains in the LH formed by uniglomerular and multiglomular projection neurons (uPNs and mPNs). Such a categorization according to hedonic valence is also visible in this study when the odor-evoked responses of glutamatergic LHNs were plotted in a PCA, taking into account the spatial response patterns as well as the intensity of activity. Although no prominent spatial domain of attractive or aversive odors was evident in the recordings, attractive odors evoked a generally stronger activity when compared to aversive odors in this subset of LHNs. A similar trend is noted in second-order uPNs. However, their response strength was neither correlated with the olfactory preference determined in behavior nor the odor response properties of LHNs. The observed significant correlation between the amount of odor-evoked activity in glutamatergic LHNs to the behavioral valence of an odor leads to a postulate that the activity strengths of higher-order olfactory neurons to odor stimulation might determine the behavioral response - an assumption that needs to be tested in future studies (Das Chakraborty, 2022).

Although this study has documented how the glutamatergic LHNs determine the innate behavioral valence, odor valence in the LH can also be achieved through learning and LHNs have also been shown to play a critical role for learned behavior. A specific class of LHNs (so-called PD2a1/b1) has been reported to mediate innate approach response in addition to learned avoidance response in an odor concentration-dependent manner. Therefore, depending upon the context, the same LHNs can mediate innate as well as learned behavior with opposing valence. Extensive interconnections between the two higher brain centers, the MB and the LH through MBONs and LHONs, signify how the MB modulates the innate olfactory pathways and the valence code in the LH (Das Chakraborty, 2022).

This study suggests that glutamatergic LHNs use different strategies to extract different features of odor information, (1) conserving the identity of an olfactory stimulus by forming an odor-specific activity map and (2) encoding the valence of an odor by integrating information from multiple olfactory channels. It is an ongoing debate regarding how neurons in the LH evaluate an odor stimulus. Gradual detection, encoding, and categorization of an odor at different olfactory processing levels can result in a simple binary choice or complex behavioral responses of an animal. The behavioral preferences are simply reflected in either to approach (positive) or to leave (negative), to copulate (positive) or to reject (negative), and to oviposit (positive) or to find another suitable oviposition site (negative) depending upon the behavioral assays. Hence, based on the context or ecological relevance, an odor can be evaluated either as 'pleasant' or 'unpleasant,' which is well reflected by the response properties of glutamatergic LHNs regarding their valence specificity and their correlation between response strength and behavioral odor preference (Das Chakraborty, 2022).

Notably, such a correlation between response intensity and behavioral preference has also been observed in previous studies, where the amplitude of food odor-evoked activity in neuropeptide F (dNPF) neurons was found to strongly correlate with food odor attractiveness. Another study that combined functional imaging with tracking of innate behavioral responses revealed that the behavioral output could be accurately predicted by a model summing up the normalized glomerular responses, in which each glomerulus contributes a small but specific part to the resulting odor preference\. At the level of the LH, LHNs then integrate the olfactory information from the glomerular responses conveyed via uPNs and mPNs. In a previous study, it was demonstrated that mPNs respond differently to attractive and aversive odors, and mediate behavioral attraction. In this context, this study complements the previous finding by showing that also uPNs display distinct valence-specific responses in the LH to attractive and aversive odors. Information from these two PN pathways becomes integrated and processed in the LH, resulting in valence-specific activities in glutamatergic LHNs, which may in turn determine the relative behavioral preference (Das Chakraborty, 2022).

mPNs inhibit the glutamatergic LHNs in an odor-selective manner, leading to an odor-specific response pattern. According to observations of this study, glutamatergic LHNs receive a stronger inhibition from mPNs in response to the odors vinegar and acetophenone than to benzaldehyde, whereas other odors, such as 2,3 butanedione, linalool, and ethyl acetate, seem not to induce an inhibition. In the absence of this inhibition, it is noted that in addition to an increased response amplitude and altered odor representation, the activity patterns of different odors became more strongly correlated and hence more similar. It is therefore conclude that the mPN-mediated selective inhibition on this glutamatergic subset of LHNs is necessary to maintain odor specificity. Along this line, a previous study has reported that mPNs provide an odor-selective input to vlpr neurons, another class of third-order LHNs. This odor-specific modulation depends on the nature of the odor and results from the stereotyped connectivity of mPNs in the AL as well as in the LH. Although this study provides evidence that uPNs are not presynaptically inhibited by mPNs, another study established that mPNs indeed inhibit uPNs in the LH, facilitating odor discrimination. In addition to uPNs and vlpr neurons, this study identifies another class of recipient neurons (glutamatergic LHNs) that receives mPN-mediated odorant-selective inhibition (Das Chakraborty, 2022).

The glutamatergic LHN population in this study comprises glutamatergic LHONs as well as LHLNs since a broad line was used that labels an ensemble of all glutamatergic LHNs. A previous study employed specific split-Gal4 lines to selectively label LHINs, LHLNs, and LHONs with different neurotransmitter identities and analyzed their connectivity using EM connectomics. By employing artificial activation of specific subsets of LHNs via CsChrimson, it was demonstrated that one class of glutamatergic LHLNs (so-called PV4a1:5) forms excitatory synapses with AV1a1 LHONs that mediate aversive behavioral responses. In the current study, since a generic Gal4-line was used to label all glutamatergic LHNs, this study could neither activate nor silence specific neuronal subsets to observe their relevance with regard to odor-guided behavior. However, when the odor response strength of LHNs was correlated to the olfactory preference, these neurons were observed to be mainly tuned to attractive odors, suggesting that they are involved in mediating approach behavior. Although it was not possible to clearly differentiate the functional properties between different populations of LHONs and LHLNs, this study provides the first understanding of how odors are integrated, transformed, and finally represented in the LH by an ensemble of glutamatergic LHNs (Das Chakraborty, 2022).

Intriguingly, the neurotransmitter identity of this class of LHNs opens up another interesting aspect: Knowing that glutamate can act as an excitatory or inhibitory neurotransmitter, as well as a coincident detector, depending upon the receptors present in the postsynaptic neurons, further experiments are needed to reveal the consequences of glutamatergic LHN input onto their postsynaptic partner neurons. Certainly, the presence of an impressive amount of vesicular glutamate in the LH points towards a significant role of glutamatergic LHNs with regard to odor coding and processing at this higher brain center (Das Chakraborty, 2022).

To identify and memorize discrete but similar environmental inputs, the brain needs to distinguish between subtle differences of activity patterns in defined neuronal populations. The Kenyon cells (KCs) of the Drosophila adult mushroom body (MB) respond sparsely to complex olfactory input, a property that is thought to support stimuli discrimination in the MB. To understand how this property emerges, the role was explored of the inhibitory anterior paired lateral (APL) neuron in the input circuit of the MB, the calyx. Within the calyx, presynaptic boutons of projection neurons (PNs) form large synaptic microglomeruli (MGs) with dendrites of postsynaptic KCs. Combining electron microscopy (EM) data analysis and in vivo calcium imaging, it was shown that APL, via inhibitory and reciprocal synapses targeting both PN boutons and KC dendrites, normalizes odour-evoked representations in MGs of the calyx. APL response scales with the PN input strength and is regionalized around PN input distribution. These data indicate that the formation of a sparse code by the KCs requires APL-driven normalization of their MG postsynaptic responses. This work provides experimental insights on how inhibition shapes sensory information representation in a higher brain centre, thereby supporting stimuli discrimination and allowing for efficient associative memory formation (Prisco, 2021).

While the importance of inhibition in reducing the overlap among stimuli representation has been postulated many decades ago and supported by more recent experimental evidence, the complete mechanism by which inhibition supports stimuli discrimination is not fully understood yet. This study shows that the inhibitory APL neuron, by participating in the structure of MGs of the Drosophila MB calyx, provides inhibition scaled to the PNs excitatory inputs to the calyx. As a result, the average strength and the distribution of postsynaptic responses in KC dendritic claws become more similar across different odour representations. It is suggested that this normalization of postsynaptic responses operated by APL is at the core of pattern separation in the MB (Prisco, 2021).

Pattern separation is obtained in the MB through the formation of a sparse response in the KC layer. The decoding of a sparse code, in general, increases the storage capacity of associative networks, thereby supporting learning and classification tasks. In fact, sparse neuronal representations are described in several organisms including mammals, songbirds, and insects. APL was reported to play a key role in maintaining KCs responses sparse, but the underlying mechanism was far from understood. KCs receive inputs from six to eight PNs on average and, due to KCs high firing threshold, require more than half of those inputs to be coactive to spike. The current data suggest that the APL neuron, by confining KC claws responses within a certain range of activation, ensures that KCs requirement of multiple coactive claws is respected even in the presence of highly variable input strengths. In other words, APL inhibition makes KC input integration dependent on the combinatorial pattern of inputs rather than on the strength of individual inputs. In support of this, blocking APL leads to an increased correspondence between input strength and KC response. Of notice, odour discrimination is achieved at multiple levels of the Drosophila olfactory pathway by different types of inhibitory neurons. Indeed, input gain control normalization has been described for GABAergic interneurons in the AL as well as for inhibitory iPNs at the lateral horn. Additionally, APL and its homolog GGN in the locust showed increased depolarization in response to increasing odour concentration. This, combined with the current findings, suggests that the normalization performed by APL might be acting not only across stimuli identities, but also among concentrations of the same stimulus (Prisco, 2021).

Structural and functional data point towards the involvement of APL in a feedforward loop from PN boutons to KC claws, as well as a closed feedback loop with PN boutons. An advantage of using recurrent circuits to provide inhibition is that such a system can deal with a wide range of input strength, as inhibition and excitation strengths are proportional. Indeed, EM analysis revealed both pre- and postsynaptic connection between APL and PN boutons, linearly proportional to each other, and the differences in the APL calcium influx in response to odours correlated to the variability measured in PNs. So far, APL has been mainly described as a feedback neuron for KCs. However, feedforward inhibitory neurons from the input population onto the next layer have been described in other neuronal networks performing pattern separation. For example, granule cells receive both feedforward and feedback inhibition from Golgi cells at the cerebellar cortex, which are driven by excitatory inputs from the mossy fibers and granule cells' axons, respectively. Moreover, it has recently been demonstrated that Golgi cells recruit scales with the mossy fibers input density, similarly to what was observed in the functional imaging experiments carried out in this study. Additionally, adaptive regulation of KCs sparseness by feedforward inhibition has already been theorized in realistic computational models of insect's MBs. Regarding KCs-to-APL connections, a positive linearity was found among pre- and postsynapses between these two cells, confirming the presence of a local feedback loop within KC dendrites and the APL at the calyx. Furthermore, it has been reported that α/β KCs receive more inhibitory synapses along their dendritic trees compared to γ and α'/β', where the majority of synapses received from the APL is localized on KC claws instead. As the ability of inhibitory synapses to shunt current from excitatory synapses depends on the spatial arrangement of the two inputs, it is speculated that the difference in APL synapses localization could contribute to some of the electrophysiological differences recorded among distinct KCs type. For example, α/β_c KCs were found to have a higher input resistances and longer membrane time constants compared to α'/β' KCs, resulting in a sigmoidal current-spike frequency function rather than a linear one. Additionally, a difference in synapses distribution can also indicate that two inhibitory mechanisms coexist at the MB calyx, similarly to what has been shown in the cerebellum where Golgi cells are responsible for both tonic inhibition, controlling granule cells spike number, gain control, and phasic inhibition, limiting the duration of granule cells responses (Prisco, 2021).

Finally, volumetric calcium imaging showed that the APL inhibition is local within the MB calyx. In particular, a difference was found in the APL calcium transients when flies were stimulated with odours that activate PN subsets with segregated bouton distribution in the calyx. These data suggest that APL inhibition onto MGs can be imagined as a gradient that peaks at the MGs active during a given stimulus and attenuates with distance. Non-spiking interneurons in insects are typically large and characterized by complex neurite branching, an ideal structure to support local microcircuits. As a matter of fact, similar examples of localized APL response as described in the current study have been reported in the Drosophila MB as well as in the APL's homolog GGN in the locust. An advantage of having local microcircuits is that it allows a single neuron to mimic the activity of several inhibitory interneurons, as described in amacrine cells of both mammals. Additionally, a parallel local-global inhibition is suggested to expand the dynamic range of inputs able to activate KCs (Prisco, 2021).

An important open question is whether the APL inhibition onto MGs of the MB calyx is more of a presynaptic phenomenon, therefore acting on PN boutons output, or a postsynaptic one on KCs claws. Functional data reveal a clear impact of APL on the postsynaptic response in MGs, while the PN boutons display a broad range of activity levels. Accordingly, silencing of the GABAA receptor Rdl in KCs increased calcium responses in the MB, including the calyx, and reduced sparseness of odour representations . However, due to the presence of presynapses from APL to PN boutons, a presynaptic component of APL inhibition is certainly possible (Prisco, 2021).

One possible caveat to the hypothesis is given by the fact that reducing GABA synthesis in APL by RNAi has been found to improve olfactory learning. However, this could be explained by a low efficiency of RNAi in this case. Indeed, incomplete silencing via RNAi increases KC output without affecting sparseness, whereas blocking APL output via shibire^ts leads to large, overlapping odour representations and impaired olfactory discrimination (Prisco, 2021).

Taken together, this study provides novel insights on how feedforward inhibition via APL shapes the postsynaptic response to olfactory inputs in the MB calyx and contributes to maintaining odour-evoked KC activity sparse. In the future, it will be interesting to investigate the impact of APL on memory consolidation, which has been associated with structural plasticity in the calyx and with changes in the KC response (Prisco, 2021).

Understanding neuronal representations of odor-evoked activities and their progressive transformation from the sensory level to higher brain centers features one of the major aims in olfactory neuroscience. This study investigated how odor information is transformed and represented in higher-order neurons of the lateral horn, one of the higher olfactory centers implicated in determining innate behavior, using Drosophila melanogaster. This study focused on a subset of third-order glutamatergic lateral horn neurons (LHNs) and characterized their odor coding properties in relation to their presynaptic partner neurons, the projection neurons (PNs), by two-photon functional imaging. Odors evoke reproducible, stereotypic, and odor-specific response patterns in LHNs. Notably, odor-evoked responses in these neurons are valence-specific in a way that their response amplitude is positively correlated with innate odor preferences. It is postulated that this valence-specific activity is the result of integrating inputs from multiple olfactory channels through second-order neurons. GRASP and micro-lesioning experiments provide evidence that glutamatergic LHNs obtain their major excitatory input from uniglomerular PNs, while they receive an odor-specific inhibition through inhibitory multiglomerular PNs. In summary, this study indicates that odor representations in glutamatergic LHNs encode hedonic valence and odor identity and primarily retain the odor coding properties of second-order neurons (Das Chakraborty, 2022).

Sweet and bitter compounds excite different sensory cells and drive opposing behaviors. However, it remains unclear how sweet and bitter tastes are represented by the neural circuits linking sensation to behavior. To investigate this question in Drosophila, this study devised trans-Tango(activity), a strategy for calcium imaging of second-order gustatory projection neurons based on trans-Tango, a genetic transsynaptic tracing technique. Spatial overlap was found between the projection neuron populations activated by sweet and bitter tastants. The spatial representation of bitter tastants in the projection neurons was consistent, while that of sweet tastants was heterogeneous. Furthermore, it was discovered that bitter tastants evoke responses in the gustatory receptor neurons and projection neurons upon both stimulus onset and offset and that bitter offset and sweet onset excite overlapping second-order projections. These findings demonstrate an unexpected complexity in the representation of sweet and bitter tastants by second-order neurons of the gustatory circuit (Snell, 2022).

Connectomes generated from electron microscopy images of neural tissue unveil the complex morphology of every neuron and the locations of every synapse interconnecting them. These wiring diagrams may also enable inference of synaptic and neuronal biophysics, such as the functional weights of synaptic connections, but this requires integration with physiological data to properly parameterize. Working with a stereotyped olfactory network in the Drosophila brain, direct comparisons were made of the anatomy and physiology of diverse neurons and synapses with subcellular and subthreshold resolution. Synapse density and location jointly predict the amplitude of the somatic postsynaptic potential evoked by a single presynaptic spike. Biophysical models fit to data predict that electrical compartmentalization allows axon and dendrite arbors to balance independent and interacting computations. These findings begin to fill the gap between connectivity maps and activity maps, which should enable new hypotheses about how network structure constrains network function (Liu, 2021).

The results show that much of the diversity in mean physiological connection weights between second-order projection neurons (PNs) and third-order lateral horn neurons (LHNs) can be explained by anatomical properties measurable in EM images. The number of synapses and LHN surface area successfully predict somatic unitary excitatory postsynaptic potential (uEPSP) amplitudes for dendritic connections, but severely underpredicts amplitudes for axonal connections. However, a more complex model incorporating neural morphology and synapse locations accurately predicted the physiology of both connection types. This highlights the insufficiency of neural point models (which ignore morphology) for predicting synaptic and neural function (Liu, 2021).

It is perhaps surprising that such a strong correspondence occurs with purely passive models, given the plethora of voltage-gated ion channels expressed in neurons. However, the quiescent network state (ex vivo preparation) and the minimal stimulation (single spike resolution in single neurons) are both favorable conditions for remaining in a passive regime. A slight tendency for underprediction of some of the smaller uEPSP amplitudes is noted, which might indicate some active boosting at the low end, or some sublinear integration at the high end of connection weights. Interestingly, a similar phenomenon has been observed in triplet recordings of PNs and LHNs. Nevertheless, the passive models should serve as a foundation in which to investigate the role of active properties (Liu, 2021).

It was found striking that the relationship between dendritic synapse density and uEPSP amplitude was largely linear, and that connections in the antennal lobe and mushroom body were consistent with this relationship. This suggests a conservation of fundamental biophysics across neurons and synapses. Notably, the capacitance of neural membrane and the resistance of intracellular medium are fairly consistent in different neurons and species. Although membrane resistance is more variable due to differences in ion channel expression and activation, the quiescent network state likely quenches many of these differences. The baseline synaptic conductance is likely fairly uniform across synapses in the Drosophila brain, but does vary in a use-dependent manner. It is thus anticipated that these predictions of synaptic function from PN-LHN connections will apply to other connections in the fly brain, but that some features, such as synaptic plasticity, will remain idiosyncratic (Liu, 2021).

The results also highlight the importance of comparing structure and function with single neuron and single spike resolution. While this study found that connectomic data accurately predicted uEPSPs evoked by single spikes in single neurons, a recent study (using the same data sources) reported lower predictive power for compound EPSP amplitudes evoked by multiple spikes in multiple neurons. Future efforts to link connectivity and physiology may thus face challenges if sufficient physiological resolution is not obtained, but the current results suggest that biophysical models could fill the gap in certain instances. For example, spatial and temporal integration across neural populations can be explicitly modeled to match commonly used experimental measures, such as voltage imaging from neuropil containing multiple cells or cell types (Liu, 2021).

The morphologies of neurons are famously diverse. Yet while morphology affects synaptic efficacy in some neurons, it may play a minimal role in others. For LHNs, this study found evidence of both phenomena: fine morphology within arbors has limited impact on synaptic efficacy, while the inter-arbor cable has a major impact. More specifically, the results show that individual arbors passively integrate synaptic input democratically. This occurs because large local variations in mEPSP amplitude are mostly offset by compensatory variations in voltage attenuation. The electrotonic structure of central Drosophila neurons may thus be similar to the dendrites of cerebellar Purkinje cells, which orchestrate a similar dendritic democracy with passive mechanisms in a heavily branched arbor. Interestingly, this configuration has been attributed to the lack of a central trunk neurite, so the branching characteristics of fly neuron arbors may be a mechanism to achieve uniform synaptic efficacy without special spatial patterning of ion channel expression or synaptic conductances. In addition, because most PN-LHN connections target a single arbor with multiple spatially distributed synapses, much of the residual variability due to synapse location will average out for larger connections. Single arbors may therefore be fundamental 'units' of computation in Drosophila neurons, which can spatially intermingle even within the same brain region. Inter-arbor cables strike a balance between interaction and independence between arbors. Interaction enables neurons to compare inputs arriving on different arbors. This is especially relevant, because axon and dendrite arbors receive their own complements of synaptic inputs. LHNs are thus reminiscent of coincidence-detector neurons in the auditory brainstem, where input from each ear impinges on different dendritic arbors, allowing the connecting cable to compare timing. Inter-arbor cables in local LHNs are longer than necessary to connect the arbors, enabling discrimination of temporal sequences on behaviorally relevant scales of ~10 msec. In contrast, independence between arbors can enable some functions to remain arbor-specific. For instance, arbor-specific structural plasticity or active conductances (e.g., voltage-gated potassium channels) could implement different transformations within each arbor prior to comparison via the inter-arbor cable. Such a configuration could enable more complex computations such as multiplication (Liu, 2021).

The abstraction of intricate morphologies into arbors and cables should prove useful for studying other neurons. Even within the fly brain, a wide range of configurations exist, including neurons with one arbor and no cable, neurons with one arbor and one cable (Kenyon cells without axonal branching), three arbors with interposed cables (optic lobe neurons, and 2-dimensional arrays of dozens of arbors (amacrine neurons). An intriguing possibility is that the arbor and cable configuration largely determines the passive biophysics of these neurons, which could provide a simple organizing framework for predicting the function of neurons with diverse neuron morphologies(Liu, 2021).

The pairing of network connectivity maps with knowledge of neuronal and synaptic physiology provides a foundation to formulate hypotheses about activity maps, because assumptions about the function of each component can be calibrated. This study takes important steps in this direction by showing how synapse densities predict uEPSP amplitudes and how morphology predicts subcellular computation. Moreover, this study demonstrated that a simplified compartmental model (the barbell model) can balance biological accuracy with computational tractability. Incorporating these results into simulations of large neural networks should allow the formulation of more precise mechanistic hypotheses about the function of previously unexplored brain circuits(Liu, 2021).

An important goal for the future will be to incorporate additional sources of knowledge to constrain other properties, such as synaptic plasticity, active conductances, and neuromodulation. For example, short-term plasticity can correlate with the number of presynaptic vesicles or the location of a synapse along the dendrite. A systematic comparison of synaptic ultrastructure with synaptic plasticity may therefore reveal other structural patterns that predict function. Active properties of neurons are less likely to be predictable from ultrastructure but could be predicted from proteomics and transcriptomics. Relating physiological measurements to these data across cell types could be used to calibrate estimates of active biophysical properties. Neuromodulation can reconfigure entire networks, yielding different functions in different conditions. In the fly, dopamine alters physiological synaptic strengths, but it is not clear if such changes are visible in EM images. High resolution mapping of cellular and synaptic biochemical processes would thus be a valuable companion to a connectome. (Liu, 2021).

Another goal will be to incorporate ongoing refinement to wiring diagrams into model calibrations. This is important, because some regions of the hemibrain connectome still have incompletely traced connections. In addition, this connectome also lacks information on electrical synapses, glia, and some subcellular structures, which have important functional roles. As this information becomes available, the calibration of cellular and synaptic predictors can be continually adjusted and improved. (Liu, 2021).

While connectivity maps of increasingly large brain volumes bring new opportunities for understanding network organization, predicting function from structure remains famously difficult. While this study have focused on the function of individual components (i.e., neurons and synapses), a central challenge will be identifying how the operation of entire networks depends on those components. Although physiological properties of synapses and neurons are most easily characterized in quiescent network states (e.g., ex vivo), many network operations occur in highly active states (e.g., in vivo). The combination of connectivity maps with validated models of synaptic and neuronal function should help to bridge this gap, by generating testable predictions of how the anatomy and physiology of neurons and synapses constrain activity maps during behavior (Liu, 2021).

Evaluating odor blends in sensory processing is a crucial step for signal recognition and execution of behavioral decisions. Using behavioral assays and 2-photon imaging, this study has characterized the neural and behavioral correlates of mixture perception in the olfactory system of Drosophila. Mixtures of odors with opposing valences elicit strong inhibition in certain attractant-responsive input channels. This inhibition correlates with reduced behavioral attraction. Defined subsets of GABAergic interneurons provide the neuronal substrate of this computation at pre- and postsynaptic loci via GABAB- and GABAA receptors, respectively. Intriguingly, manipulation of single input channels by silencing and optogenetic activation unveils a glomerulus-specific crosstalk between the attractant- and repellent-responsive circuits. This inhibitory interaction biases the behavioral output. Such a form of selective lateral inhibition represents a crucial neuronal mechanism in the processing of conflicting sensory information (Mohamed, 2019).

This study analyzed the integration of binary odor mixtures of opposing hedonic valences and demonstrate how glomerular-specific inhibition and crosstalk results in an appropriate behavioral output. Glomeruli that strongly respond to the attractive odor are inhibited by the repellent odor in the mixture, which is mediated by defined subsets of GABAergic local interneurons (LNs; see Circuit model for glomerulus-specific crosstalk in the fly AL). Heterogeneity in responses to mixtures has been shown in previous studies where excitation of some glomeruli by one of the mixture components can inhibit the glomeruli activated by the other component. Similar to invertebrates, evidence for non-linearity of mixture interactions has been reported in individual mitral/tufted cells (PNs analogs) in the olfactory bulb of vertebrates. As an alternative scenario it is also conceivable that instead of inhibiting the attractant-coding pathway to shift the behavior towards aversion, the response of the repellent-responsive glomeruli could be boosted via lateral excitation. Lateral excitation has been described to drive synergistic interaction between the binary mixture of cis-vaccenyl acetate and vinegar. Although odors representing sex and food are mutually reinforcing, a binary mixture of odors with opposing valences means a conflicting input. It is therefore postulated that, in contrast to reinforcing input, conflicting sensory input is processed via lateral inhibition in the fly AL. An assumption that would be intriguing to be tested in the future (Mohamed, 2019).

No inhibition of the attractant-responsive glomeruli when stimulating with MIX(+) (a binary mixture of ethyl acetate and benzaldehyde). This lack of inhibition is probably due to the strong ORN input leading to high presynaptic firing rates in the attractant-responsive glomeruli. Consequently, lateral inhibition deriving from the aversive circuit has only a low impact and does not decrease the excitation of the attractant-responsive glomeruli (Mohamed, 2019).

Obviously not all glomeruli that are activated by an attractive odor are inhibited by a repellent in a mixture and might not contribute to the attractiveness of an odor. This observation makes sense in the light of accumulating evidence suggesting that the innate behavioral output is correlated either to the summed weights of specific activated glomeruli or to the activity of single processing channels. The latter argument is supported by the finding that only very few, special glomeruli seem to be valence-specific and induce clear attraction or aversion behavior upon artificial activation (Mohamed, 2019).

It is important to mention that the subset of repellent-responsive glomeruli does also respond to non-aversive and even partly attractive odors, such as E2-hexenal and ethyl benzoate. However, an attractive odorant may indeed activate some aversive input channels beside their main activation of the attractive circuitry (or the other way around). What actually matters is the behavioral output that is consequently elicited when a specific glomerulus becomes activated. For example, ORNs that respond to CO2 are also activated by ethyl benzoate and E2-hexenal. However, the CO2 circuit has been clearly demonstrated to mediate behavioral aversion. Following this argument, artificial activation of glomeruli DL1 and/or DL5 leads to aversive behavior, while silencing DM1 and DM4 abolished attraction to the attractant. These experiments provide evidence that activation of the repellent- and attractant-responsive glomeruli causes a valence-specific behavior, and can therefore be defined as attractive or aversive input channels, respectively (Mohamed, 2019).

Interestingly, one exception was observed in the data set: although the repellent odor geosmin reduced the attraction to balsamic vinegar in the mixture, no mixture inhibition was observed. The detection of geosmin is one of the rare cases, where an odor is detected by only one receptor type and consequently activates only one glomerulus. Similar specialized pathways have been described for the detection of sex pheromones and CO2. Glomeruli processing these ecologically labeled lines differ from broadly tuned glomeruli with regard to their neuronal composition. Hence, it is conceivable that the narrowly tuned geosmin-responsive glomerulus does not exhibit strong interglomerular interactions and has therefore a different impact on the attractant-responsive glomeruli. Mixture interactions between geosmin and attractive odors might be implemented in higher processing centers which contain circuit elements mediating interactions between odors (Mohamed, 2019).

Lateral inhibition, which is believed to enhance contrast and to facilitate discrimination of similar stimuli, is an important motif throughout the nervous system. In mice, dense center-surround inhibition refines mitral cell representation of a glomerular map56, while other evidence showed that lateral inhibition can be rather selective and biased between different mitral cells. In accordance with the olfactory bulb, the AL exhibits broad, selective or even both forms of lateral inhibition, whereby certain glomeruli can show different sensitivities towards an inhibitory input. Lateral inhibition in the Drosophila AL is largely mediated through GABA. Most of the GABAergic inhibition in the Drosophila AL has been shown to take place predominantly on the presynaptic site mediated through GABAA and GABAB receptors. In addition, PNs also receive GABAergic inhibition via GABAA and/or GABAB receptors from LNs. Notably, this study found that two out of four attractant-responsive glomeruli are inhibited on the pre- and postsynaptic levels (via GABAB- or GABAA-receptors), while the other two glomeruli are inhibited only presynaptically through GABAB-type receptors. Previous results have so far shown that GABAA-type receptors contribute weakly to lateral inhibition and shape the early phase of odor responses. However, the data demonstrate that GABAA-type receptors largely mediate mixture-induced inhibition during the full period of the odor presentation which is reminiscent to tonic inhibition in the mammalian system (Mohamed, 2019).

This study shows that mixture-induced lateral inhibition of the attractant-responsive glomeruli was abolished when GABA synthesis was silenced in mostly patchy LNs. Hence the data suggest, in consistency with previous studies, that LNs with more selective innervations mediate glomerulus-specific interactions and rather contribute to mixture processing, while pan-glomerular LNs (e.g., GH298-Gal4 and H24-Gal4), that globally release GABA, might be involved in gain control (Mohamed, 2019).

Interestingly, the repellent-responsive glomeruli DL1 and DL5 did not show any mixture interaction, but mediate the lateral inhibition of the attractant-responsive glomeruli. Two possible scenarios would provide the neuronal substrate for this mechanism dependent on either the donor (i.e. LNs) or the receiver (i.e. glomerulus) side. First, since glomeruli vary dramatically in their GABA sensitivity and consequently their sensitivity to LN activation14, lateral inhibition is heterogeneous across different glomeruli. Second, lateral inhibition is biased among different glomeruli due to a glomerulus-specific synaptic distribution of pre- and postsynapses of GABAergic LNs, i.e. the GABA release is not uniform. This assumption is supported by data revealing that GABAergic LNs possess a higher density of postsynapses in DL1 and DL5 than in the attractant-responsive glomeruli. In line with the current findings, EM based data from the larvae AL describe GABAergic, oligoglomerular 'choosy' LNs with a clear polarity contributing to postsynaptic inhibition for most glomeruli, while they receive inputs from only a small glomerular subset. Hence, there is strong evidence that some glomeruli can drive lateral inhibition in other glomeruli. Both scenarios could either occur separately or reinforce each other. Moreover, it might be ecological relevant not to inhibit the input of the aversive pathways since these are associated with life-threatening situations that should be coded reliably and rather override an attractive input (Mohamed, 2019).

In contrast to expectations, sole photoactivation of DL1 or DL5 or stimulation with the repellent alone did not induce inhibition in the attractant-responsive glomeruli. This might be due to the low spontaneous activity of ORNs innervating the attractant-responsive glomeruli, which correlates with spontaneous fluctuations in the membrane potential of the postsynaptic PNs. Consequently, inhibitory responses (i.e. hyperpolarizations) are difficult to capture with calcium imaging (Mohamed, 2019).

In other sensory systems, lateral inhibitory connections of neuronal subsets involved in sensory processing have been elucidated in great detail, such as in the retina of mice or the rat visual cortex. Also for the Drosophila AL, previous studies suggested that glomerular subgroups are connected via inhibitory LNs. However, these studies could neither pinpoint the precise connections nor their significance for behavioral perception. The data provide evidence for a specific inhibitory crosstalk between identified glomeruli and substantiate the existence of selective lateral inhibition in the fly AL. The postulated network circuits offer insights into the principle of sensory integration. It will be intriguing to see whether neuron-specific crosstalk represents a general phenomenon to integrate multiple and rather conflicting input channels in other sensory modalities (Mohamed, 2019).

Local interneurons (LNs) in the Drosophila olfactory system exhibit neuronal diversity and variability, yet it is still unknown how these features impact information encoding capacity and reliability in a complex LN network. This study used two strategies to construct a diverse excitatory-inhibitory neural network beginning with a ring network structure and then introduced distinct types of inhibitory interneurons and circuit variability to the simulated network. The continuity of activity within the node ensemble (oscillation pattern) was used as a readout to describe the temporal dynamics of network activity. Inhibitory interneurons were found to enhance the encoding capacity by protecting the network from extremely short activation periods when the network wiring complexity is very high. In addition, distinct types of interneurons have differential effects on encoding capacity and reliability. Circuit variability may enhance the encoding reliability, with or without compromising encoding capacity. Therefore, this study has described how circuit variability of interneurons may interact with excitatory-inhibitory diversity to enhance the encoding capacity and distinguishability of neural networks. This work has evaluate the effects of different types and degrees of connection diversity on a ring model, which may simulate interneuron networks in the Drosophila olfactory system or other biological systems (Tsai, 2018).

Drosophila olfactory local interneurons (LNs) in the antennal lobe are highly diverse and variable. How and when distinct types of LNs emerge, differentiate, and integrate into the olfactory circuit is unknown. Through systematic developmental analyses, this study found that LNs are recruited to the adult olfactory circuit in three groups. Group 1 LNs are residual larval LNs. Group 2 are adult-specific LNs that emerge before cognate sensory and projection neurons establish synaptic specificity, and Group 3 LNs emerge after synaptic specificity is established. Group 1 larval LNs are selectively reintegrated into the adult circuit through pruning and re-extension of processes to distinct regions of the antennal lobe, while others die during metamorphosis. Precise temporal control of this pruning and cell death shapes the global organization of the adult antennal lobe. These findings provide a road map to understand how LNs develop and contribute to constructing the olfactory circuit (Liou, 2018.

Local inhibitory neurons control the timing of neural activity in many circuits. To understand how inhibition controls timing, it is important to understand the dynamics of activity in populations of local inhibitory interneurons, as well as the mechanisms that underlie these dynamics. This study describes the in vivo response dynamics of a large population of inhibitory local neurons (LNs) in the Drosophila melanogaster antennal lobe, the analog of the vertebrate olfactory bulb, and dissects the network and intrinsic mechanisms that give rise to these dynamics. Some LNs respond to odor onsets ("ON" cells) and others to offsets ("OFF" cells), whereas still others respond at both times. Moreover, different LNs signal odor concentration fluctuations on different timescales. Some respond rapidly, and can track rapid concentration fluctuations. Others respond slowly, and are best at tracking slow fluctuations. A continuous spectrum of preferred stimulation timescales was found among LNs, as well as a continuum of ON-OFF behavior. Using in vivo whole-cell recordings, it was shown that the timing of an LN's response (ON vs OFF) can be predicted from the interplay of excitatory and inhibitory synaptic currents that it receives. Meanwhile, the preferred timescale of an LN is related to its intrinsic properties. These results illustrate how a population of inhibitory interneurons can collectively encode bidirectional changes in stimulus intensity on multiple timescales, and how this can arise via an interaction between synaptic and intrinsic mechanisms (Nagel, 2016).

Go to top

In nature, animals form memories associating reward or punishment with stimuli from different sensory modalities, such as smells and colors. It is unclear, however, how distinct sensory memories are processed in the brain. This study established appetitive and aversive visual learning assays for Drosophila that are comparable to the widely used olfactory learning assays. These assays share critical features, such as reinforcing stimuli (sugar reward and electric shock punishment), and allow direct comparison of the cellular requirements for visual and olfactory memories. It was found that the same subsets of dopamine neurons drive formation of both sensory memories. Furthermore, distinct yet partially overlapping subsets of mushroom body intrinsic neurons are required for visual and olfactory memories. Thus, these results suggest that distinct sensory memories are processed in a common brain center. Such centralization of related brain functions is an economical design that avoids the repetition of similar circuit motifs (Vogt, 2014).

Devising a transparent electric shock grid module made it possible to apply the same visual stimulation in aversive and appetitive conditioning assays. Also an integrated platform was developed for fully automated high-throughput data acquisition using customized software to control the presentation of electric shock and visual stimuli while making video recordings of behavior. In these assays, memory performance is based on altered visual preference in walking flies, a task likely to be less demanding than the constant flight required for flight simulator learning. These advantages facilitate behavioral examination of many genotypes (Vogt, 2014).

Circuits underlying olfactory and visual memory can be optimally compared when the sugar reward and electric shock punishment are matched between the two modalities. Visual and olfactory memories share the same subsets of dopamine neurons that convey reinforcing signals. This shared requirement of the transmitter system between visual and olfactory learning has been described in crickets. However, the pharmacological manipulation used in these studies does not allow further circuit dissection (Vogt, 2014).

For electric shock reinforcement, identified neurons in the PPL1 cluster, such as MB-MP1, MB-MV1 and MB-V1, drive aversive memories in both visual and olfactory learning, while the MB-M3 neurons in the PAM cluster seem to be involved specifically in aversive olfactory memory. Thus, overlapping sets of dopamine neurons appear to represent electric shock punishment in both visual and olfactory learning with olfactory aversive memory probably recruiting a larger set. Previous studies have shown that the MB-M3 neurons induce aversive olfactory memory that increases stability of other memory components. Olfactory memories last longer than visual memories potentially due to the recruitment of additional dopamine neurons (Vogt, 2014).

In appetitive conditioning, PAM cluster neurons play crucial roles in both olfactory and visual memories. Which cell types in these clusters are involved and whether there is a cellular distinction between olfactory and visual memory requires further analysis at the single cell level. Most importantly, all these neurons convey dopamine signals to restricted subdomains of the MB. The blockade of octopamine neurons did not impair appetitive visual memories with sucrose. The involvement of octopamine neurons may be more substantial when non-nutritious sweet taste rewards are used, as has been shown in olfactory learning (Vogt, 2014).

In addition to these shared reinforcement circuits in the MB, the necessity of MB output for visual memory acquisition and retrieval is also consistent with olfactory conditioning. Taken together, these results suggest that the MBs harbor associative plasticity for visual memories and support the conclusion that similar coincidence detection mechanisms are used to form memories within the MBs. Centralization of similar brain functions spares the cost of maintaining similar circuit motifs in different brain areas and may be an evolutionary conserved design of information processing. Such converging inputs of different stimuli into a multisensory area have even been described in humans (Vogt, 2014).

'Flight simulator' visual learning was shown to require the central complex but not the MBs. Although this appears to contradict the current study, it is noted that there are important differences between the behavioral paradigms employed. In the flight simulator, a single tethered flying Drosophila is trained to associate a specific visual cue with a laser beam punishment, to later on avoid flying towards this cue in the test. Although this study controlled for visual context consistency and the 'operant component' of the flight simulator training, any other difference could account for the differential requirement of brain structures. Given that flies during flight show octopamine-mediated modulation of neurons in the optic lobe, similar state-dependent mechanisms might underlie different requirement of higher brain centers. Thus, it is critical to design comparable memory paradigms (Vogt, 2014).

This study together with previous results in associative taste learning highlights the fact that the role of the MB in associative learning is not restricted to one sensory modality or reinforcer. This study found that olfactory and visual memories recruit overlapping, yet partly distinct, sets of Kenyon cells (see Circuit model of olfactory and visual short-term memories). In contrast to the well-described olfactory projection neurons, visual inputs to the MB remain unidentified. No anatomical evidence has been reported in Drosophila for direct connections between optic lobes and MBs although such connections are found in other insects. Also afferents originating in the protocerebrum were found to provide multi-modal input to the MB lobes of cockroaches. Thus, Drosophila MBs may receive indirect visual input from optic lobes, and the identification of such a visual pathway would significantly contribute to understanding of the MB circuits (Vogt, 2014).

Given the general requirement of the γ lobe neurons, visual and olfactory cues may be both represented in the γ neurons. Consistently, the dopamine neurons that convey appetitive and aversive memories heavily project to the γ lobe. In olfactory conditioning, the γ lobe was shown to contribute mainly to short-term memory. This converging evidence from olfactory and visual memories suggests a general role for the γ lobe in short-lasting memories across different sensory modalities. Previous studies found that the MB is also involved in sensorimotor gating of visual stimuli or visual selective attention. Therefore, the MB circuits for visual associative memories might be required for sensorimotor gating and attention (Vogt, 2014).

Interestingly, the contribution of the α'/β' lobes is selective for olfactory memories. This Kenyon cell class is more specialized to odor representation, as the cells have the broadest odor tuning and the lowest response threshold among the three Kenyon cell types (Vogt, 2014).

The role of α/β neurons in visual memories is also limited. The α/β neurons might play more modulatory roles in specific visual memory tasks, such as context generalization, facilitation of operant learning and occasion setting. This modulatory role of the α/β neurons is corroborated in olfactory learning, where they are preferentially required for long-lasting memories (Vogt, 2014).

Differentiated but overlapping sensory representations by KCs may be conserved among insect species. In honeybees, different sensory modalities are represented in spatially segregated areas of the calyx, whereas the basal ring region receives visual and olfactory inputs. The MB might thus have evolved to represent the sensory space of those modalities that are subject to associative modulation (Vogt, 2014).

The mushroom body in the fruitfly Drosophila melanogaster is an associative brain centre that translates odour representations into learned behavioural responses. Kenyon cells, the intrinsic neurons of the mushroom body, integrate input from olfactory glomeruli to encode odours as sparse distributed patterns of neural activity. This study has developed anatomic tracing techniques to identify the glomerular origin of the inputs that converge onto 200 individual Kenyon cells. Each Kenyon cell integrates input from a different and apparently random combination of glomeruli. The glomerular inputs to individual Kenyon cells show no discernible organization with respect to their odour tuning, anatomic features or developmental origins. Moreover, different classes of Kenyon cells do not seem to preferentially integrate inputs from specific combinations of glomeruli. This organization of glomerular connections to the mushroom body could allow the fly to contextualize novel sensory experiences, a feature consistent with the role of this brain centre in mediating learned olfactory associations and behaviours (Caron, 2013).

Olfactory perception in the fly is initiated by the binding of an odorant to an ensemble of olfactory sensory neurons (OSNs) in the antennae, resulting in the activation of a unique and topographically fixed combination of glomeruli in the antennal lobe (AL). The discrimination of odours therefore requires the integration of information from multiple glomeruli in higher olfactory centres. AL projection neurons (PNs) extend dendrites into a single glomerulus and project axons that bifurcate to innervate two distinct brain regions, the lateral horn and the MB. The invariant circuitry of the lateral horn is thought to mediate innate behaviours, whereas the MB translates olfactory sensory information into learned behavioural responses. PN axons that innervate the MB terminate in large boutons. A given KC extends a small number of dendritic “claws”, with each claw receiving information from only one PN bouton. A single bouton connects to multiple KC claws to form a discrete anatomic structure, the microglomerulus. Each KC projects an axon to one of the three different classes of MB lobes, α/β, α’/β’, or γ, where it synapses upon a relatively small number of extrinsic output neurons (Caron, 2013).

Electrophysiological and optical imaging studies show that odorants activate sparse subpopulations of KCs distributed across the MB without spatial preference. Individual KCs could be connected to preferential combinations of glomeruli that are co-ordinately activated by behaviourally relevant odours. Alternatively, KCs may not receive structured input; rather the glomerular inputs may be random, a feature that maximizes the diversity of odour representations in the MB. This study has exploited the specialized structure of the PN-KC synapse to characterize the glomerular origin of the PNs that converge onto individual KCs (Caron, 2013).

Photoactivatable green fluorescent protein (PA-GFP) was expressed in all neurons of the fly and a single KC was photolabelled. It was observed that individual photolabelled KCs elaborate between 2 and 11 dendritic claws restricted to the main olfactory calyx. The axonal projections of a labelled KC can be traced into either the α/β, α’β’, or γ lobes of the MB. Texas red dextran was then electroporated into the centre of a single KC claw, filling the PN bouton innervating that claw. Retrograde transfer of the dye labels a single PN and its associated AL glomerulus, providing further evidence that an individual KC claw receives input from only a single glomerulus (Caron, 2013).

It was verified that this tracing method identifies functional connections between PNs and KCs. Functional imaging was performed on flies that express the calcium indicator GCaMP3 in most KCs to identify the claws activated by the stimulation of a single glomerulus. Electroporation of dye into an activated microglomerulus labels a single PN that innervates the stimulated glomerulus. Thus, the electroporation of dye into a KC claw allows faithful identification of the PN to which it is functionally connected (Caron, 2013).

The strategy of photolabelling a single KC and sequential electroporation of dye into each of its claws was used to define glomerular inputs to an individual KC. In initial experiments, PA-GFP was expressed in all neurons and 100 randomly chosen KCs were photolabelled in 100 different flies. Among the 100 photolabelled KCs, 84 α/β KCs, 14 α’/β’ KCs, but only 2 γ KCs were identified. Each MB contains about 1000 α/β KCs, 370 α’β’ KCs, and 670 γ KCs. γKCs are underrepresented in this initial data set. This is likely to result from the spatial segregation of their cell bodies, which renders γ KCs less accessible to photolabelling. Most α/βKCs, but not the α/β and α’β’ KCs, express Fruitless (Fru). An additional 100 γ KCs were targeted for photoactivation in flies expressing PA-GFP under the control of the Fru promoter (Caron, 2013).

Texas red dextran was sequentially electroporated into different claws of a photolabelled KC. It is technically difficult to fill all the claws of a KC and on average 3 glomerular inputs were identified per KC. In fewer than 5% of the samples, the number of labelled PNs differed from the number of claws filled, reflecting either unsuccessful or imprecise electroporation. Samples with more labelled PNs than expected were discarded. The low frequency of unsuccessful claw fills indicates that claws extending from a given KC were filled with equal efficiency independent of size. Thus the size of a claw was not a selection criterion in these experiments (Caron, 2013).

A total of 683 inputs that synapse on 200 KCs were identified. 654 of these inputs connect to PNs innervating 49 of the 51 AL glomeruli. PNs innervating the DA3 and VL1 are absent from the data set but boutons were identified from these PNs in the MB calyx. 29 of the claws receive input from brain regions other than the AL. Interestingly, 11 of these claws are innervated by PNs that derive from pseudoglomeruli in the proximal antennal protocerebrum, a thermosensing centre in the fly brain that receives input from distinct heat- and cold-sensing neurons in the antennae. The remaining 18 PNs innervated different uncharacterized regions of the brain (Caron, 2013).

It was observe that the distribution of the glomerular inputs to KCs is not uniform. Inputs from the DA1 and DC3 glomeruli are most frequent, with each accounting for 5.1% of the total connections. The non-uniform distribution reflects the fact that the size and number of calycal boutons formed by PNs varies across glomeruli. For instance, the PNs associated with the DA1 and DC3 glomeruli form more numerous boutons in comparison with the PNs of less frequently represented glomeruli. It was also observed that there is a small but significant difference between the inputs to the α/β and γ KCs (p < 0.001). All subsequent statistical analyses were therefore performed separately on both the α/β and γ data sets, but failed to reveal any significant difference between the two data sets. Therefore, only the results obtained from the full data set are shown (Caron, 2013).

Statistical analyses of the 665 connections allowed searching for structure among the connections between glomeruli and KCs. First, it wae determined whether the KCs receiving input from a given glomerulus have a higher probability of receiving additional input from that same glomerulus. Of the 200 KCs in the data set, only 11 receive two inputs from the same glomerulus, and none receive three or more such inputs. It was determined whether the frequency of convergent input from a single glomerulus is significantly above or below chance expectations by randomly shuffling the connections in the data set between the different KCs, while maintaining the number of connections each of them receives. This shuffling maintains the frequency of glomerular connections observed in the experimental data, but eliminates any potential, non-random patterns of inputs onto individual KCs. This shuffling is used in all subsequent statistical analyses. The frequency of multiple connections from the same glomerulus in the observed and shuffled data sets is not significantly different. Thus, no KCs were observed that receive preferential inputs from a single glomerulus. Rather, individual KCs integrate information from multiple different glomeruli (Caron, 2013).

It was next determined whether KCs are connected to any preferential pair, trio, or quartet of glomeruli. Of the 1378 (53*52/2) different pairs of glomeruli that could converge onto an individual KC, 508 distinct pairs appear in the data set. 310 of these pairs connect to only one of the 200 KCs analysed, whereas certain pairs of glomeruli connect to multiple KCs. There are combinations of glomerular trios that connect with two different KCs, and one case in which two KCs receive inputs from the same quartet of glomeruli. However, the observed frequency with which the different pairs, trios and quartets converge onto different KCs is consistent with expectations from the shuffled data set. Thus, the identity of a glomerulus connected to a KC provides no predictive information as to the identity of the remaining glomerular inputs onto that neuron (Caron, 2013).

Glomeruli can be grouped based on biological properties shared by their associated OSNs (sensilla type, odour specificity) and PNs (developmental origin and topography of their axonal projections). KCs might receive preferential input from one or another of these glomerular categories. For example, the OSNs innervating the AL are derived from three sensillar types (basiconic, coeloconic, and trichoid sensilla) that project to three classes of glomeruli tuned to different odour categories. If individual KCs were tuned to a particular class of odours, they might preferentially integrate inputs from one type of sensillum. Statistical analyses, however, reveal that KCs that receive an input from one sensillar type are no more or less likely to receive additional inputs from this or any other type of sensillum than is predicted by chance. Sensillar type, however, provides only a coarse correlate of odour tuning. Therefore, glomeruli were first grouped based on the similarity of their odour response profiles and again no structure was observed in the inputs to a KC that correlated with odour tuning (Caron, 2013).

Glomeruli were classified on the basis of the properties of their PNs. PN axons from different glomeruli project to broad but stereotyped domains in the lateral horn and calyx of the MB. Input to an individual KC could be shaped by the topography of PN projections. Analysis of the distribution of inputs to a given KC, however, fails to reveal any preferential PN connectivity that reflects the organization of their projection in either the MB calyx or lateral horn. KCs do not preferentially integrate information from glomeruli innervated by PNs sharing a developmental origin. In addition, KCs do not select their input based on topographical constraints as suggested by a previous study. Finally, three glomeruli are innervated by Fru-expressing OSNs and PNs. Preferential pairing of inputs from Fru⁺ PNs onto individual KCs were not observed. Moreover, although most γ KCs express Fru, there is no preferential input from Fru⁺ glomeruli to γ KCs (Caron, 2013).

Next, an unbiased search was performed for structure by examining correlations within the connectivity matrix between the 53 glomeruli (51 AL glomeruli and 2 pseudoglomeruli) and the 200 KCs. Correlations were extracted by performing a principal component analysis of this matrix. This analysis failed to reveal structure in the input to KCs other than that inherent in the non-uniform distribution of glomerular inputs (Caron, 2013).

These data are consistent with a model in which each KC receives input from a combination of glomeruli randomly chosen from the non-uniform distribution of glomerular projections to the MB. Classification of either glomeruli or KCs on the basis of several shared developmental, anatomic, and functional features fails to reveal structured input onto individual KCs. Members of a given PN class do not preferentially converge onto an individual KC nor do members of a KC class receive specific and distinguishing PN inputs. A given KC can integrate information from glomeruli activated by food odours, pheromones, CO₂ and even temperature. Recent data suggests that the extrinsic output neurons of the MB that are responsible for the different forms of learned behaviour are anatomically segregated and synapse with KC axons within a specific MB lobe. Interestingly, similar glomerular inputs are observed for the KCs that innervate the different lobes of the MB. This random input to individual KCs provides a mechanism to contextualize a rich diversity of novel KC responses (Caron, 2013).

It is important to note that the tracing procedure that was developed only allows characterization of the inputs to a single KC per fly. It is therefore possible that the inputs to every KC are determined but this developmental program results in a distribution of glomerular inputs that appears random. However, it is difficult to conceive of a development mechanism that could dictate the identity of inputs to each of the seven claws of the 2000 KCs. Moreover, the logic of employing complex and unlikely identity codes to achieve an uncorrelated distribution of inputs is elusive. Indeed, a previous study examined the electrophysiological response of KCs to different odours in a line of flies that labels only 23 α/β neurons but failed to identify replicate KCs with shared odour response profiles. These observations support the conclusion that the complement of glomerular inputs to KCs differs in different individuals. In addition, it is not possible, from the analysis of the glomerular inputs to 200 KCs, to exclude the existence of small subsets of KCs that received determined inputs from the AL. Nonetheless, the data are most consistent with a model in which the majority of individual KCs receive input from a random collection of glomeruli, a finding with important implications for odour processing in the MB (Caron, 2013).

If the connections from AL to MB are indeed random, a given odour will activate a different ensemble of KCs in different flies. However, in an individual fly, a given odour will consistently activate the same ensemble. This representation must acquire valence through experience or unsupervised activity dependent plasticity to dictate an appropriate behavioural output. Uncorrelated glomerular input to KCs affords the fly with the ability to impart meaning to a diversity of novel and unpredictable sensory stimuli that it may encounter throughout its life. Plasticity at highly convergent synapses between KC axons and MB extrinsic neurons could mediate experience-dependent behavioural output, an elemental feature of MB function. Thus, the fly has evolved an olfactory circuit with a connectivity that optimizes its ability to contextualize and respond appropriately to a rich array of olfactory experiences (Caron, 2013).

The mechanisms by which this immature activity is generated, shaped, and terminated vary from system to system. In the embryonic OSNs, the transition from irregular spike-trains to continuous discharge may require the expression of olfactory receptors (OR), because in larvae mutant for the co-receptor Orco Or83b, necessary for OR function, this transition does not occur normally. Since Orco is expressed before the onset of spontaneous activity, it is suggested that the change in the pattern of OSN spontaneous activity is likely to be driven, at least in part, by the onset and level of expression of specific ORs. However, this might not be the only factor shaping spontaneous activity patterns over development, and other factors such as expression of other ion channels may also play a role. This might explain why 16 h AEL Orco mutants have indistinguishable levels of activity when compared with controls, yet the variability in their spike train is significantly increased (Prieto-Godino, 2012).

Previous studies have suggested that spontaneous activity is essential for the normal development of vertebrate OSNs, but that there is no such requirement in insects. However, this study found that there is a role for OSN activity in the development of the larval olfactory network. OSN activity regulates the morphology of OSN terminals independently of activity in neighbouring axons, and without activity terminals appear immature and occupy larger territories. This is similar to what has been described in zebrafish and mouse OSN terminals devoid of activity. There is also a report of a similar phenotype found in the AL of third instar Drosophila larvae after synaptic release was blocked in a large subset of OSNs. The results show that while immature terminal morphology is a cell autonomous phenotype that is independent of activity levels in neighbouring OSN axons, the expansion of OSN terminals is limited by interactions among the OSN terminals. Interestingly a similar process has been found to regulate the morphology and terminal expansion of retinotectal axons. Thus the control of axonal terminal extension via activity-dependent interactions may be a general process in the wiring of nervous systems. The nature of inter-axonal interactions that limit terminal growth remains unknown and is one example of how future work using amenable experimental systems such as the one provided by the larval olfactory network in Drosophila larvae may reveal general mechanisms operating during the assembly of neural circuitry (Prieto-Godino, 2012).

Animals' response to a stimulus in one sensory modality is usually influenced by other modalities. One important type of multisensory integration is the cross-modal modulation, in which one sensory modality modulates (typically inhibits) another. Identification of the mechanisms underlying cross-modal modulations is crucial for understanding how sensory inputs shape animals' perception and for understanding sensory processing disorders. However, the synaptic and circuit mechanisms that underlie cross-modal modulation are poorly understood. This is due to the difficulty of separating cross-modal modulation from multisensory integrations in neurons that receive excitatory inputs from two or more sensory modalities(5)-in which case it is unclear what the modulating or modulated modality is. This study reports a unique system for studying cross-modal modulation by taking advantage of the genetic resources in Drosophila. Gentle mechanical stimuli was shown to inhibit nociceptive responses in Drosophila larvae. Low-threshold mechanosensory neurons inhibit a key second-order neuron in the nociceptive pathway through metabotropic GABA receptors on nociceptor synaptic terminals. Strikingly, this cross-modal inhibition is only effective when nociceptor inputs are weak, thus serving as a gating mechanism for filtering out weak nociceptive inputs. These findings unveil a novel cross-modal gating mechanism for sensory pathways (Pan, 2023).

Evolution has generated an enormous variety of morphological, physiological, and behavioral traits in animals. How do behaviors evolve in different directions in species equipped with similar neurons and molecular components? This study adopted a comparative approach to investigate the similarities and differences of escape behaviors in response to noxious stimuli and their underlying neural circuits between closely related drosophilid species. Drosophilids show a wide range of escape behaviors in response to noxious cues, including escape crawling, stopping, head casting, and rolling. This study found that D. santomea, compared with its close relative D. melanogaster, shows a higher probability of rolling in response to noxious stimulation. To assess whether this behavioral difference could be attributed to differences in neural circuitry, focused ion beam-scanning electron microscope volumes of the ventral nerve cord of D. santomea were generated to reconstruct the downstream partners of mdIV, a nociceptive sensory neuron in D. melanogaster. Along with partner interneurons of mdVI (including Basin-2, a multisensory integration neuron necessary for rolling) previously identified in D. melanogaster, two additional partners of mdVI were identified in D. santomea. Finally, this study showed that joint activation of one of the partners (Basin-1) and a common partner (Basin-2) in D. melanogaster increased rolling probability, suggesting that the high rolling probability in D. santomea is mediated by the additional activation of Basin-1 by mdIV. These results provide a plausible mechanistic explanation for how closely related species exhibit quantitative differences in the likelihood of expressing the same behavior (Zhu, 2023).

To effectively control their bodies, animals rely on feedback from proprioceptive mechanosensory neurons. In the Drosophila leg, different proprioceptor subtypes monitor joint position, movement direction, and vibration. This study investigate how these diverse sensory signals are integrated by central proprioceptive circuits. Signals for leg joint position and directional movement converge in second-order neurons, revealing pathways for local feedback control of leg posture. Distinct populations of second-order neurons integrate tibia vibration signals across pairs of legs, suggesting a role in detecting external substrate vibration. In each pathway, the flow of sensory information is dynamically gated and sculpted by inhibition. Overall, these results reveal parallel pathways for processing of internal and external mechanosensory signals, which are proposed to mediate feedback control of leg movement and vibration sensing, respectively. The existence of a functional connectivity map also provides a resource for interpreting connectomic reconstruction of neural circuits for leg proprioception (Chen, 2021).

This study reports the anatomical structure and functional organization of second-order circuits for leg proprioception in Drosophila. Due to the lack of clear hierarchical structure within the VNC leg neuropil, it has been challenging to infer the flow of proprioceptive sensory signals with existing tools. Therefore genetic driver lines were generated that label specific subtypes of leg proprioceptors and classified candidate second-order neurons based on hemilineage identity. Optogenetics and calcium imaging were used to map the functional connectivity between leg proprioceptors and second-order neurons, followed by EM reconstruction to validate synaptic connectivity and in vivo calcium imaging to understand the function of second-order neurons during leg movement. Spatially targeted and subtype-specific optogenetic stimulation were used to analyze integration of FeCO signals within a subset of second-order neuron classes (Chen, 2021).

Overall, this work reveals the logic of sensory integration in second-order proprioceptive circuits: some populations of second-order neurons integrate tibia vibration signals across pairs of legs, suggesting a role for detection of external substrate vibration. Signals for leg joint position and directional movement converge in other second-order neurons, revealing pathways for local feedback control of leg posture. It is anticipated that this functional wiring diagram (see Summary diagram of circuits processing leg proprioceptive signals from the Drosophila FeCO, based on experiments in this study) will also help guide the interpretation of anatomical wiring diagrams determined through EM reconstruction of VNC circuits (Chen, 2021).

Proprioceptors in the Drosophila FeCO can be classified into three subtypes: club neurons encode bidirectional tibia movement and vibration frequency; claw neurons encode tibia position (flexion or extension); and hook neurons encode the direction of tibia movement.19 Our results show the existence of two distinct central pathways for processing signals from club vs. claw and hook neurons (see Summary diagram of circuits processing leg proprioceptive signals from the Drosophila FeCO, based on experiments in this study). It is proposed that neurons downstream of the club mediate sensing of small mechanical vibrations in the external environment, whereas neurons downstream of the claw and hook provide proprioceptive feedback to motor circuits for controlling the posture and movement of the legs. This division of central pathways for external and internal sensing may be a common motif across limbed animals. Work in a variety of species, including a recent study in mice, has found that many animals can detect low-amplitude, high-frequency, substrate-borne vibrations (Chen, 2021).

Flies may use vibration sensing to monitor acoustic signals in the environment, such as during courtship behavior, or to detect approaching threats. The distinct anatomical organization of neurons downstream of the club vs claw and hook also supports a segregation of vibration sensing and motor control feedback pathways. 9Ba and 10Ba neurons arborize in the mVAC, a common target of descending neuron axons. In contrast, 13Bb arborize in the IntNp, which contains the dendritic branches of the leg motor neurons and premotor interneurons. Based on these differences, it was hypothesized that vibration-sensing neurons interact with ascending and descending signals to and from the brain, whereas neurons downstream of hook and claw axons contribute to local motor control through direct or indirect connections to motor neurons. Leg motor neurons receive position- and movement-tuned proprioceptive input, consistent with feedback from claw and hook neurons (Chen, 2021).

Additional connectomic reconstruction is needed to determine which second-order neurons mediate these feedback connections, but 13Bb neurons are promising pre-motor candidates. VNC neurons postsynaptic to claw and hook axons receive only local input, from individual legs. In contrast, second-order neurons postsynaptic to club axons integrate signals across multiple legs. For example, GABAergic 9Bb neurons pool information from left and right legs in a single VNC segment, whereas cholinergic 10Ba neurons receive convergent input from left and right legs across different segments. Integrating club input across legs may improve detection of external vibration signals, while proprioceptive signals from the claw and hook may be initially processed in parallel to support postural control of individual legs. Bilateral integration also occurs in second-order auditory circuits downstream of the Drosophila Johnston's organ: mechanosensory signals from the two antennae are processed in parallel by second-order neurons in the AMMC but then converge in third-order neurons in the wedge (Chen, 2021).

Although second-order neurons in the vibration pathway integrate club signals across legs, multiple classes of second-order neurons in the motor pathway (13Bb and 8Aa) integrate signals across different FeCO subtypes from the same leg. Using new genetic driver lines that subdivide claw neurons into extension- and flexion-tuned subtypes, it was found that extension-tuned claw and hook neurons converge on 13Bb neurons. These cells are hypothesized to mediate resistance reflexes that stabilize tibia position in response to external perturbations. Prior work in the stick insect has shown that tibia resistance reflexes rely on position and directional movement signals from the FeCO (Chen, 2021).

In support of this hypothesis, another class of neurons in the 13B hemilineage, 13Ba, also encode tibia extension and drive tibia flexion when optogenetically activated. This functional connectivity map reveals interesting parallels with sensorimotor circuitry in the larval Drosophila VNC. Although fly larvae do not have legs, they do possess body wall proprioceptors (class I sensory neurons) and use chordotonal neurons to sense external vibrations in a manner analogous to club neurons in the adult FeCO. As in the adult, larval neurons belonging to lineage 9 (basin neurons) and lineage 8 (eve lateral interneurons) integrate signals from chordotonal sensory neurons and proprioceptors. These examples suggest that some lineage connectivity motifs are likely conserved across the larval and adult nervous systems, which are already known to possess molecular and general anatomical similarities (Chen, 2021).

The results identify a prominent role for inhibition in central processing of proprioceptive information from the FeCO. Of the eight identified second-order cell classes, six are putative inhibitory neurons (i.e., release GABA or glutamate). In other sensory circuits, local inhibitory processing contributes to sharpening spatial and temporal dynamics as well as reducing sensory noise through crossover inhibition (Chen, 2021).

By pharmacologically blocking GABAa and GluCl receptors, we identified a role for inhibition in controlling adaptation within second-order neurons (e.g., 10Ba and 13Bb neurons; Figure 5C). In other cases (20/22Ab or 9Ba; Figure 5A), inhibition was strong enough to completely mask proprioceptive inputs from FeCO axons. We hypothesize that this inhibition may be tuned in certain behavioral contexts, for example, during active movements, to gate the flow of proprioceptive feedback signals in a context-dependent manner. Synaptic transmission in Drosophila can be mediated by chemical synapses, which can be visualized with EM, or electrical gap junctions, which are not typically identifiable at the resolution of current EM volumes. FeCO neurons release acetylcholine but also express gap junction proteins (shakB; data not shown). We therefore used pharmacology to test for the presence of gap junctions between sensory and central neurons. MLA, an effective antagonist of nicotinic acetylcholine receptors in Drosophila eliminated functional connectivity between club and 9Bb neurons but only reduced functional connectivity between club and 10Ba neurons. We observed similar results downstream of the claw: MLA blocked functional connectivity between claw and 13Bb neurons but only reduced functional connectivity between claw and 13Ba neurons (data not shown). These results suggest that second-order proprioceptive circuits receive mixed chemical and electrical input from FeCO neurons. More work is needed to confirm these observations and to investigate the functional significance of why pathways might use one means of signal transmission over the other. One hypothesis is that chemical synapses exhibit adaptation (e.g., synaptic depression), whereas electrical synapses may be more advantageous for sustained synaptic transmission (Chen, 2021).

Each may provide different advantages for pathways that control behavior on a variety of timescales, from slow postural reflexes to rapid escape behaviors (Chen, 2021).

Central processing of sensory signals from the FeCO has been previously studied in other insects, especially the locust and stick insect. In both species, second-order interneurons encode combinations of tibia movement and position and also integrate multimodal signals from different proprioceptive organs (Chen, 2021).

Vibration signals detected by the FeCO appear to be processed by largely segregated populations of VNC interneurons. However, these conclusions were based on mapping of sensory receptive fields, and it was not previously possible to identify specific sources of sensory input, as is done in this study. Overall, comparison of our functional connectivity results in Drosophila with the prior work in other insect species suggests general evolutionary conservation of VNC circuits for leg proprioception and motor control. Although it is currently difficult to identify homologous cell types across insect species, future efforts could leverage conserved developmental programs: the organization of neuroblasts that give rise to the VNC is similar across insect species separated by 350 Ma of evolution (Chen, 2021).

This is an important advantage of using developmental lineages to define VNC cell classes-locusts and stick insects also possess 9A, 10B, and 13B neurons, which could someday be identified based on molecular markers of lineage identity (Chen, 2021).

The functional connectivity approach that was employed in this study has both benefits and drawbacks. On the positive side, it allowed screening a large connectivity matrix of genetically identified sensory and central neurons. Compared to other methods for anatomical mapping (e.g., EM), the use of optogenetics and calcium imaging allowed measuring of connection strength and dynamics across multiple individuals. It was found that second-order VNC neurons varied significantly in their functional connectivity strength and temporal dynamics. 5-fold differences were observed in peak calcium signals in response to optogenetic stimulation with the same light intensity. This range could be due to differences in GCaMP expression or intracellular calcium buffering, but could also reflect differences in synaptic strength between pre- and postsynaptic partners (Chen, 2021).

One limitation of functional connectivity is that it is not possible to measure all possible combinations of pre- and postsynaptic partners. For example, a previous study provided evidence that 9Aa neurons receive input from hook and club neurons, which was not observed in the current screen. This discrepancy could be due to the fact that the driver lines that were used do not label the specific subset of hook and club cells presynaptic to 9Aa neurons. Alternatively, it may be due to differences in signal transmission driven by optogenetic stimulation versus natural tibia movements, as was the case for 9Bb neurons (Chen, 2021).

Functional connectivity mapping also cannot resolve whether inputs are direct, due to the slow kinetics of GCaMP6. This study therefore used sparse, targeted EM tracing to validate some of the functional connections that were identified between FeCO and VNC neurons. A more detailed comparison of functional and anatomical connectivity will require dense, comprehensive reconstruction of the VNC neuropil. Automated reconstruction and manual proofreading have recently led to draft wiring diagrams of neural circuits in the adult Drosophila central brain (Chen, 2021).

Nervous systems have the ability to select appropriate actions and action sequences in response to sensory cues. The circuit mechanisms by which nervous systems achieve choice, stability and transitions between behaviors are still incompletely understood. To identify neurons and brain areas involved in controlling these processes, a large-scale neuronal inactivation screen was combined with automated action detection in response to a mechanosensory cue in Drosophila larva. Behaviors were analyzed from 2.9x105 larvae, and 66 candidate lines were identified for mechanosensory responses out of which 25 for competitive interactions between actions. The neurons in these lines were further characterized in detail and their connectivity was analyzed using electron microscopy. The neurons in the mechanosensory network were found to be located in different regions of the nervous system consistent with a distributed model of sensorimotor decision-making. These findings provide the basis for understanding how selection and transition between behaviors are controlled by the nervous system (Masson, 2020).

In order to identify neurons and brain regions underlying competitive interactions and transitions between actions during mechanosensory responses, a high-throughput inactivation screen was performed where individual neurons and groups of neurons were silenced (using tetanus-toxin) in 567 genetic GAL4 lines in Drosophila larva, and the effects of these manipulations on larval behavioral responses to a mechanosensory cue were examined (Masson, 2020).

The behavioral response of wild-type larvae to the stimulus (air-puff) were characterized and larvae were found to perform a probabilistic sequence of five different actions. An automated approach was developed and used that detects and distinguishes five different discrete behaviors that larvae perform in response to the air-puff. Evidences suggest that the discrete action description is relevant when compared to a continuum approach as parameters associated to larva dynamics tend to naturally cluster. The representation is found to be stable even for large number of larvae while their characteristics (amplitude of actions, duration, size of the larva, shape etc.) can vary significantly. Yet it is pointed out that it does not mean that all behaviors and actions that larvae exhibit are necessarily described as only discrete actions (Masson, 2020).

This analysis was used to describe phenotypes that result from manipulation of different populations of neurons or single neuron types. Phenotypes were found that are consistent with a specific role of neurons in sensory processing or motor control, competitive interactions, and sequence transitions. Neuronal expression data for all of the GAL4 lines used in this screen have been previously published). The number of neurons that were targeted in the tested lines varies from 1 to 7 pairs on average and smaller number of the GAL4 lines the driver is restricted to a single neuron type. The morphology of top hits were analyzed in more detail using single-cell FLP-out and their connectivity was analyzed using electron-microscopy reconstruction (see Putative pathways in the mechanosensory network)(Masson, 2020).

A framework was developed for selectively identifying circuit elements underlying competitive interactions and sequence transitions. Sensory-processing, sensorimotor decisions, and sequence generation are intertwined processes as the latter two will depend on how the sensory information is processed, and the sequence production mechanistically might depend on competitive interaction between distinct actions as suggested by models of sequence generation like competitive queuing or chains of disinhibitory loops. Nevertheless, the reasoning described below was used to identify neurons selectively involved in competitive interactions that underlie sensorimotor decisions and sequence generation (Masson, 2020).

It was reasoned that, if the stimulus cannot be processed and thus perceived accurately the animals might respond less, by performing less of all or some of the actions. If the sensory processing is affected in the opposite way (hypersensitivity), animals might respond more, and perform more of all or some of the actions normally observed. Thus, the neurons that gave such inactivation phenotypes (of less of one or more actions; or more of one or more actions) could be involved in any aspect of sensory processing or motor control. It cannot be excluded that these larvae responded less because the inactivation of the neurons modulates the overall animal state (Masson, 2020).

However, inactivation of neurons involved in mediating competition between actions is expected to result in increased probability of one action and a decreased probability of one or more other actions (or the converse) as the neuron implementing the competitions will promote one action while inhibiting competing options. Based on this logic, 25 hits (GAL4 lines) were identified that were top candidates for selectively mediating affected competitive interactions. Morphologically the neurons in these lines were characterized using light microscopy of multicolor flip-out and for some of the neurons determined their connectivity by identifying them in the electron microscopy volume. It was found that some of these neurons received input from chordotonal sensory neurons, chordotonal related interneurons or multidendritic class III sensory neurons while others were pre-motor neurons. In addition, other neurons were found that project to or are located in the brain. Taken altogether, the GAL4 lines that were identified as hits drive in neurons that are located in the ventral nerve cord (both abdomen and thorax region), suboesophageal ganglion and brain. This suggests that the networks for competitive interactions between actions that occur in response to air-puff are distributed across the nervous system (Masson, 2020).

The idea that sensorimotor decisions are made 'through a distributed consensus that emerges in competitive populations' and that interactive behaviors require sensorimotor and selection system to function in parallel have emerged in various fields, but it has been challenging to elucidate the neuronal architecture that would implement such sensorimotor decisions. The Drosophila larva, because of its numerical simplicity, small size and the existence of multiple experimental approaches for structural and functional connectivity studies, behavioral genetics, optogenetics etc. is an ideal system for investigating how the outcomes of these competitive interactions at the different sites are integrated across the nervous system to give rise to coherent sensorimotor behaviors (Masson, 2020).

The neural architecture that controls the productions of probabilistic action sequences and establishes the order of the individual elements in the sequence is also poorly understood. This study identified a number of hit line phenotypes that were consistent with an implication of the neurons in ensuring proper ordering of individual elements in the sequence. For example, the neurons in the R45D11 line could be inhibiting reversals from Bend to Crawl and promoting transitions from Bend to Back, while neurons in the R69E06 line could be promoting transitions from Bend to Back-up while preventing reversals from Bend to Hunch. In previous work on a two- element Hunch-Bend sequence in response to an air-puff, it was proposed that transitions to the next element in the sequence and reversal to the previous element are controlled through two different motifs: lateral disinhibition from the neuron driving one behavior onto the neuron driving the following behavior and feedback disinhibition that provides a positive feedback that stabilizes the second behavior and prevents reversals back onto the previous actions (Jovanic, 2016).

It was speculated that chains of such disinhibitory loops could be a general mechanism for generating longer behavioral sequences. In the case of longer sequences (more than two elements) the maintenance of a selected action (through a positive feedback) after the transition from the previous action has occurred would need to be balanced with promoting the transition from the current onto the following action in the sequence. The candidate neurons in the R45D11 and the R69E06 could represent a starting point for investigating these mechanisms as their phenotype are consistent with preventing reversals from Bend to Hunch and Crawl and promoting transitions from Bend to Back-up (that represent nearly 80% of transitions from Bend). Another category of phenotypes, increase in transitions from Hunch to Back-up and decreased from Hunch to Bend, suggests that asymmetric competitive interactions exist between transitions to Bend and Back-up (from Hunches) where the transitions from Hunch to Bend inhibit transitions from Hunch to Back-up but not the other way around. Such a mechanism would allow a progression of a sequence in a probabilistic way where the transitions from Hunch to Bend are more likely (50%) than to Back-up (less than 15%) (Masson, 2020).

In summary, this screen provides a roadmap for investigating the neural circuit mechanisms underlying the different computations during mechanosensory responses. It also offers a starting point for identifying the mechanisms underlying the competitive interactions between behaviors as well as the transition between individual actions in probabilistic sequences across the nervous system. While the number of neurons that were targeted in the tested lines varies from one to seven pairs on average, and sometimes more, in the case when the lines label multiple neuron types, intersectional strategies can be used to further refine the expression patterns. In the larva, a volume of electron microscope data has been acquired and more than 60% of the nervous system has been reconstructed through collaborative efforts. The synaptic partners of the identified candidate neurons can therefore be further reconstructed in the electron microscopy volume. Combined with EM reconstruction, physiology, and modeling the candidate lists of neurons can be used to relate circuit structure and function across the nervous system and unravel the principles of how the nervous system selects actions and produces action sequences in response to external stimuli (Masson, 2020).

Proprioceptors provide feedback about body position that is essential for coordinated movement. Proprioceptive sensing of the position of rigid joints has been described in detail in several systems; however, it is not known how animals with a flexible skeleton encode their body positions. Understanding how diverse larval body positions are dynamically encoded requires knowledge of proprioceptor activity patterns in vivo during natural movement. This study used high-speed volumetric swept confocally aligned planar excitation (SCAPE) microscopy in crawling Drosophila larvae to simultaneously track the position, deformation, and intracellular calcium activity of their multidendritic proprioceptors. Most proprioceptive neurons were found to activate during segment contraction, although one subtype was activated by extension. During cycles of segment contraction and extension, different proprioceptor types exhibited sequential activity, providing a continuum of position encoding during all phases of crawling. This sequential activity was related to the dynamics of each neuron's terminal processes, and could endow each proprioceptor with a specific role in monitoring different aspects of body-wall deformation. This study demonstrates this deformation encoding both during progression of contraction waves during locomotion as well as during less stereotyped, asymmetric exploration behavior. The results provide powerful new insights into the body-wide neuronal dynamics of the proprioceptive system in crawling Drosophila, and demonstrate the utility of the SCAPE microscopy approach for characterization of neural encoding throughout the nervous system of a freely behaving animal (Vaadia, 2019).

This study demonstrates a new approach for live volumetric imaging of sensory activity in behaving animals, leveraging an optimized form of high-speed SCAPE microscopy. This methodology was used to examine the activity patterns of a heterogeneous collection of proprioceptive neurons during crawling, as well as during more complex movements such as head turning and retraction, to determine how larvae sense body-shape dynamics. Imaging revealed 3D distortion of proprioceptive dendrites during movement and GCaMP activity that occurred coincident with dendritic deformations. It is noted that the results are consistent with a complementary study (He, 2019), which examined ddaD and ddaE dorsal proprioceptors and also demonstrated increased activity during dendrite folding. The He study elucidated that this deformation-dependent signaling is reliant on the mechanosensory channel TMC (Vaadia, 2019).

This survey of the full set of hypothesized multidendritic proprioceptors in behaving larvae revealed that most neurons (all class I neurons, dmd1, and vbd) increase activity during segment contraction. By contrast, dbd neurons showed increased activity during segment stretch, which is consistent with previous electrophysiological recordings of dbd in a dissected preparation. The temporal precision afforded by high-speed SCAPE microscopy further revealed that different proprioceptors exhibit sequential onset of activity during forward crawling. Timing of activity was associated with distinct dendrite morphologies and movement dynamics, suggesting that proprioceptors monitor different features of segment deformation. The complementary sensing of segment contraction versus stretch in class I, dmd1, and vbd versus dbd neurons provides an additional measure of movement that is conceptually similar to the responses of Golgi tendon organs versus muscle spindles in mammals. Combined, these results indicate that this set of proprioceptors function together to provide a continuum of sensory feedback describing the diverse 3D dynamics of the larval body (Vaadia, 2019).

Prior work suggested that the proprioceptors analyzed in this study have partially redundant functions during forward crawling because silencing different subsets caused similar behavioral deficits, namely slower crawling, whereas silencing both subsets had a more severe effect. Slow locomotion may be a common outcome in a larva that is lacking in part of its sensory feedback circuit, yet the results suggest that each cell type has a unique role. The demonstration of the varying activity dynamics of proprioceptors during crawling and more complex movements indicates that diverse sensory information is available to the larva, and suggests that feedback from a combination of these sensors could be used to infer aspects of speed, angle, restraint, and overall body deformation. This feedback system is likely to be important for a wide range of complex behaviors, such as body bending and nociceptive escape (Vaadia, 2019).

How can an understanding of proprioceptor activity patterns inform models of sensory feedback during locomotion? Electron microscopic reconstruction has shown that ddaD, vbd, and dmd1 proprioceptors synapse onto inhibitory premotor neurons (period-positive median segmental interneurons, A02b), which promote segment relaxation and anterior wave propagation. Thus, activity of these sensory neurons may signal successful segment contraction and promote forward locomotion, in part by promoting segment relaxation. Furthermore, vpda neurons provide input onto excitatory premotor neurons A27h, which acts through GABAergic dorsolateral (GDL) interneurons to inhibit contraction in neighboring anterior segments, thereby preventing premature wave propagation. In this way, vpda feedback could contribute to proper timing of contraction in anterior segments during forward crawling. In contrast to other proprioceptors, dbd neurons are active during segment stretch. Their connectivity also tends to segregate from contraction-sensing neurons, and understanding how the timing of this input promotes wave propagation is an important future question. This study's dynamic recordings of the function of these neurons during not just crawling but also exploration behavior provide essential new boundary data for testing putative network models derived from this anatomical roadmap (Vaadia, 2019).

SCAPE's high-speed 3D imaging capabilities enabled 10 VPS imaging of larvae during rapid locomotion. Fast volumetric imaging not only prevented motion artifacts but also revealed both the 3D motion dynamics and cellular activity associated with crawling behavior. SCAPE's large, 1-mm-wide field of view allowed multiple cells along the larva to be monitored at once, while providing sufficient resolution to identify individual dendrite branches. Because SCAPE data are truly 3D, dynamics could be examined in any section or view. Additionally, fast two-color imaging enabled simultaneous 3D tracking of cells, monitoring of GCaMP activity, and correction for motion-related intensity effects. The demonstration that larvae that are compressed during crawling exhibit altered dendrite deformation, and thus altered proprioceptive signaling, underscores the benefit of being able to image unconstrained larvae, volumetrically, in real time. Furthermore, rapid volumetric imaging allowed for the analysis of sensory responses during non-stereotyped, exploratory head movements in 3 dimensions, revealing activity patterns that could be utilized for encoding of complex, simultaneous movements. This finding also demonstrates the quantitative nature of SCAPE data and its high signal to noise, which enabled real-time imaging of neural responses without averaging from multiple neurons (Vaadia, 2019).

This study provides an example of how high-resolution, high-speed volumetric imaging enabled investigation of the previously intractable question of how different types of proprioceptive neurons encode forward locomotion and exploration behavior during naturalistic movement. Imaging could readily be extended to explore a wider range of locomotor behaviors such as escape behavior, in addition to other sensory modalities such as gustation and olfaction. Detectable signals reveal rich details including the firing dynamics of dendrites and axonal projections during crawling. Waves of activity in central neurons within the ventral nerve cord can also be observed. It is expected that the in vivo SCAPE microscopy platform utilized in this study could ultimately allow complete activity mapping of sensory activity during naturalistic behaviors throughout the larval CNS. Using SCAPE, it is conceivable to assess how activity from proprioceptive neurons modulates central circuits that execute motor outputs, which will provide critical information for a dissection of the neural control of behavior with whole-animal resolution (Vaadia, 2019).

Sensory stimuli with graded intensities often lead to yes-or-no decisions on whether to respond to the stimuli. How this graded-to-binary conversion is implemented in the central nervous system (CNS) remains poorly understood. This study shows that graded encodings of noxious stimuli are categorized in a decision-associated CNS region in the ventral cord of Drosophila larvae, and then decoded by a group of peptidergic neurons for executing binary escape decisions. GABAergic inhibition gates weak nociceptive encodings from being decoded, whereas escalated amplification through the recruitment of second-order neurons boosts nociceptive encodings at intermediate intensities. These two modulations increase the detection accuracy by reducing responses to negligible stimuli whereas enhancing responses to intense stimuli. These findings thus unravel a circuit mechanism that underlies accurate detection of harmful stimuli (Hu, 2020).

This study identified a neural network that categorizes noxious stimuli of graded intensities to generate binary escape decisions in Drosophila larvae, and a gated amplification mechanism was unraveled that underlies such binary categorization. In responding to the noxious stimuli, whereas failure in prompt responses may cause harm, excessive escape responses to negligible stimuli would lead to the loss of resources for survival. The gated amplification mechanism could reduce the responses to negligible stimuli whereas enhancing the responses to intense stimuli. In this way, the accuracy in deciding whether to escape from the stimuli is enhanced (Hu, 2020).

Information processing in the nervous system is affected by noise, which may be embedded in external sensory stimuli (e.g., sensory noise) or generated within the nervous system (e.g., electric noise). A recent study in C. elegans shows that activation mediated by electrical synapses and disinhibition mediated by glutamatergic chemical synapses form an AND logic gate to integrate the presentation of the salience of attractive odors. The AND-gate computation in worm AIA interneurons requires multiple sensory neurons to report the presence of attractive odors and, consequently, filters out the noise embedded in the sensory stimuli. Another study on the olfactory system of adult Drosophila reported a mechanism to address the noise that is produced within the nervous system. A three-layered feedforward network averages the noise to enhance peak detection accuracy and then uses coincidence detection to distinguish real signals arrived synchronously from noise caused by spontaneous neural activities. In the nervous system, the noise can be produced at each stage of the sensori-motor transformation. Compared with the two mechanisms mentioned above, which filter out the existing noise, the graded-to-binary conversion through the gated amplification mechanism reported in this study makes the converted signals less vulnerable to the noise produced at later stages of sensori-motor transformation. This is because after the graded signals become binary, the signals are more separated (either suppressed or amplified) according to stimulus intensities and, consequently, the same level of noise is less likely to cause the binary signals to falsely pass the decision threshold than the graded ones. As a result, the ambiguous encoding range of the stimulus intensity is narrowed and the frequency of false decisions is reduced, as demonstrated by computational modeling (Hu, 2020).

Thresholding of gradually accumulated sensory evidence has been considered to be fundamental for generating yes-or-no decisions. For example, a recent study in mammals has shown that visual evidence of danger can be gradually accumulated by recurrent circuits to overcome the threshold for escape behaviors. Such a mechanism takes time to build up decision-associated activities for decisions with higher accuracy, which leads to the well-known speed-accuracy trade-off in perceptual decision making. However, the current findings add a new dimension to the processing of sensory evidence for perceptual decision making: different from recurrent networks, the recruitment of a number of SONs can instantaneously boost the decision-associated activity to reach the decision threshold, which ensures decision speed. Because the gated amplification mechanism reported here also ensures the detection accuracy, such a mechanism might bypass the speed-accuracy trade-off in sensory signal detection (Hu, 2020).

An electron microscopy connectome study reported 13 types of second-order neurons (SONs) in the Drosophila larval nociceptive system, each of which has distinct connectivity and functions. For example, Basin-4, DnB, and Wave neurons also receive mechanosensory inputs, whereas A08n does not. Moreover, Wave neurons detect stimulus positions on larval body walls. Furthermore, serotonergic modulation acts on this network during development to adjust the nociceptive responses, providing a mechanism for larvae to adjust the escape threshold according to their developmental environment. However, because at least 5 types of SONs are both required and sufficient for larval escape behaviors, it remains a mystery why there exist so many seemingly redundant neurons at the same level in the network. The nociceptive system is a dedicated protective system that responds to potential tissue-damaging insults, so both speed and accuracy of the perceptual decision-making process are important. This is probably why the nociceptive system uses an amplification network formed by a large number of SONs to dissociate time from accuracy in the perceptual decision-making process and avoid the trade-off between decision speed and accuracy (Hu, 2020).

Novel unbiased computational toolsets were developed for automatically analyzing the functional connectivity of all neural structures, including both somas and neurites in the larval VNC. Using these toolsets, a decision-associated CNS region, the PMC, was identified that covers the neuropil structure TP. The TP is concentrated with large amounts of neurites, especially those of peptidergic neurons. Although this anatomical structure was identified previously, its function is unknown. The finding of this study of its important function in sensori-motor transformation suggests that this region is possibly a hub for information exchange and integration. The detailed anatomical and functional connectivity of the TP could be a fascinating direction for future studies (Hu, 2020).

In summary, this study postulates a neural basis for converting graded sensory inputs to yes-or-no behavioral decisions. A previous study showed that neurons in the rat posterior parietal cortex encode a graded value of accumulating evidence whereas those in the prefrontal cortex have a more categorical encoding that indicates the decisions. Thus, the categorization of sensory evidence by making graded encodings binary in perceptual decision making is likely an evolutionarily conserved process. In this study, advantage was taken of the powerful genetic model Drosophila to unravel how such computation might be implemented at the cellular and molecular level. Finally, because whole-CNS functional imaging analysis is a key approach to decipher the neural basis for sensori-motor integration and perceptual decision making, it is anticipated that the computational tools developed in this study will facilitate investigations in these fields (Hu, 2020).

The mechanisms by which synaptic partners recognize each other and establish appropriate numbers of connections during embryonic development to form functional neural circuits are poorly understood. This study combined electron microscopy reconstruction, functional imaging of neural activity, and behavioral experiments to elucidate the roles of (1) partner identity, (2) location, and (3) activity in circuit assembly in the embryonic nerve cord of Drosophila. Postsynaptic partners were found to be able to find and connect to their presynaptic partners even when these have been shifted to ectopic locations or silenced. However, orderly positioning of axon terminals by positional cues and synaptic activity is required for appropriate numbers of connections between specific partners, for appropriate balance between excitatory and inhibitory connections, and for appropriate functional connectivity and behavior. This study reveals with unprecedented resolution the fine connectivity effects of multiple factors that work together to control the assembly of neural circuits (Valdes-Aleman, 2020).

The human nervous system is organized into circuits with specifically matched and tuned cell-to-cell connections essential for proper function. During development, neurons navigate through the nervous system to reach their target location. Surrounded by numerous cells along their trajectories and in their target areas, developing neurons ignore most cells and connect only to specific partners (Valdes-Aleman, 2020).

The absolute numbers of synapses between specific partners can vary across individuals, hemispheres, or repeated network modules in the same individual. However, recent electron microscopy (EM) reconstructions in multiple Drosophila larvae suggest that, at least in some circuits, the relative numbers of synapses between partners are precisely regulated. Thus, the fraction of inputs a neuron receives from a specific partner, relative to its total number of inputs, is remarkably conserved across individuals, across larval stages, and even between larva and adult. For example, the fraction of input varied by an average factor (fold change; i.e., the ratio of two fractions) of 1.07 ± 0.22 between different first instar larvae (n = 13 homologous connections) and 1.09 ± 0.20 from first to third instar (n = 12 homologous connections). Similarly, the average input a mushroom body output neuron receives from a modulatory neuron in the larva and adult is 3.4% and 3.3%, respectivelyh. These examples of conserved fractions of synaptic input across individuals and life stages raise several key questions: (1) How important are the precise numbers of connections between neurons for normal behavior? (2) How are the precise numbers of connections between partners specified? and (3) How is the appropriate balance between excitatory and inhibitory connections in the circuit achieved (Valdes-Aleman, 2020)?

The chemoaffinity hypothesis proposes that pre- and postsynaptic partners express specific matching combinations of cell surface molecules that enable them to seek out and recognize each other during development. However, relatively few examples of partner-recognition molecules have been identified, so it is unclear whether their use is a general principle or if they are used only in some systems. It is also unknown if these partner-recognition mechanisms specify precise numbers of synapses between partners, or only instruct two neurons to form synapses, but not how many (Valdes-Aleman, 2020).

Alternative hypotheses propose that neurons seek out specific locations in the nervous system, rather than specific partners, indiscriminately connecting to whichever neurons are present there. Consistent with this, neurons have been shown to use non-partner-derived positional cues, such as third-party guidepost cells or gradients of repellents, to select their termination and synaptogenesis area independently of their partners. Additionally, activity-dependent mechanisms are thought to refine connections through Hebbian and/or homeostatic plasticity mechanisms. Neurons that fire together preferentially wire together in many areas of the vertebrate nervous system through positive feedback. At the same time, homeostatic mechanisms restore activity toward a specific set point through negative feedback, imposing competition and preventing runaway excitation or complete silencing of the circuit. However, the extent to which activity modulates numbers versus the strength of existing synapses is still an open question (Valdes-Aleman, 2020).

These questions have been difficult to address because they require manipulating candidate factors that could influence connectivity, visualizing synapses between uniquely identified partners, and relating observed structural changes to effects on functional connectivity and behavior. This study therefore used the tractable Drosophila larva as a model system with the following advantages: (1) excellent genetic tools for selective manipulation of uniquely identified neurons; (2) a compact nervous system amenable to rapid imaging with synaptic resolution, and (3) a rich behavioral repertoire with well-established quantitative assay (Valdes-Aleman, 2020).

Recently, comprehensive synaptic-resolution connectivity maps of the circuitry downstream of the mechanosensory Chordotonal (hereafter 'mechanosensory') neurons and nociceptive multidendritic class IV (hereafter 'nociceptive') neurons in an abdominal segment of a first instar larva have been generated. Portions of this circuit were also reconstructed in two different abdominal segments (A1 and A3) of two different first instar individuals and at two different life stages: first (A1) and third instar (A3) (Valdes-Aleman, 2020).

This study selectively altered the location or activity of the mechanosensory neurons and generated new EM volumes of the manipulated samples to investigate the effects on connectivity. These anatomical studies were completed with functional connectivity and behavioral assays. This study reveals that proper location, partner identity, and activity are all required to achieve appropriate connectivity and behavior (Valdes-Aleman, 2020).

In some systems the position of pre- or postsynaptic terminals is specified by non-partner-derived positional cues. In other systems, molecules have been identified that mediate partner matching. However, it was unclear whether both mechanisms could operate in the same system and whether either mechanism specifies numbers of connections between partners (Valdes-Aleman, 2020).

Although developing sensory axons use non-partner-derived positional cues to select their final termination area in the Drosophila nerve cord, the current results suggest that position alone does not specify connectivity and that partner recognition also exists. When the location of sensory axons was altered, their postsynaptic partners extended ectopic branches and formed synaptic connections with them. The shifted axons did not gain any new strongly connected partners at their ectopic location, providing further evidence of remarkable partner selectivity. It is hard to imagine which cue, other than the mechanosensory axons themselves, instructed partner dendrites to form these ectopic branches and synapses. Nevertheless, the final proof of the existence of the partner-derived cues will be their identification in the future (Valdes-Aleman, 2020).

If partner-recognition molecules are sufficient for selective synaptogenesis irrespective of the location of partners, why is the precise location of sensory neuron axon terminals so tightly regulated by non-partner-derived positional cues? Despite partner neurons' connecting in ectopic locations, they did not establish appropriate numbers of synapses, resulting in defective responses to mechanosensory stimuli. This indicates that precise positioning of presynaptic mechanosensory axons is necessary for the formation of appropriate number of synapses (Valdes-Aleman, 2020).

It is not known why some partners received more synapses from shifted mechanosensory axons and others fewer than in wild-type. One possibility could be the involvement of third-party guidepost cells in synaptogenesis which would not be present in the aberrant location. Another speculation is that some neurons are better than others at finding their misplaced partners. Yet another possibility could be that shifting mechanosensory neurons initially resulted in fewer or weaker synaptic connections. This could have triggered compensatory homeostatic changes in the balance of excitation and inhibition within the circuit by increasing mechanosensory connections onto excitatory interneurons and reducing those onto inhibitory interneurons. This latter possibility could explain why similar connectivity effects were observed when sensory neurons were shifted and when they were inactivated during development (Valdes-Aleman, 2020).

Finally, in addition to changes in synapse numbers, silencing or shifting presynaptic partners could have also induced changes in synaptic strength and electrical properties (e.g., through changes in ion channel composition) that could account for some of the observed effects in behavior and functional connectivity. Furthermore, changes in the shapes of arbors could potentially affect electrical signal propagation. Future patch-clamp recordings following the same experimental manipulations could reveal the extent to which this occurs (Valdes-Aleman, 2020).

Activity plays a major role in refining the patterns of neuronal connections during development, especially in vertebrates. However, the effects induced within the network in response to selective silencing of specific neuron types are not fully understood (Valdes-Aleman, 2020).

The role activity plays in the development of the insect central nervous system is less clear. Some studies have shown that a lack of sensory activity during development does not affect neuron morphology or the capacity to form connections. Other studies have reported neural circuits can adapt their morphology, connectivity, or behavior in response to changes in developmental activity. However, a comprehensive synaptic-resolution analysis of the effects of silencing a specific neuron type on the numbers of connections between partners was lacking (Valdes-Aleman, 2020).

EM reconstructions revealed that silenced mechanosensory neurons connected to the appropriate partners, but with inappropriate numbers of synapses. Interestingly, excitatory multisensory interneurons (Basin) received a higher fraction of input from silenced mechanosensory neurons than in controls, while inhibitory interneurons (Ladder and Griddle) received a lower fraction. Selective silencing of mechanosensory neurons also increased input from a different sensory modality (nociceptive) onto Basin interneurons and decreased their input from inhibitory interneurons. This overall effect is similar to observations in the cortex, where sensory deprivation induces network-level homeostasis that alters the balance of excitation and inhibition. Synaptic scaling in the cortex is thought to be multiplicative, such that all excitatory connections onto an excitatory neuron are scaled equally when excitatory drive onto that neuron is reduced. In contrast, the inhibitory connections onto excitatory neurons are reduced. Although the majority of studies in the cortex focus on homeostatic plasticity of functional connections, this study demonstrated a drastic plasticity in the number of synaptic connections between partners. This apparent homeostasis of synapse numbers may follow similar multiplicative rules, because this study found that both mechanosensory and nociceptive inputs onto Basin interneurons were increased when mechanosensory neurons were silenced (Valdes-Aleman, 2020).

It was found that larvae with permanently silenced mechanosensory neurons not only had increased structural connections between nociceptive and Basin neurons but also stronger functional connections and behavioral responses to nociceptive stimuli. This structural and behavioral compensation is reminiscent of findings in mammals, in which if one sensory modality is removed, another modality is restructured and improved (Valdes-Aleman, 2020).

Interestingly, silencing mechanosensory neurons during development permanently decreased responses to mechanosensory stimuli, even days after restoring activity. This is also reminiscent of findings in mammals, in which deprivation of visual input during an early critical period permanently impairs vision. However, this result appears at odds with the increased structural and functional connections from silenced mechanosensory neurons onto the excitatory Basins. A possible explanation is the reduction of mechanosensory connections onto inhibitory neurons under the same conditions. Unlike nociceptive neurons, the mechanosensory neurons have more inhibitory than excitatory postsynaptic partners, and these inhibitory interneurons play a role in triggering mechanosensory behaviors through disinhibition. Silencing the mechanosensory neurons may therefore result in a permanent reduction in disinhibition in the circuit with permanent consequences on behavior (Valdes-Aleman, 2020).

In summary, although partner-recognition molecules can ensure neurons recognize and connect only with appropriate partners, they are not sufficient to robustly specify appropriate numbers of synapses. Conversely, although neither precise location of presynaptic terminals nor neuronal activity in presynaptic partners directly instructs partner specificity, both are crucial to achieve appropriate numbers of connections, appropriate strengths of functional connections, appropriate balance of excitation and inhibition, and appropriate behavior. This study reveals with unprecedented resolution how location, identity, and activity work together to give rise to appropriately wired neural circuits and appropriate behaviors (Valdes-Aleman, 2020).

Starvation enhances olfactory sensitivity that encourage animals to search for food. The molecular mechanisms that enable sensory neurons to remain flexible and adapt to a particular internal state remain poorly understood. The roles of GABA and insulin signaling in starvation-dependent modulation of olfactory sensory neuron (OSN) function was studied in the Drosophila larva. The GABAB-receptor and insulin-receptor play important roles during OSN modulation. Using an OSN-specific gene expression analysis, this study explored downstream targets of insulin signaling in OSNs. The results suggest that insulin and GABA signaling pathways interact within OSNs and modulate OSN function by impacting olfactory information processing. It was further shown that manipulating these signaling pathways specifically in the OSNs impact larval feeding behavior and its body weight. These results challenge the prevailing model of OSN modulation and highlight opportunities to better understand OSN modulation mechanisms and their relationship to animal physiology (Slankster, 2020).

Starvation increases olfactory sensitivity that enhances an animal's search for food. This has been shown in insects, worms, and mammals including humans. However, the mechanisms by which an animal's starved state modulates sensory neuron function remain poorly understood. Understanding of these mechanisms significantly improved in the last decade or so from studies that showed how neuromodulators enable changes in the gain of peripheral sensory inputs. The prevailing mechanistic model for olfactory sensory neuron (OSN) modulation by the animal's starved state is that during the animal's starved-state, lower insulin signaling frees production of the short neuropeptide F receptor (sNPFR1), which increases sNPF signaling. Higher sNPF signaling increases presynaptic facilitation of OSNs, which leads to enhanced responses to odors. Interestingly, insulin and neuropeptide Y (the mammalian ortholog of sNPF) signaling also feature in the vertebrate olfactory bulb (Slankster, 2020).

However, the above model is incomplete and several questions remain. For instance, the model does not account for the role of GABA signaling, which plays important roles during both starvation and olfactory behavior in flies and mammals. The model also does not account for interactions between GABA and insulin signaling pathways that are known to impact neuromodulation in both fly and mammalian systems: For instance, GABAB-Receptor (GABA_BR) mediates a GABA signal from fly brain interneurons, which may be involved in the inhibitory control of Drosophila insulin like peptide (DILP) production; In mammalian CNS neurons, insulin increases the expression of GABA_AR on the postsynaptic and dendritic membranes; GABA administration to humans resulted in a significant increase in circulating Insulin levels under both fasting and fed conditions. Finally, the model does not account for the ultimate targets of insulin/GABA/sNPF signaling that alter OSN sensitivity to odors and its function (Slankster, 2020).

The above questions are significant because the mechanisms driving neural circuit modulations are fundamental to understanding of how neural circuits support animal cognition and behavior. If these mechanisms are better understood, it would be possible to learn how flexibility and the ability to adapt to a particular internal state are built into the sensory circuit. Understanding the mechanisms by which the starved state of an animal modulates its olfactory sensitivity and thereby controls its food-search behavior is important for both olfactory and appetite research. Finally, this connection cannot be ignored in light of the obesity epidemic and the demonstration that obese adults have reduced olfactory sensitivity (Slankster, 2020).

This study builds upon the prevailing model and argue that GABA and insulin signaling pathways interact within OSNs to mediate starvation-dependent modulation of its function and that defects in these signaling pathways impact larval food-search and feeding behaviors, which in turn impact weight gain. The Drosophila larval system is used in this study to demonstrate evidence in support of this argument. Using larval behavior assays, this study shows that GABA_BR and insulin receptor (InR) are required for starvation dependent increases in larval olfactory behavior. Using a novel OSN-specific gene expression analysis, this study shows that insulin and GABA signaling pathways interact within OSNs and modulate OSN function by impacting odor reception, olfactory information processing, and neurotransmission. Finally, this study shows that manipulating these signaling pathways specifically in the OSNs impact larval feeding behavior and its body weight (Slankster, 2020).

Insulin and GABA signaling pathways interact within OSNs and likely modulate OSN function by impacting odor reception (Orco), olfactory information processing (Rut), and/or neurotransmission (Syt1). Defects in GABA/insulin signaling pathways impact the animal's feeding behavior and body weight. These findings suggest a hitherto unsuspected role for GABA signaling in starvation-dependent modulation of OSN function, a role that is likely downstream of insulin signaling. They also raise questions about how individual OSNs may be differentially modulated by the animal's starved state. Finally, these findings imply a potential relationship between nutrient sensing and animal physiology (Slankster, 2020).

GABA and insulin signaling play important roles during both starvation and olfactory behavior. While GABA signaling in different regions of the animal brain is known to mediate starvation-dependent behavior, its role in specific olfactory neurons during starvation is unclear. Similarly, insulin has long been considered as an important mediator of state dependent modulation of feeding behavior. However, its precise role in olfactory neurons during starvation is controversial. According to the prevailing model, insulin signaling decreases upon starvation. However, a previous study showed that there is a three-fold increase in DILP-6 (Drosophila Insulin like Peptide) mRNA expression in larval tissue including fat bodies upon starvation (Slaidina, 2009), which is inconsistent with the model described in this paper. While the significance of DILP-6 increase in larval tissue during starvation is as yet unclear, consistent with the prevailing model, this study shows that InR and DILP-6 expression in larval head samples decrease upon starvation (Slankster, 2020).

This study also shows that higher insulin signaling increases expression levels of GABA_BRs in OSNs. This result is in line with several other studies in flies and mammals that have suggested possible interactions between GABA signaling and insulin signaling in different regions of the brain. The most relevant example supporting the current observation is noted in mice where insulin increases the expression of GABAARs on the postsynaptic and dendritic membranes of CNS neurons. Other examples show how GABA signaling might influence insulin signaling. For instance, in flies, GABA signaling from interneurons has been shown to affect insulin signaling by regulating DILP production; In humans, GABA administration significantly increases circulating insulin levels under both fasting and fed conditions; In diabetic rodent models, combined oral administration of GABA and an anti-diabetic drug (Sitagliptin) promoted beta cell regeneration and reduced blood glucose levels. Overall, this study adds to this growing body of literature and strongly suggests that GABA and insulin signaling pathways interact within larval OSNs to mediate OSN modulation (Slankster, 2020).

It is noted that starvation enhanced larval attraction toward only a subset of the odors tested. A related question in the field is whether starvation enhances an animal's ability to detect food-odors or all odors. Studies are inconclusive so far. Some studies have shown that starvation enhances an animal's ability to detect both food-related odors and nonfood-related odors. While similar results have also been shown in humans, the findings regarding the relevance of odor to feeding are rather mixed. This study along with previous studies raise the possibility that starvation differentially modulates individual OSNs. Indeed, individual OSNs exhibit functional diversity that may lend them to differential modulation by the animal's internal state. This diversity may stem from heterogeneous GABA_BR levels on the terminals of individual OSNs that determine differential presynaptic gain control. It is reasonable to speculate that heterogeneous GABA_BR and/or InR levels in individual OSNs could contribute to differential modulation of OSNs by the animal's starved state, which in turn impacts behavior toward only a subset of odors (Slankster, 2020).

An inability to regulate sensitivity to food odors at appropriate times leads to irregular feeding habits, which in turn leads to weight gain. Obesity researchers will readily acknowledge that while several obvious risk factors for obesity (e.g., genetics, nutrition, metabolism, environment etc.) have been heavily researched, the relationship between nutrient sensing/sensory behavior and obesity remains grossly understudied. The present study sets the stage to further explore this relationship. Interestingly, several of the signaling molecules described in this study that play a role in OSN modulation have also been implicated in hyperphagia and obesity phenotypes. For instance, overexpression of sNPF in Drosophila and NPY injection in the hypothalamus of rats leads to increased food-intake and bigger and heavier phenotypes. Genetically obese rats have low levels of insulin in the brain including in the olfactory bulb and imbalanced insulin signaling via insulin receptors is associated with obesity phenotypes. Adenylyl cyclase (rut) deficient mice were found to be obese and both Adenylyl cyclase and Synaptotagmin have been targeted for anti-obesity drug development. These studies provide added significance to the observation that manipulating mechanisms mediating starvation-dependent modulation of OSNs impact feeding behavior and weight gain in larvae (Slankster, 2020).

Indeed, food odors can be powerful appetitive cues. A previous study showed that larvae engage in appetitive cue-driven feeding behavior and that this behavior required NPF signaling within dopaminergic neurons in higher-order olfactory processing centers (Wang, 2013). The current studies show that manipulating GABA_BR signaling in first-order OSNs impact appetitive cue-driven feeding behavior in larvae. While it remains to be seen whether parallel regulations during different stages of olfactory information processing impact feeding behavior, further studies are needed to reveal the mechanistic relationship between GABA_BR/InR signaling in OSNs, feeding behavior, and changes in body-weight (Slankster, 2020).

Based on the evidence so far, a motivating model is proposed for future investigations (see Model for OSN modulation). In this model, InR expressed on the terminals of larval OSNs act as sensors for the internal state of the animal. Its concerted activity with GABA_BR impacts OSN function either at the level of odor reception by affecting the expression of Orco or at the level of olfactory signal transduction by affecting the expression of Rut or at the level of neurotransmission by affecting the expression of Syt1 and sNPFR1. It is acknowledged that more exhaustive gene expression analyses are required to identify other molecular players downstream of InR and GABA_BR. It would also be valuable to investigate the relationship between InR expression levels on the terminals of individual OSNs and the sensitivity of individual OSNs to modulation by the animal's starved state (Slankster, 2020).

A valid concern in this study is that an innate attraction of larvae toward an odorant does not necessarily equate to food-search behavior. However, it is argued that attractiveness toward an odor source is a reliable measure of food-search behavior because an animal's ability to efficiently smell and move toward an odor source necessarily predicates most forms of such behavior. Another possibility to be considered is that changes in OSN sensitivity, food-search and/or feeding behaviors are independently regulated. For instance, it has been noted that starvation-induced hyperactivity in adult Drosophila was independently regulated from food consumption behavior in the flies. Blocking octopamine signaling in a small group of octopaminergic neurons located in the subesophageal zone (SEZ) of the fly brain neurons eliminated starvation induced hyperactivity but not the increase in food consumption. While such a possibility cannot be ruled out, the evidence presented in this study support the argument that starvation induced-changes in OSN function is related to the observed changes in food search and feeding behaviors. It is acknowledged that other studies have opted to keep larvae on sucrose with the intention of starving them of amino acids and other nutrients. So, the non-starved conditions in the present study actually represents partial starvation of macronutrients other than sugar. This was done to control the nutrient intake in the non-starved state with the intention of measuring the impact of individual macronutrients on OSN modulation in future studies. Finally, while this study tested the hypothesis that increases in body-weight of mutant genotypes are due to altered food consumption, alternate hypotheses that body-weight increases may be due to altered metabolism or increased fat accumulation haven not been tested(Slankster, 2020).

Overall, this study conducted in a simple, tractable, and highly conserved model system builds upon the prevailing model of starved-state dependent modulation of OSN function. It highlights and offers unique opportunities that are now possible to address the inadequate understanding of OSN modulation mechanisms at the resolution of single neurons, which in turn would lead to a better understand how flexibility and the ability to adapt to a particular internal state are built into the sensory circuit (Slankster, 2020).

Several features of the adult nervous systems develop in a "critical period," (CP) during which high levels of plasticity allow neural circuits to be tuned for optimal performance. Through an analysis of long-term olfactory habituation (LTH) in female Drosophila, this study provides new insight into mechanisms by which CPs are regulated in vivo LTH manifests as a persistently reduced behavioural response to an odorant encountered for four continuous days and occurs together with the growth of specific, odorant-responsive glomeruli in the antennal lobe. The CP for behavioral and structural plasticity induced by ethyl butyrate (EB) or carbon dioxide (CO(2)) closes within 48 hours after eclosion. The elaboration of excitatory projection neuron (PN) processes likely contribute to glomerular volume increases: both occur together and similarly require cAMP signalling in the antennal lobe inhibitory local interneurons (iLNs). Further, the CP for structural plasticity could be extended beyond 48 hours if EB- or CO(2)-responsive olfactory sensory neurons (OSNs) are silenced after eclosion; thus, OSN activity is required for closing the CP. Strikingly, silencing of glomerulus-selective OSNs extends the CP for structural plasticity only in respective target glomeruli. This indicates existence of a local, short-range mechanism for regulating CP closure (Chodankar, 2020).

Serotonin plays different roles across networks within the same sensory modality. Previous whole-cell electrophysiology in Drosophila has shown that serotonergic neurons innervating the first olfactory relay are inhibited by odorants. This study shows that network-spanning serotonergic neurons segregate information about stimulus features, odor intensity and identity, by using opposing coding schemes in different olfactory neuropil. A pair of serotonergic neurons (the CSDns) innervate the antennal lobe and lateral horn, which are first and second order neuropils. CSDn processes in the antennal lobe are inhibited by odors in an identity independent manner. In the lateral horn, CSDn processes are excited in an odor identity dependent manner. Using functional imaging, modeling, and EM reconstruction, it was demonstrated that antennal lobe derived inhibition arises from local GABAergic inputs and acts as a means of gain control on branch-specific inputs that the CSDns receive within the lateral horn (Zhang, 2019).

Habituation is the process that enables salience filtering, precipitating perceptual changes that alter the value of environmental stimuli. To discern the neuronal circuits underlying habituation to brief inconsequential stimuli, a novel olfactory habituation paradigm was developed, identifying two distinct phases of the response that engage distinct neuronal circuits. Responsiveness to the continuous odor stimulus is maintained initially, a phase termed habituation latency; it requires Rutabaga Adenylyl-Cyclase-depended neurotransmission from GABAergic Antennal Lobe Interneurons and activation of excitatory Projection Neurons (PNs) and the Mushroom Bodies. In contrast, habituation depends on the inhibitory PNs of the middle Antenno-Cerebral Track, requires inner Antenno-Cerebral Track PN activation and defines a temporally distinct phase. Collectively, these data support the involvement of Lateral Horn excitatory and inhibitory stimulation in habituation. These results provide essential cellular substrates for future analyses of the molecular mechanisms that govern the duration and transition between these distinct temporal habituation phases (Semelidou, 2018).

Drosophila is a premier system for molecular approaches to understand habituation because of its advanced molecular and classical genetics. In fact, it is a well-established model for habituation of various sensory modalities such as taste, vision, mechanosensory and escape responses, reflecting that habituation is apparent in most, if not all, circuits and modalities of the nervous system. However, in most of these paradigms, the circuits engaged to process the stimulus and establish the experimentally measured attenuated behavioral response are unclear. Importantly, the advanced understanding of the Drosophila olfactory circuitry and stimulus processing facilitates exploration of the mechanisms mediating decreased stimulus responsiveness and habituation to inconsequential odors. Such a recently described paradigm of olfactory habituation in Drosophila required 30 min of odor exposure and was mediated entirely by antennal lobe neurons. In contrast, habituation to repetitive 30 s odor pulses required functional Mushroom Bodies, neurons on the central brain also implicated in associative learning and memory in flies (Semelidou, 2018).

To resolve this paradox, this study focused on the early behavioral dynamics of habituation upon continuous odor stimulation. To that end, a novel habituation paradigm was developed and characterized to rather brief continuous odors. The behavioral responses define two distinct phases, an initial phase termed habituation latency, when stimulus responsiveness is maintained, which is followed by a significant response decrement reflecting habituation. Analogous response dynamics have been reported for footshock habituation. In addition, whether these phases engage and are mediated by distinct neuronal circuits was. The results highlight the stimulus duration-dependent activation of specific neuronal subsets and their distinct roles in securing timely habituation latency and habituation induction (Semelidou, 2018).

This study describes a novel olfactory habituation paradigm to brief odor stimuli and operationally defines two distinct phases in the response dynamics. The initial period of ~120 s is termed habituation latency and is characterized by maintenance of responsiveness to the odor. This is followed by manifestation of the habituated response, characterized behaviorally by attenuated osmotaxis. Focusing on the behavioral dynamics early in the process complements previous work olfactory habituation to continuous odor stimulation in Drosophila. A number of criteria differentiate these two paradigms from other types of habituation to olfactory stimuli as discussed below (Semelidou, 2018).

Drosophila habituate equally well to continuous or pulsed olfactory stimuli. This likely reflects the nature of olfactory stimuli, which typically are continuous rather than pulsed. On the other hand, habituation of the startle response to ethanol vapor may specifically require short (30 s) pulses due to its sedative properties and this may also be reflected by the rather long 6 min ITIs compared to the 15 s to 2.5-min intervals used herein for OCT. Short odor pulses are also required for the odor-mediated jump and flight response habituation, suggesting that pulsing may be necessary to evoke the startle response per se (Semelidou, 2018).

An important property shared with all habituation paradigms in Drosophila and other systems is spontaneous recovery of the response. This is another differentiating parameter among habituation paradigms in Drosophila. For the olfactory habituation paradigms, whereas 6 min suffice for spontaneous recovery after 4 and 30 min continuous odor exposure, 15-30 to surprisingly 60 min are required for recovery in the olfactory startle paradigms. Habituation to mechanosensory stimuli typically also requires shorter spontaneous recovery times, with habituation of the giant fiber-mediated jump-and-flight response requiring a mere 2 min and electric footshock habituation 6 min. Interestingly, other non-mechanosensory habituation paradigms require long spontaneous recovery periods with 30 min for habituation of the proboscis extension reflex (PER), and surprisingly, 2 hr for habituation of odor-induced leg response. It is posited that these differences reflect the engagement of distinct neuronal circuits mediating habituation to these diverse stimuli and the properties and connections of the neuronal types that comprise them (Semelidou, 2018).

Overall, these data suggest that latency and habituation to brief odor exposure involve modulation of lateral horn (LH) output, a neuropil innately encoding response valence to odor stimuli. It is propose dthat habituation latency involves processes that are not permissive to, or actively prevent stimulus devaluation. Latency duration depends on stimulus strength and is consistent with the notion that it is adaptive not to devalue strong, hence potentially important stimuli, expediently. In fact, it is posited that habituation latency serves to facilitate associations with concurrent stimuli, a requirement for associative learning. Shortened latency leading to premature habituation is predicted to compromise associative learning (Semelidou, 2018).

Importantly, maintaining responsiveness early upon odorant exposure requires activity of GABAergic inhibitory neurons, which are essential for lateral inhibition of antennal lobe glomeruli. LN activation appears to prevent saturation by strong continuous odors and hence reduce PN activity. Therefore, shortening habituation latency by blocking GABAergic neurotransmission in the antennal lobe may effectively reduce stimulus intensity, expediting habituation as suggested by the dilute odor experiments. This interpretation is further supported by the decreased habituation latency upon silencing the iACT PNs conveying olfactory signals to the MBs and the LH, but not by the mACT neurons innervating only the LH. Since iACT PNs are mainly excitatory, it appears that response maintenance requires excitatory signaling to the LH and the MBs (Semelidou, 2018).

All MB neuronal types except the γ, are essential for habituation latency. This suggests that at least part of the excitatory signal conveyed by the iACT PNs impinges upon the αβ and α' β' MB neurons, which is consistent with their role in associative learning and the proposal that habituation latency facilitates it. Neurotransmission from the MBs to LH neurons mediating aversive responses likely engages MB output neurons (MBONs), to maintain the valence and intensity of the odor and sustain aversion. Distinct MBONs are known to drive both attraction and aversion to odors and their potentially differential involvement in habituation is currently under investigation (Semelidou, 2018).

Dishabituation results in stimulus value recovery and apparently resets habituation latency. Clearly it requires neurotransmission via the GH146-marked neurons and MBs because silencing these neurons disables dishabituation, consistent with their role in response maintenance. These results lead to the hypothesis that dishabituating stimuli might converge on the MBs and/or iACT, possibly stimulating excitatory neurotransmission to the LH, to reinstate stimulus aversion. This hypothesis is currently under investigation as well (Semelidou, 2018).

In contrast, habituation requires prolonged or repeated exposure to the odorant and functional iACT and mACT PNs converging on the LH. Interestingly, the mainly GABAergic mACT PNs receive input both from the olfactory sensory neurons and the excitatory iACT PNs. Their depolarization also activates the excitatory iACT neurons via direct chemical synapses. This apparent feedback loop may be required for mACT activation after prolonged exposure to aversive odors, since these neurons were reported to respond mainly to attractive stimuli. It is proposed that prolonged aversive odor exposure enhances iACT activation, which in turn leads to habituation, while shorter exposure does not activate the iACT neurons, reflected by their dispensability for habituation latency. Importantly, the mACT innervates the LH downstream of the iACT PNs, providing feedforward inhibition. These characteristics likely underlie the necessary and sufficient role of mACT PNs in habituation upon 4-min odor stimulation. Collectively, these results are consistent with the proposal that mACT activation inhibits the innate LH-mediated avoidance response to the aversive odorant, establishing habituation (see A model of the neuronal subsets underlying (A) Habituation Latency and (B) Habituation, after exposure to aversive stimuli.). However, full mACT activation appears to also require iACT neurotransmission, which if abrogated eliminates habituation but is insufficient to establish it on its own (Semelidou, 2018).

Because the MZ699 Gal4 driver also marks ventrolateral protocerebrum (vlpr) neurons it is possible that they also play a role in habituation. In fact, vlpr neurons function in aversive odor responses, are activated by excitatory iACT PNs, inhibited by the inhibitory PNs, and are afferent to the LH. Thus, they could act in parallel or synergistically to mACT PNs to establish the habituated response. As no specific vlpr driver is available, it is impossible at the moment to address this possibility directly. Briefly then, the current collective results strongly suggest a novel role for the inhibitory PNs innervating the LH, and possibly vlpr neurons, in inhibition of the innate response and habituation. The kinetics of inhibitory projection neuron activation and their output on downstream neurons could serve as a measure of the duration of odor exposure. Upon prolonged exposure, these neurons mediate inhibition of odor avoidance, thus devaluing the stimulus (Semelidou, 2018).

Analysis of the neuronal subsets underlying habituation has focused on aversive odors. However, considering the neuronal clusters involved in the process, it would be relatively safe to assume that the results extend to attractive odor habituation as well. It is possible that the neuronal circuitry comprised of PNs, the LH and MBs may be mediating habituation independently of odor valence. However, specific neuronal clusters may differ in odor valance-dependent activation or inhibition of other circuit components with opposing effects on the behavioral readout. For example, inhibitory PNs (iPNs) mediate attraction by releasing GABA in the LH to inhibit avoidance. If inhibited themselves, the resultant attenuated attraction will likely drive a behavioral output of habituation to an attractive odor (Semelidou, 2018).

In accord with this notion, attractive and aversive odors are represented in different AL glomerular clustersand this valence-dependent organization is preserved into higher brain centers. In fact, the posterior-dorsal LH responds to attractive and its ventral complement to aversive odors, while third order neurons convey information from ventral LH to the vlpr and from the dorsal LH to the superior medial protocerebrum. This organization potentially reflects differential recruitment of these neuronal clusters in habituation to aversive and attractive odors. The circuits involved in habituation to attractive odors and their specific contribution to the process will be the focus of future work (Semelidou, 2018).

Although behaviorally there is significant osmotactic attenuation after both 4 and 30 min aversive odor exposure, the experiments suggest that these represent distinct types of olfactory habituation. Habituation after 4 min of odor exposure does not require the MBs, but rather the projection neurons innervating the LH. Habituation after 30 min of exposure is also independent of MB function, but appears to be entirely mediated by iLNs and reside within the AL. This clear difference suggests that the specific potentiation of inhibitory synapses shown to underlie habituation after 30 min of exposure is not necessary for habituation to the brief 4-min exposure. Additionally, while Rut is required within the iLNs during the latency period upon brief odor exposure, it is surprisingly required within the same neurons for habituation to long odor exposure. Therefore, Rut-driven activity within the iLNs yields opposing time-dependent behavioral outputs in accord with the abovementioned notion that the same circuit components may drive opposing outputs (Semelidou, 2018).

Furthermore, the fact that mechanosensory stimuli are not effective dishabituators after 30 min of odor exposure as they are after 4 min, augments the conclusion these are different types of olfactory habituation and suggests that distinct dishabituators likely recruit different neuronal subsets to modulate the habituated response. Such neuronal circuits and the effect of different dishabituators in response recovery are currently under investigatio (Semelidou, 2018).

Altogether, the results indicate different mechanisms for 4 min and 30 min habituation to aversive odors with the former mediated by the interaction between iPNs, ePNs and their targets in the LH, while the latter is based on the inhibition of ePNs by iLNs at the AL level. However, it is possible that the potentiated PN inhibition would decrease their output to the LH to drive reduced avoidance. This argues that the LH could be involved in the behavioral output indicating habituation after 30 min of OCT exposure as well. An AL-mediated reduction in the perceived intensity or valence of a chronically present odor probably serves an adaptive evolutionary role distinct from short exposure to the same stimulus. In fact, filtering away the chronic odor at the antenna, the first olfactory synaptic station, might facilitate evaluation of additional odors at higher order neurons such as the MBs or the LH (Semelidou, 2018).

Significantly, this interpretation is congruent with timescale habituation in mice, where short-timescale odor habituation is mGluR-dependent and mediated by the anterior piriform cortex while long-timescale habituation requires NMDAR and is mediated by the olfactory bulb. In addition, studies in mice, rats and primates have shown that habituation of the higher order neurons is faster and more prominent than in olfactory bulb neurons. Therefore, temporal and spatial principles for olfactory habituation appear broadly conserved between insects and mammals, despite their evolutionary distance (Semelidou, 2018).

Noxious stimuli trigger a stereotyped escape response in animals. In Drosophila larvae, class IV dendrite arborization (C4 da) sensory neurons in the peripheral nervous system are responsible for perception of multiple nociceptive modalities, including noxious heat and harsh mechanical stimulation, through distinct receptors. Silencing or ablation of C4 da neurons largely eliminates larval responses to noxious stimuli, whereas optogenetic activation of C4 da neurons is sufficient to provoke corkscrew-like rolling behavior similar to what is observed when larvae receive noxious stimuli, such as high temperature or harsh mechanical stimulation. How C4 da activation triggers the escape behavior in the circuit level is still incompletely understood. This study identified segmentally arrayed local interneurons (medial clusters of C4 da second-order interneurons [mCSIs]) in the ventral nerve cord that are necessary and sufficient to trigger rolling behavior. GFP reconstitution across synaptic partners (GRASP) analysis indicates that C4 da axons form synapses with mCSI dendrites. Optogenetic activation of mCSIs induces the rolling behavior, whereas silencing mCSIs reduces the probability of rolling behavior upon C4 da activation. Further anatomical and functional studies suggest that the C4 da-mCSI nociceptive circuit evokes rolling behavior at least in part through segmental nerve a (SNa) motor neurons. These findings thus uncover a local circuit that promotes escape behavior upon noxious stimuli in Drosophila larvae and provide mechanistic insights into how noxious stimuli are transduced into the stereotyped escape behavior in the circuit level (Yoshino, 2017).

Insects and mammals share similarities of neural organization underlying the perception of odors, taste, vision, sound, and gravity. This study observed that insect somatosensation also corresponds to that of mammals. In Drosophila, the projections of all the somatosensory neuron types to the insect's equivalent of the spinal cord segregated into modality-specific layers comparable to those in mammals. Some sensory neurons innervate the ventral brain directly to form modality-specific and topological somatosensory maps. Ascending interneurons with dendrites in matching layers of the nerve cord send axons that converge to respective brain regions. Pathways arising from leg somatosensory neurons encode distinct qualities of leg movement information and play different roles in ground detection. Establishment of the ground pattern and genetic tools for neuronal manipulation should provide the basis for elucidating the mechanisms underlying somatosensation (Tsubouchi, 2017).

Only three distinct types of sensory information are transmitted directly to the brain by primary neurons [leg gustatory sensilla (gs), chordotonal organs (co), and wing and haltere campaniform sensilla (cs)]. Such connection has also been reported in other insects, suggesting that this might be a general feature across insecta. Whereas only a small portion of leg co neurons project directly to the brain, most wing and haltere cs neurons innervate the brain; these cs neurons are known to detect various aspects of wing-beat force during flight to provide feedback control. Direct projections to the brain would be important for these neurons to enable fast transmission of information about rapidly changing sensory parameters during flight (Tsubouchi, 2017).

It was found that ground detection for wind-induced suppression of locomotion (WISL), which would require slower temporal resolution than flight control, is mediated by both direct and indirect pathways. Primary neurons and secondary interneurons of the same sensory modality tend to converge in specific subregions of the brain, forming modality-specific somatosensory representation. In spite of the similar axon trajectory in the brain, these neurons convey information about leg movement in different ways (Tsubouchi, 2017).

Interneurons associated with the leg co and es terminate in neighboring but different regions of the lateral brain, yet some of them have shared roles in WISL control. Because their signals are transmitted to distinct parts of the brain, yet-unidentified higher-order neurons in the brain should converge those signals to the motor control circuitry (Tsubouchi, 2017).

In this respect, it is important to note that most ascending secondary interneurons identified in this study have presynaptic output sites, not only in the brain but also in the VNC. Local circuitry in the leg neuropil is important for controlling leg movement. Those local neurons are likely candidates that receive output from the ascending interneurons, because axon terminals of sensory neurons hardly have postsynaptic sites. Similar local output has also been found in other sensory modalities; many olfactory and visual projection interneurons have collateral output synapses in the antennal lobe and optic lobes (Tsubouchi, 2017).

There are three pairs of leg neuropils. Among them, the foreleg neuropil has specialized arborization of the gs neurons that exist only in the foreleg. Other than this, no substantial differences of arborization patterns were found between the fore-, mid-, and hindleg neuropils (Tsubouchi, 2017).

The present results provide data for a systematic comparison of the insect somatosensory system with its mammalian counterparts. Insects and mammals share similarities of neural organization underlying the perception of odors, taste, vision, sound, and gravity, and the current data also reveal marked similarity for the mechanosensory system. In insects, some primary neurons project directly to distinct parts of the ventral and lateral brain, whereas others terminate within the VNC. Likewise, in mammals, some neurons project directly to the ventral brain at the medulla oblongata, whereas others terminate within the spinal cord. Modality-specific pathways tend to converge in different subregions of the medulla, as well as in the thalamus of the mammalian brain. Similarly, direct and indirect pathways tend to converge in common subregions of the insect brain, and neurons conveying information about different somatosensory modalities tend to terminate in different subregions. As in mammals, these subregions often lie adjacent to each other in certain parts of the brain; for example, the entire terminal arborizations of the leg co and es secondary interneurons are confined in a 40-μm-wide, 150-μm-tall cylindrical volume in the lateral brain (Tsubouchi, 2017).

Somatosensory signals are sent predominantly to the ipsilateral brain side in insects and contralateral in mammals. Considering that descending neurons tend to project ipsilaterally in insects but contralaterally in mammals, however, somatosensory signals and motor control computation are processed primarily in the same side of the brain in both cases (Tsubouchi, 2017).

Layers of sensory axon terminals in the insect VNC and mammalian spinal cord are also organized in a similar order. Insect multidendritic neurons and mammalian free nerve endings share various characteristics in common: Their dendrites both have free endings without forming particular sense organs to detect pain, temperature, and other submodalities. The md neurons project to the most ventral layer of the VNC, whereas free nerve endings innervate the most dorsal layer of the spinal cord. Axons from the insect external sensilla and mammalian hair receptors, both of which detect haptic contact to the tips of the bristles and hairs, terminate in the second-ventral and second-dorsal layers, respectively. Insect chordotonal organ and mammalian muscle spindle, as well as insect campaniform sensilla and mammalian Golgi tendon organ, also show similarity with respect to their functions in motor control. These receptor systems supply afferents to the most dorsal and most ventral layers in insects and mammals, respectively. A fly's stretch receptors and mammalian Merkel cell neurites-as well as Meissner, Ruffini, and Pacinian corpuscles-terminate in the third-ventral and third-dorsal layer, respectively. Although correspondence between them is less obvious, they similarly detect deformation of the exoskeleton and skin. Thus, functionally comparable somatosensory terminals are layered in reverse order between the two systems. Considering that the dorsoventral axis of the mammalian body is developmentally upside down compared with the insect one, the corresponding order of sensory arrangements is actually conserved exactly between the two systems (Tsubouchi, 2017).

Do corresponding somatosensory cell types express common genes? Modality-dependent molecular specialization is not apparent even within insects or mammals, because the same genes are often expressed in multiple cell types and only a few genes share expression in the corresponding cell types across taxa. This might be a rather general feature; receptor molecules as well as developmental origins of the sensory organs are not identical between insects and mammals also in olfactory and auditory systems, yet sensory centers in the brain share architectural similarities (Tsubouchi, 2017).

With this somatosensory analysis, transphyletic correspondence of neuronal circuitry has been found in all of the sensory modalities. Corresponding organization has been suggested also for associative centers and motor systems. The fact that essentially all important components of the brain system share conserved features across the two evolutionary clades, which have been separated since at least the end of the Ediacaran period more than 550 million years ago, would suggest that basic development programs for the orderly and secrete segregation of those circuits may have evolved before deuterostome-protostome or deuterostomia-ecdysozoa divergence (Tsubouchi, 2017).

Modulatory neurons project widely throughout the brain, dynamically altering network processing based on an animal's physiological state. The connectivity of individual modulatory neurons can be complex, as they often receive input from a variety of sources and are diverse in their physiology, structure, and gene expression profiles. To establish basic principles about the connectivity of individual modulatory neurons, a pair of identified neurons was examined, the 'contralaterally projecting, serotonin-immunoreactive deutocerebral neurons' (CSDns), within the olfactory system of Drosophila. Specifically, the neuronal classes were determined providing synaptic input to the CSDns within the antennal lobe (AL), an olfactory network targeted by the CSDns, along with the degree to which CSDn active zones are uniformly distributed across the AL. Using anatomical techniques, the CSDns were found to receive glomerulus-specific input from olfactory receptor neurons (ORNs) and projection neurons (PNs), and network-wide input from local interneurons (LNs). Furthermore, the number of CSDn active zones was quantified in each glomerulus; CSDn output was fount to be not uniform, but rather heterogeneous across glomeruli and stereotyped from animal to animal. Finally, it was demonstrated that the CSDns synapse broadly onto LNs and PNs throughout the AL, but do not synapse upon ORNs. These results demonstrate that modulatory neurons do not necessarily provide purely top-down input, but rather receive neuron class-specific input from the networks that they target, and that even a two cell modulatory network has highly heterogeneous, yet stereotyped pattern of connectivity (Coates, 2017).

Neuromodulation confers flexibility to anatomically-restricted neural networks so that animals are able to properly respond to complex internal and external demands. However, determining the mechanisms underlying neuromodulation is challenging without knowledge of the functional class and spatial organization of neurons that express individual neuromodulatory receptors. This study describes the number and functional identities of neurons in the antennal lobe of Drosophila melanogaster that express each of the receptors for one such neuromodulator, serotonin (5-HT). Although 5-HT enhances odor-evoked responses of antennal lobe projection neurons (PNs) and local interneurons (LNs), the receptor basis for this enhancement is unknown. Endogenous reporters of transcription and translation for each of the five 5-HT receptors (5-HTRs) were used to identify neurons, based on cell class and transmitter content, that express each receptor. Specific receptor types are expressed by distinct combinations of functional neuronal classes. For instance, the excitatory PNs express the excitatory 5-HTRs (5-HT2 type and 5-HT7), the 5-HT1 type receptors are generally inhibitory, and distinct classes of LNs each express different 5-HTRs. This study therefore provides a detailed atlas of 5-HT receptor expression within a well-characterized neural network, and enables future dissection of the role of serotonergic modulation of olfactory processing (Sizemore, 2016).

Neuromodulators often act through diverse sets of receptors expressed by distinct network elements and in this manner, differentially affect specific features of network dynamics. Knowing which network elements express each receptor for a given neuromodulator provides a framework for making predictions about the mechanistic basis by which a neuromodulator alters network activity. This study provides an 'atlas' of 5-HTR expression within the AL of Drosophila, thus revealing network elements subject to the different effects of serotonergic modulation. In summary, different receptors are predominantly expressed by distinct neuronal populations. For example, the 5-HT2B is expressed by ORNs, while the 5-HT2A and 7 are expressed by cholinergic PNs. Additionally, each receptor was found to be expressed by diverse populations of LNs, with the exception the 5-HT1B. For instance, 5-HT1A is expressed by GABAergic and peptidergic (TKK and MIP) LNs, while 5-HT2A and 2B are not expressed by peptidergic LNs. However, the vPNs are the exception to the general observation that distinct neuronal classes differ from each other in the 5-HTRs and the implications of this are discussed below. Together, these results suggest that within the AL, 5-HT differentially modulates distinct populations of neurons that undertake specific tasks in olfactory processing (Sizemore, 2016).

A recurring theme of neuromodulation is that the expression of distinct receptor types by specific neural populations allows a single modulatory neuron to differentially affect individual coding features. For instance, GABAergic medium spiny neurons (MSNs) in the nucleus accumbens express either the D1 or D2 dopamine receptor allowing dopamine to have opposite effects on different MSNs via coupling to different Galpha subunits (reviewed in56). MSNs that differ in dopamine receptor expression also differ in their synaptic connectivity. Dopamine activates D1-expressing MSNs that directly inhibit dopaminergic neurons in the ventral tegmental area (VTA), and inhibits D2-expressing MSNs that inhibit GABAergic VTA interneurons thus inducing suppression of dopamine release. In this manner, a single neuromodulator differentially affects two populations of principal neurons via different receptors to generate coordinated network output. This principle also holds true for the effects of 5-HT within the olfactory bulb. For instance, 5-HT enhances presynaptic inhibition of olfactory sensory neurons by 5-HT2C-expressing juxtaglomerular cells57, while increasing excitatory drive to mitral/tufted cells and periglomerular cells via 5-HT2A-expressing external tufted cells. Similarly, distinct classes of AL neurons were observed to differ in their expression of 5-HTRs. For instance, ePNs express the 5-HT2A, 5-HT2B and 5-HT7 receptors, while peptidergic LNs predominantly express the 5-HT1A receptor. This suggests that the cumulative effect of 5-HT results from a combination of differential modulation across neuronal populations within the AL. Interestingly, although it was found that 5-HT2B is expressed by ORNs, previous reports found that 5-HT does not directly affect Drosophila ORNs. In this study, ORNs were stimulated using antennal nerve shock in which the antennae were removed in order to place the antennal nerve within a suction electrode. Thus, if 5-HT2B is localized to the ORN cell body, removal of the antennae would eliminate any effect of 5-HT on ORNs. In several insects, 5-HT within the antennal haemolymph modulates ORN odor-evoked responses. Therefore, it is plausible ORNs are modulated by a source of 5-HT other than the CSD neurons within the AL. Serotonergic modulation of LN activity has widespread, and sometimes odor specific, effects on olfactory processing. LNs allow ongoing activity across the AL to shape the activity of individual AL neurons, often in a glomerulus specific manner creating non-reciprocal relationships. It is fairly clear that 5-HT directly modulates LNs, although 5-HT almost certainly affects synaptic input to LNs. Serotonin modulates isolated Manduca sexta LNs in vitro and, consistent with the current results, a small population of GABAergic LNs in the AL of Manduca also express the 5-HT1A receptor. Furthermore, 5-HT has odor-dependent effects on PN odor-evoked activity, suggesting that odor specific sets of lateral interactions are modulated by 5-HT. Different populations of LNs were found to express different sets of 5-HT receptors, however LNs were categorized based on transmitter type, so it is possible that these categories could be even further sub-divided based on morphological type, synaptic connectivity or biophysical characteristics. Regardless, the results suggest that 5-HT modulates lateral interactions within the AL by selectively affecting LN populations that undertake different tasks. For instance, the TKKergic LNs that express the 5-HT1A receptor provide a form of gain control by presynaptically inhibiting ORNs32. The results suggest that 5-HT may affect TKK mediated gain control differently relative to processes undertaken by other LN populations. Furthermore, the expression of the TKK receptor by ORNs is regulated by hunger, allowing the effects of TKK to vary with behavioral state. It would be interesting to determine if the expression of 5-HTRs themselves also vary with behavioral state as a means of regulating neuromodulation within the olfactory system (Sizemore, 2016).

Although it was primarily found that individual populations of AL neurons chiefly expressed a single or perhaps two 5-HTR types, the vPNs appear to be an exception. As a population, the vPNs express all of the 5-HTRs and the vPNs that express each 5-HTR did not appear to differ in terms of the proportion of those neurons that were GABAergic or cholinergic (roughly 3:2). Unfortunately, the approach does not allow determination of the degree to which individual vPNs co-express 5-HTRs. However, it is estimated that there are ~51 vPNs and even if this is an underestimate, there is likely some overlap of receptor types as a large number of vPNs expressed the 5-HT1A, 1B, 2B and 7 receptors. It is possible that a single vPN expresses one 5-HTR in the AL and a different 5-HTR in the lateral horn. However, the current approach only allows identification of which neurons express a given 5-HTR, not where that receptor is expressed. The CSD neurons ramify throughout both ALs and both lateral horns, thus vPNs could have differential spatial expression of individual 5-HTRs. Individual neurons expressing multiple 5-HTRs has been demonstrated in several neural networks. For instance, pyramidal cells in prefrontal cortex express both the 5-HT1A and 5-HT2A7. This allows 5-HT to have opposing effects that differ in their time course in the same cell. In terms of the vPNs, the results suggest that the current understanding of the diversity of this neuron class is limited. The expression of receptors for different signaling molecules could potentially be a significant component to vPN diversity (Sizemore, 2016).

Neuromodulators are often released by a small number of neurons within a network, yet they can have extremely diverse effects depending upon patterns of receptor expression. For the most part, individual populations of AL neurons differed in the receptor types that they expressed. This suggests that 5-HT differentially acts on classes of neurons that undertake distinct tasks in olfactory processing. In the case of the vPNs, this differential modulation may be fairly complex due to the diversity within this neuronal class. The goal of this study was to establish a functional atlas of 5-HTR expression in the AL of Drosophila. This dataset therefore provides a mechanistic framework for the effects of 5-HT on olfactory processing in this network (Sizemore, 2016).

Glutamatergic neurons are abundant in the Drosophila central nervous system, but their physiological effects are largely unknown. This study investigated the effects of glutamate in the Drosophila antennal lobe, the first relay in the olfactory system and a model circuit for understanding olfactory processing. In the antennal lobe, one-third of local neurons are glutamatergic. Using in vivo whole-cell patch clamp recordings, this study found that many glutamatergic local neurons are broadly tuned to odors. Iontophoresed glutamate hyperpolarizes all major cell types in the antennal lobe, and this effect is blocked by picrotoxin or by transgenic RNAi-mediated knockdown of the GluClα gene, which encodes a glutamate-gated chloride channel. Moreover, antennal lobe neurons are inhibited by selective activation of glutamatergic local neurons using a nonnative genetically encoded cation channel. Finally, transgenic knockdown of GluClα in principal neurons disinhibits the odor responses of these neurons. Thus, glutamate acts as an inhibitory neurotransmitter in the antennal lobe, broadly similar to the role of GABA in this circuit. However, because glutamate release is concentrated between glomeruli, whereas GABA release is concentrated within glomeruli, these neurotransmitters may act on different spatial and temporal scales. Thus, the existence of two parallel inhibitory transmitter systems may increase the range and flexibility of synaptic inhibition (Liu, 2013).

Although glutamatergic neurons are abundant in the Drosophila brain, the role of glutamate as a neurotransmitter in the Drosophila CNS has received little study. In the antennal lobe, where approximately one-third of LNs are glutamatergic, the physiological effects of glutamate have never been characterized. This study shows that glutamate is an inhibitory transmitter that shapes the responses of PNs to olfactory stimuli (Liu, 2013).

In the past, glutamate has been proposed to mediate lateral excitation between olfactory glomeruli. The results of this study demonstrate that the main effect of glutamate is inhibition, not excitation. The possibility cannot be ruled out that glutamate has small excitatory effects, but no evidence was found of excitation even when GluClα was knocked down genetically or inhibited pharmacologically. It is noted that there is in fact lateral excitation in the antennal lobe, which exists in parallel with lateral inhibition. However, lateral excitation is mediated not by glutamate, but by electrical coupling between LNs and PNs (Liu, 2013).

All of the effects of glutamate on PNs were eliminated by knocking down GluClα. The dominant role for GluClα is notable, given how many other glutamate receptors are in the genome. The results are particularly surprising in light of two recent studies that have reported behavioral effects of knocking down an NMDA receptor subunit (NR1) in PNs. Further experiments will be needed to clarify the role of NR1 (Liu, 2013).

There is a precedent for the idea that glutamate can be an inhibitory neurotransmitter in the Drosophila brain. Specifically, several studies have reported that bath-applied glutamate inhibits the large ventrolateral neurons of the Drosophila circadian clock circuit. Collectively, these studies suggest roles for both ionotropic and metabotropic glutamate receptors in glutamatergic inhibition. Regardless of which glutamate receptors are involved, these studies are consistent with the conclusion that glutamate is an important mediator of synaptic inhibition (Liu, 2013).

The idea that glutamate can be inhibitory has important implications for neural coding. One particularly interesting case is the motion vision circuit of the Drosophila optic lobe. Two neuron types, L1 and L2, both receive strong synaptic inputs from photoreceptors, and they respond equally to contrast increments (“on”) and decrements (“off”). However, based on conditional silencing experiments, L1 is thought to provide input to an on pathway, and L2 to an off pathway. Therefore, opponency must arise downstream from L1 and L2. According to recent evidence, L1 is glutamatergic, whereas L2 is cholinergic. In light of the current data, that result suggests that L1 may actually be inhibitory, which would be sufficient to create opponency in the on and off pathways (Liu, 2013).

Glutamate can act as an inhibitory neurotransmitter in the Caenorhabditis elegans olfactory circuit, and this fact too has implications for neural coding of odors in this organism. In the worm, a specific type of glutamatergic olfactory neuron inhibits one postsynaptic neuron via GluCl, while also exciting another postsynaptic neuron via an AMPA-like receptor. This arrangement creates a pair of opponent neural channels that respond in an anticorrelated fashion to odor presentation or odor removal, analogous to opponent channels in the visual system (Liu, 2013).

This study has shown that the cellular actions of Glu-LNs are broadly similar to the actions of GABA-LNs. Specifically, both types of LNs inhibit PNs and other LNs. In addition, both GABA and glutamate inhibit neurotransmitter release from ORNs. Thus, both neurotransmitters inhibit all of the major cell types in the antennal lobe circuit. However, Glu-LNs and GABA-LNs are not functionally identical. In particular, it was found that the vesicular glutamate transporter is mainly confined to the spaces between glomeruli, whereas the vesicular GABA transporter is abundant within glomeruli. This finding implies that glutamate and GABA are released in largely distinct spatial locations. Consistent with this implication, no individual synaptic connections from Glu-LNs onto PNs were found, whereas a substantial rate of connections was found from GABA-LNs onto PNs. Nevertheless, PNs were found to be hyperpolarized by coactivation of multiple Glu-LNs, and PNs are disinhibited by knockdown of GluCl specifically in PNs (Liu, 2013).

These results can be reconciled by a model where the sites of glutamate release are distant from PN glutamate receptors. As a result, glutamate would need to diffuse some distance to inhibit PNs. Coactivation of multiple Glu-LNs would increase extracellular glutamate concentrations, overwhelming uptake mechanisms and allowing glutamate to diffuse further. In this scenario, glutamatergic inhibition should be most important when LN activity is intense and synchronous. By comparison, GABAergic inhibition of PNs does not require LN coactivation, implying a comparatively short distance between presynaptic and postsynaptic sites. There is a precedent in the literature for the idea that different forms of inhibition can be differentially sensitive to LN coactivation, due to the spatial relationship between presynaptic and postsynaptic sites. In the hippocampus, GABA_A receptors are closer than GABA_B receptors to sites of GABA release, and so activation of individual interneurons produces GABA_A but not GABA_B currents, whereas coactivation of many interneurons produces both GABA_A and GABA_B currents (Liu, 2013).

The pharmacology of glutamate-gated conductances in antennal lobe neurons is similar to the pharmacology of GABA_A conductances in these neurons. This result should prompt a reevaluation of studies that used picrotoxin to block inhibition in the antennal lobe. Given the current results, it seems likely that these studies were reducing both glutamatergic and GABAergic inhibition (Liu, 2013).

It is perhaps surprising that knocking down GluClα in PNs had such a substantial effect on PN odor responses, given that picrotoxin alone has comparatively modest effects. The solution to this puzzle may lie in the finding that glutamate regulates not only PNs but also GABA-LNs. Importantly, GABA-LNs are spontaneously active and provide tonic inhibition to PNs. Hence, in the intact circuit, glutamatergic inhibition of GABA-LNs should tend to disinhibit PNs. Picrotoxin prevents Glu-LNs from inhibiting GABA-LNs and should tend to potentiate GABAergic inhibition. The effects of GABA are mediated in part by GABA_B receptors, which are not sensitive to picrotoxin. Thus, picrotoxin likely has bidirectional effects on the total level of inhibition in the circuit. By contrast, knockdown of GluClα specifically in PNs should not directly affect GABA-LNs and should not produce these complex effects. These results illustrate how a cell-specific genetic blockade of a neurotransmitter system can have more dramatic effects than a global pharmacological blockade of the same system (Liu, 2013).

This study reveals that an LN can have push-pull effects on a single population of target cells. For example, Glu-LNs directly inhibit PNs, but they should also disinhibit PNs, via the inhibition of GABA-LNs. This architecture may allow for more robust gain control and rapid transitions between network states and is similar to the wiring of many cortical circuits, where corecruitment of excitation and inhibition is a common motif (Liu, 2013).

Why might the existence of two parallel inhibitory transmitters be useful? The data argue that GABA and glutamate may act on different spatial and temporal scales. Because these two inhibitory systems comprise different cells, receptors, and transporters, they can be modulated independently. Because their properties are encoded by different genes, they can also evolve independently. This organization should confer increased flexibility in adapting synaptic inhibition to a changing environment (Liu, 2013).

The antennal lobe of Drosophila is perhaps one of the best understood neural circuits, because of its well-described anatomical and functional organization and ease of genetic manipulation. Olfactory lobe interneurons - key elements of information processing in this network - are thought to be generated by three identified central brain neuroblasts, all of which generate projection neurons. One of these neuroblasts, located lateral to the antennal lobe, also gives rise to a population of local interneurons, which can either be inhibitory (GABAergic) or excitatory (cholinergic). Recent studies of local interneuron number and diversity suggest that additional populations of this class of neurons exist in the antennal lobe. This implies that other, as yet unidentified, neuroblast lineages may contribute a substantial number of local interneurons to the olfactory circuitry of the antennal lobe. This study identified and characterized a novel glutamatergic local interneuron lineage in the Drosophila antennal lobe. MARCM (mosaic analysis with a repressible cell marker) and dual-MARCM clonal analysis techniques to identify this novel lineage unambiguously, and to characterize interneurons contained in the lineage in terms of structure, neurotransmitter identity, and development. The glutamatergic nature of these interneurons was demonstrated by immunohistochemistry and an enhancer-trap strain was used that reports the expression of the Drosophila vesicular glutamate transporter (DVGLUT). The neuroanatomical features of these local interneurons at single-cell resolution, and the marked diversity in their antennal lobe glomerular innervation patterns was documented. Finally, the development of these dLim-1 and Cut positive interneurons was tracked during larval and pupal stages. This study has identified a novel neuroblast lineage that generates neurons in the antennal lobe of Drosophila. This lineage is remarkably homogeneous in three respects. All of the progeny are local interneurons, which are uniform in their glutamatergic neurotransmitter identity, and form oligoglomerular or multiglomerular innervations within the antennal lobe. The identification of this novel lineage and the elucidation of the innervation patterns of its local interneurons (at single cell resolution) provides a comprehensive cellular framework for emerging studies on the formation and function of potentially excitatory local interactions in the circuitry of the Drosophila antennal lobe (Das, 2011).

Naive Drosophila larvae show vigorous chemotaxis toward many odorants including ethyl acetate (EA). Chemotaxis toward EA is substantially reduced after a 5-min pre-exposure to the odorant and recovers with a half-time of ~20 min. An analogous behavioral decrement can be induced without odorant-receptor activation through channelrhodopsin-based, direct photoexcitation of odorant sensory neurons (OSNs). The neural mechanism of short-term habituation (STH) requires the (1) Rutabaga adenylate cyclase; (2) transmitter release from predominantly GABAergic local interneurons (LNs); (3) GABA-A receptor function in projection neurons (PNs) that receive excitatory inputs from OSNs; and (4) NMDA-receptor function in PNs. These features of STH cannot be explained by simple sensory adaptation and, instead, point to plasticity of olfactory synapses in the antennal lobe as the underlying mechanism. These observations suggest a model in which NMDAR-dependent depression of the OSN-PN synapse and/or NMDAR-dependent facilitation of inhibitory transmission from LNs to PNs contributes substantially to short-term habituation (Larkin, 2010).

Experience-induced plasticity of synapses is believed to be a fundamental mechanism of learning and memory. However, central synaptic changes that underlie memory have not been clearly defined, even for relatively simple nonassociative learning processes such as habituation (Larkin, 2010).

During habituation, unreinforced exposure to a repeated or prolonged stimulus results in a reversible decrease in response to that stimulus. Habituation probably serves as an important building block for more complex cognitive function. By allowing unchanging or irrelevant stimuli to be ignored, it allows cognitive resources to be focused on more salient stimuli (Larkin, 2010 and references therein).

The neural basis of short-term habituation (STH) is best studied in the marine snail, Aplysia californica. In this organism STH (lasting ~30 min) of the defensive gill-withdrawal reflex in response to tactile stimulation of the siphon is thought to arise from presynaptic depression of transmitter release at sensorimotor synapses. However, even here, presynaptic plasticity may not be cell-autonomous, potentially requiring, for instance, activity of yet-to-be-identified interneurons (Larkin, 2010).

Several forms of habituation have been described in Drosophila and are often shown to require the function of genes that regulate cAMP-dependent forms of associative memory. For instance, habituation of proboscis extension reflex as well as odor-evoked startle reflex in adult Drosophila requires rutabaga (rut)-encoded Ca2+/calmodulin-sensitive adenylyl cyclase. In addition, habituation of the ethanol-induced startle response requires the shaggy/GSK-3 signaling pathway. Despite such pioneering observations, the mechanisms of these various forms of habituation, even whether the primary neuronal changes are purely sensory or involve plasticity of central synapses (involving centrally located interneurons that may integrate various different kinds of modulatory, inhibitory, and excitatory inputs), remain poorly understood (Larkin, 2010).

Recent advances in understanding the circuitry that underlies Drosophila olfactory behavior, as well as the development of new tools to perturb identified neurons in vivo, has opened the opportunity for understanding mechanisms of olfactory habituation at the level of the underlying neural circuitry (Larkin, 2010).

In the larval olfactory system, 21 olfactory sensory neurons (OSNs), each expressing a single odorant receptor (together with the broadly expressed Or83b co-receptor), synapse, respectively, onto 21 cognate projection neurons (PNs) within 21 glomeruli in the larval antennal lobe (AL). Local, predominantly GABAergic interneurons (LNs) synapse widely within the antennal lobe, interlinking different glomeruli. Various neuromodulatory synapses also form on the larval antennal lobe and mushroom body. Thus, odorant-stimulated signals in sensory neurons are processed in the antennal lobe, modulated by motivational or emotional states, and relayed through projection neurons to higher brain centers (Larkin, 2010).

Previous work has shown that in Drosophila larvae, olfactory chemotaxis decreases after odorant pre-exposure. This study shows that this behavioral habituation, alternatively referred to as 'adaptation' by some previous investigators, arises from mechanisms of synaptic plasticity. This study demonstrates that odorant receptor activation is not necessary for olfactory habituation; however, local interneuron activity and projection neuron signaling is necessary. These observations suggest a model in which habituation occurs by a pathway in which NMDA receptors in projection neurons signal depression of OSN-PN synapses and/or facilitation of LN-PN synapses (Larkin, 2010).

Previous studies have not clearly discriminated between peripheral and central mechanisms. Indeed, the term 'adaptation,' better applied to sensory neuron changes such as receptor desensitization, has often been used interchangeably with the term 'habituation', which is usually restricted to behavioral changes arising from central synaptic mechanisms (Larkin, 2010). .

The form of larval olfactory STH characterized in this study displays at least some of the defining behavioral characteristics of habituation. First, there is a behavioral decrement in response to repeated or sustained application of a particular stimulus. Second, STH shows spontaneous recovery with time in the absence of the habituating stimulus. And third, STH is susceptible to dishabituation when habituated larvae are presented with of a strong or noxious stimulus. The property of dishabituation is particularly significant, as an important way of distinguishing between habituation and either fatigue or sensory adaptation. Dishabituation shows that the habituated animal retains the capability to respond and suggests that the attenuated behavioral response arises from some form of active suppression. Thus, the behavioral data suggest (1) that the term 'habituation' may be better used in place of 'adaptation,' while referring to the behavioral phenomenon that was studied; and (2) that STH probably arises from central synaptic mechanisms, rather than sensory neuron adaptation (Larkin, 2010).

Three main lines of data support the conclusion that STH arises from a central synaptic mechanism that resides in the antennal lobe, rather than from adaptation of olfactory receptor signaling in the OSN. First, behavioral decrements similar to STH can be induced by direct depolarization of OSNs, indicating that STH may potentially be induced by processes stimulated by activation action-potential firing in OSNs, independently of olfactory receptor activation. Second, and more striking, STH requires synaptic-vesicle exocytosis from local interneurons during the process of odorant exposure, when STH is being established. This requirement is incompatible with an exclusively sensory mechanism. Third, STH requires the function of NMDA receptors on postsynaptic projection neurons. This last observation also provides a particularly strong argument for a synaptic mechanism, indicating a need for plasticity of OSN and/or LN synapses made onto dendrites of projection neurons in the antennal lobe. Given that OSNs are excitatory and LNs are primarily inhibitory, it appears most likely that NMDAR functions in PNs to depress excitatory OSN-PN synapses and/or to potentiate inhibition by strengthening the LN-PN synapse. It is suggestd that the LN-PN mechanism may be involved because (1) LN transmission seems necessary for both induction and expression of habituation; and (2) the process of dishabituation could be attractively explained as arising from the inhibition of local inhibitory synapses through descending neuromodulation. A requirement for facilitation of the LN-PN synapse would be consistent with previous studies (Sachse, 2007) showing that adult-long-term olfactory habituation is associated with an increase in odor-evoked calcium fluxes in GABAergic processes within the Drosophila antennal lobe (Larkin, 2010).

Based both on experimental and theoretical arguments, a simple model is suggested for short-term olfactory habituation. Since this is a model, no claim is being made to to having ruled out additional major contributing mechanisms, It is suggested that during initial odorant pre-exposure, dendritic NMDA receptors on projection neurons detect and respond to membrane depolarization occurs coincident with transmitter release from LNs. Calcium entry through dendritic NMDA receptors may trigger a local retrograde signal required for facilitation of transmitter release from the LNs. Although existing data do not rule out functions for rutabaga in higher larval brain centers, it is suggested that either the generation of a retrograde signal in PN dendrites or the presynaptic response of LNs to this signal could be dependent on the rut adenylate cyclase. In habituated animals, facilitation of GABA release would reduce odor-evoked projection neuron outputs to higher brain centers, thereby reducing olfactory behavior. As NMDAR signaling would only occur at active glomeruli, this mechanism can account not only for the observed odor selectivity of habituation, but also the instances of cross-habituation (Larkin, 2010).

Such a model also naturally suggests a hypothesis for the mechanism of dishabituation: namely, that dishabituating stimuli cause release of neuromodulators that act to reduce GABA release from local inhibitory synapses (Larkin, 2010).

Given the remarkable similarities in the anatomical organization of insect and mammalian olfactory systems, a significant conservation of olfactory mechanisms would be expected. In rodents, at least two forms of habituation have been described, lasting 2-3 and 30-60 min, respectively: the latter equivalent in timescale to larval STH described in this study. Consistent with a similar underlying mechanism, the more persistent form of olfactory habituation can be blocked by an N-methyl-D-aspartate (NMDA) receptor antagonist in the olfactory bulb, a structure homologous to the insect antennal lobe. Thus, larval STH described in this study has some similarities to a previously characterized form of mammalian olfactory habituation. Analysis of the underlying mechanisms is therefore likely to provide directly transferable insights in mammalian olfaction. The data make the prediction that the activity of mammalian olfactory interneurons, either periglomerular or granule cells, is critical for the establishment and display of at least one timescale of olfactory habituation (Larkin, 2010).

In addition to providing some insight into mechanisms of olfactory habituation in mammals, it possible that circuit mechanisms of larval olfactory habituation are relevant to other forms of behavioral habituation. In at least three previous instances, increased inhibition has been associated with attenuated behavior. For example, habituation of an escape reflex mediated by the lateral giant fibers in the crayfish has been associated with enhanced GABAergic transmission onto giant fibers. Similarly, LTP of inhibitory synapses controlling excitability of the Mauthner cell has been associated with reduced escape behavior in goldfish. Furthermore, ethanol, a potentiator of GABA synapses, has been shown to enhance habituation of a motor pathway in the frog spinal cord. Could these different instances of habituation all involve circuit mechanisms similar to those used in Drosophila larval olfactory behavior (Larkin, 2010)?

In all brain regions, principal/projection neurons are subject to inhibitory feedback modulation and a pathway that has been appreciated as potentially essential for neuronal homeostasis. Potentiation of inhibitory feedback triggered by the pattern of principle cell activation would be predicted to preferentially dampen this particular output pattern. Thus, the circuit mechanism suggest in this study is theoretically generalizable to other and more complex forms of habituation. Further experiments will be required to determine the validity of this very testable hypothesis (Larkin, 2010).

The importance of habituation has been underlined by the fact that deficits in sensory gating and pre-pulse inhibition (PPI), processes with similarities to habituation, have been linked with various neurological problems, including autism and schizophrenia. Indeed, a circuit model for understanding schizophrenia has specifically proposed that altered negative feedback in the hippocampus may underlie both positive and negative symptoms of schizophrenia (Larkin, 2010).

In addition, defects in habituation or habituation-like processes have been described in Fragile X syndrome and migraines. It has also been shown to have important effects relating to learning disabilities, age-related changes in learning, and substance abuse. If mechanisms of olfactory habituation prove to be general, then studies of olfactory plasticity may prove relevant for other forms of cognition as well as for human neurological disease (Larkin, 2010).

Early sensory processing can play a critical role in sensing environmental cues. This study investigated the physiological and behavioral function of gain control at the first synapse of olfactory processing in Drosophila. Olfactory receptor neurons (ORNs) express the GABA_B receptor (GABA_BR) and its expression expands the dynamic range of ORN synaptic transmission that is preserved in projection neuron responses. Strikingly, it was found that different ORN channels have unique baseline levels of GABA_BR expression. ORNs that sense the aversive odorant CO₂ do not express GABA_BRs nor exhibit any presynaptic inhibition. In contrast, pheromone-sensing ORNs express a high level of GABA_BRs and exhibit strong presynaptic inhibition. Furthermore, a behavioral significance of presynaptic inhibition was revealed by a courtship behavior in which pheromone-dependent mate localization is impaired in flies that lack GABA_BRs in specific ORNs. Together, these findings indicate that different olfactory receptor channels may employ heterogeneous presynaptic gain control as a mechanism to allow an animal's innate behavioral responses to match its ecological needs (Root, 2008).

The stereotypic organization of the Drosophila olfactory system and the identification of the family of odorant receptor genes make the fly an attractive system to study olfactory mechanisms. An adult fly expresses about 50 odorant receptor genes and each ORN typically expresses just one or a few receptor genes. ORNs detect odors in the periphery and send axons to glomeruli in the antennal lobe. Each glomerulus receives axons from about 20 ORNs expressing the same receptor genes and dendrites of a few uniglomerular projection neurons (PNs), which propagate olfactory information to higher brain centers. This numerically simple olfactory system coupled with genetic markers to label most of the input channels provides an opportunity to dissect synaptic function and information processing (Root, 2008).

The Drosophila antennal lobe is populated with GABAergic local interneurons (LNs) that release GABA in many if not all glomeruli. GABA exerts its modulatory role via two distinct receptor systems, the fast ionotropic GABA_A receptor, which is sensitive to the antagonist picrotoxin, and the slow metabotropic GABA_B receptor, which is sensitive to the antagonist CGP54626. Pharmacological blockade of the GABA receptors demonstrate that GABA-mediated hyperpolarization suppresses PN response to odor stimulation in a non-uniform fashion. Electron microscopy studies of the insect antennal lobe show that GABAergic LNs synapse with PNs, which support the well established olfactory mechanism of lateral inhibition. GABAergic LNs also synapse onto ORNs and imaging studies in mouse suggest that activation of GABA_BRs in ORN terminals suppress neurotransmitter release in ORNs (Root, 2008).

It was hypothesized that setting the appropriate olfactory gain for environmental cues is important for adjusting an organism's sensitivity to its environment. A recent study shows that GABA_BR mediated presynaptic inhibition provides a mechanism to modulate olfactory gain. Electrical recordings show that interglomerular presynaptic inhibition suppresses the olfactory gain of PNs to potentially increase the dynamic range of the olfactory response. Likewise, gain modulation may not be uniform among different glomeruli, which could reflect a tradeoff between sensitivity and dynamic range in different olfactory channels. For example, high sensitivity may be crucial for some environmental cues, such as those that require an immediate behavioral response, whereas a larger dynamic range may be more advantageous for other odors where precise spatial and temporal information may be critical for optimal performance (Root, 2008).

This study investigated the physiological and behavioral function of gain control in early olfactory processing. Interneuron-derived GABA was shown to activate GABA_BRs on ORN terminals, reducing the gain of ORN-to-PN synaptic transmission. Different types of ORNs exhibit different levels of presynaptic inhibition and this heterogeneity in presynaptic inhibition is preserved in antennal lobe output projection neurons. Interestingly, pheromone-sensing ORNs exhibit high levels of GABA_BR expression and behavioral experiments indicate that GABA_BR expression in a population of pheromone ORNs is important for mate localization, suggesting that presynaptic gain control is important for the olfactory channel-specific fine-tuning of behavior (Root, 2008).

Two-photon imaging was used to monitor activity in selective neural populations in the antennal lobe. Specific blockade of GABA_BRs reveals a scalable presynaptic inhibition to suppress olfactory response at high odor concentrations. Pharmacological and molecular experiments suggest that GABA_BRs are expressed in primary olfactory receptor neurons. Furthermore, the level of presynaptic inhibition is different in individual glomerular modules, which is tightly linked to the level of GABA_BR expression. The importance of presynaptic GABA_BRs in olfactory localization was investigated, and it was found that reduction of GABA_BR expression in the presynaptic terminal of ORNs impairs the ability of male flies in locating potential mates (Root, 2008).

Heterogeneity was found in the levels of presynaptic inhibition among different glomeruli. Varying GABA_BR2 expression level in ORNs with molecular manipulations is sufficient to produce predictable alterations in presynaptic inhibition in specific glomeruli. Together these experiments argue that presynaptic GABA_BR expression level is a determinant of glomerulus-specific olfactory gains in the antennal lobe. A recent report revealed that there is a non-linear transformation between ORNs and PNs that is heterogeneous between glomeruli. In other words, PNs innervating a given glomerulus have a unique response range for its ORN input. Given that ORNs are the main drivers of PN response, it is plausible that the heterogeneity in presynaptic inhibition contributes to the heterogeneity in ORN to PN transformations observed by Bhandawat and colleagues. Additionally, heterogeneity in GABA release by LNs could also contribute to heterogeneity in presynaptic inhibition. It is interesting to note that when presynaptic inhibition is abolished, heterogeneity remains in the input-output curves of PN response to the four different odors in these experiments, suggesting that other mechanisms such as probability of vesicle release contribute to the heterogeneity as well (Root, 2008).

Theoretical analysis of antennal lobe coding has recently suggested that the non-linear synaptic amplification in PNs provides an efficient coding mechanism for the olfactory system. According to this model, the optimal distribution of firing rates across a range of odorants should be flat without clusters. Firing rates of a given ORN responses cluster in an uneven distribution. Conversely, PNs exhibit a more equalized firing rate distribution than ORNs. According to the optimum coding theory, the high amplification of ORN to PN transformation generates a more even distribution of PN firing rates that should facilitate odor discrimination. However, this model of olfactory coding poses a potential problem. The high gain in this synaptic amplification reduces the dynamic range of PNs, causing a loss of information about concentration variation that could be important for an animal to localize odor objects. Presynaptic inhibition may provide a mechanism to expand the dynamic range of the olfactory system. For some glomerular modules that mediate innate behaviors such as avoidance of the stress odorant CO₂, there is a potential trade off for odor sensitivity and dynamic range. The lack of GABA_BR in the CO₂ sensing ORNs could be important to maintain high sensitivity (Root, 2008).

Pheromones play an important role in Drosophila mating behaviors and the current results indicate that pheromone sensing ORNs have high levels of GABA_BR, which is correlated with a high level of presynaptic inhibition in these ORNs. Mate localization is impaired in the absence of presynaptic inhibition in one pheromone sensing glomerulus. It is interesting to note that in addition to the pheromone sensing ORNs, the palpal ORNs also exhibit high GABA_BR expression level. Although the behavioral role of the palpal ORNs has not been determined, it is possible that they are also important for odor object localization. There are two potential mechanisms for the role of GABA_BR in olfactory localization. GABA_BR-mediated activity-dependent suppression of presynaptic transmission on a short time scale provides a mechanism for dynamic range expansion. On a longer time scale, activity-dependent suppression provides a mechanism for adaptation, hence a high pass filter to allow the detection of phasic information. Further experiments will be necessary to determine which property is important for olfactory localization (Root, 2008).

Intraglomerular and interglomerular presynaptic inhibition mediated by GABA_BRs have been described in the mammalian olfactory system. Intraglomerular presynaptic inhibition was suggested as a mechanism to control input sensitivity while maintaining the spatial maps of glomerular activity. Interglomerular presynaptic inhibition was proposed as a mechanism to increase the contrast of sensory input. A recent report revealed a similar gain control mechanism by interglomerular presynaptic inhibition in the Drosophila olfactory system where GABA_BR expression in ORNs was shown to scale the gain of PN responses. Interestingly, most if not all of the presynaptic inhibition was suggested to be lateral. In contrast, this study study does not seek to distinguish between intra- and interglomerular presynaptic inhibition, however evidence was found that the VA1lm glomerulus receives significant intraglomerular presynaptic inhibition. Thus, despite significant differences between the insect and mammalian olfactory systems, the inhibitory circuit in the first olfactory processing center appears remarkably similar (Root, 2008).

Based on whole cell recordings of PNs in response to ORN stimulation, Olsen (2008) suggests that both GABA_AR and GABA_BR are expressed in ORNs to mediate presynaptic inhibition and that GABA_AR signaling is a large component of lateral presynaptic inhibition. In contrast, this study, which employed direct optical measurements of presynaptic calcium and synaptic vesicle release, suggests that GABA_BRs but not GABA_ARs are involved in presynaptic inhibition. To resolve these discrepancies further molecular experiments will be important to determine conclusively whether ORNs express GABA_AR and whether the receptor contributes to gain control. Furthermore, the antennal lobe is a heterosynaptic system comprised of at least three populations of neurons that include ORNs, LNs and PNs. Therefore, how these different populations of neurons respond to GABA signaling and what contribution they make to olfactory processing in the antennal lobe is a critical question for future investigation (Root, 2008).

This study has demonstrated differential presynaptic gain control in individual olfactory input channels and its contribution to the fine-tuning of physiological and behavioral responses. Synaptic modulation by the intensity of receptor signaling is reminiscent of the mammalian nervous system where expression levels of AMPA glutamate receptors play an important role in regulating synaptic efficacy. Furthermore, presynaptic regulation of GABA_BR signaling provides a mechanism to modulate the neural activity of individual input channels without much interference with overall detection sensitivity because this mechanism of presynaptic inhibition would only alter responses to high intensity stimuli. In parallel, it is tempting to speculate that global modulation of interneuron excitability should alter the amount of GABA release across channels, thus providing a multi-channel dial of olfactory gain control that may reflect the internal state of the animal (Root, 2008).

Odors are detected by sensory neurons that carry information to the olfactory lobe where they connect to projection neurons and local interneurons in glomeruli: anatomically well-characterized structures that collect, integrate and relay information to higher centers. Recent studies have revealed that the sensitivity of such networks can be modulated by wide-field feedback neurons. The connectivity and function of such feedback neurons are themselves subject to alteration by external cues, such as hormones, stress, or experience. Very little is known about how this class of central neurons changes its anatomical properties to perform functions in altered developmental contexts. A mechanistic understanding of how central neurons change their anatomy to meet new functional requirements will benefit greatly from the establishment of a model preparation where cellular and molecular changes can be examined in an identified central neuron. This study examined a wide-field serotonergic neuron in the Drosophila olfactory pathway and mapped the dramatic changes that it undergoes from larva to adult. Expression of a dominant-negative form of the ecdysterone receptor prevents remodeling. Different transgenic constructs were used to silence neuronal activity, and defects are reported in the morphology of the adult-specific dendritic trees. The branching of the presynaptic axonal arbors is regulated by mechanisms that affect axon growth and retrograde transport. The neuron develops its normal morphology in the absence of sensory input to the antennal lobe, or of the mushroom bodies. However, ablation of its presumptive postsynaptic partners, the projection neurons and/or local interneurons, affects the growth and branching of terminal arbors. These studies establish a cellular system for studying remodeling of a central neuromodulatory feedback neuron and also identify key elements in this process. Understanding the morphogenesis of such neurons, which have been shown in other systems to modulate the sensitivity and directionality of response to odors, links anatomy to the development of olfactory behavior (Singh, 2007).

Changes in the pattern of arborization of a mature neuron can come about as a consequence of removal of its afferent inputs or targets, chronic stress or other environmental inputs, such as delivered during learning or exercise. Many of these changes are effected through the action of growth factors and developmental signals acting in concert with steroid hormones and neuronal activity to modify the cytoskeleton or synaptic properties relevant to an altered functional setting. Metamorphosis in Drosophila (a period during which mature larval neurons are often altered to take on new adult functions) provides a context where the mechanistic underpinnings of such neuronal change can be genetically dissected (Singh, 2007).

This study used a genetic method to mark the serotonin-immunoreactive deutocerebral interneurons (CSDn), recently identified on the basis of serotonin immunoreactivity. While this preparation identifies a central neuron, it also has an important feature that allows the analysis of mechanisms underlying the changes it undergoes during remodeling. This system, because of the random nature of the RN2-FLP action, results in bilateral, unilateral or no excision of the FRT element in the Tub-FRT-CD2-FRT-Gal4 construct in the CSDn. Thus, it was possible to choose and analyze preparations where the CSDn from only one hemisphere was labeled: this facility is vital as it allows the analysis of contralateral and ipsilateral projections of the CSDn, without this being obscured by projections of the neuron from the other hemisphere to the same target sites. The GFP reporter in the RN2-Flp, Tub-FRT-CD2-FRT-Gal4, UAS mCD8-GFP strain is first detected very late in embryogenesis (stage 20), after the neuron has acquired its mature larval pattern. These features thus provide a preparation where an identified central neuron, whose function is known, can be followed and genetically manipulated as it changes its form in response to external and internal cues during metamorphosis (Singh, 2007).

The neuron, present during the larval stages, undergoes well-defined changes during pupation to give rise to a more complex adult architecture. What are the factors that regulate the stereotyped pruning and re-growth of arbors in the CSDn during metamorphosis? The results suggest that the interaction of external factors and autonomous properties (some of which could be identified) establish the homeostasis required during branching and establishment of the adult form (Singh, 2007).

Arbors from the larval neuron are removed by pruning over the first 20 hours of pupation before the adult pattern is elaborated. The EcR-B1 isoform, whose expression is typically seen in neurons that alter their larval form and contribute to the circuitry in the adult, is detected in CSDn. Down-regulating EcR in the CSDn during metamorphosis results in a failure of remodeling and the 'adult' neuron retains a larval morphology. The detailed mechanisms by which EcR signaling acts to bring about sculpting of cell shape are not totally understood and reports on Manduca sexta indicate that steroid-induced modifications in dendritic shape can be regulated by activity-dependent mechanisms (Singh, 2007).

Studies on the cellular and molecular mechanisms of pruning events during metamorphosis could provide valuable insights into understanding of degeneration in higher systems. These events require ubiquitin-mediated proteolysis, and it is known that local activity of caspases is involved in dendritic pruning in an identified sensory neuron. Degeneration of specific branches is followed by migration of glial cells into the site of activity. The role of these glia in bringing about pruning and in clearing debris from the vicinity requires further study (Singh, 2007).

The assembly of complex circuits is dependent on a carefully orchestrated interplay of intrinsic and extrinsic cues. Does activity play a role in determining neuronal shape? Spontaneous and evoked activity in the CSDn were silenced using different methods and changes were observed in the dendritic arbors as well as in presynaptic terminals. The effects on the terminals and dendrites are possibly due to distinct mechanisms and will be discussed separately (Singh, 2007).

The strongest effects on presynaptic terminal branching were produced by expression of TeTxLC, which blocks synaptic release, and a dominant-negative Shi protein, which affects receptor-mediated endocytosis. Apart from blocking neuronal activity by abrogating synaptic vesicle release, both treatments could potentially affect axon growth. Consistent with this is the observation that TeTxLC expression affects re-growth of CSDn terminals during metamorphosis, while pruning occurred normally. Weak anatomical defects have also been described in other, non-modulatory neurons, some of which could be explained by a role in the regulation of levels of cell adhesion molecules (Singh, 2007).

Increases in size and branching pattern of the dendritic trees is a robust effect occurring notably when neuronal activity was silenced by Kir_2.1expression. In the third instar larva, expression of TNT-G leads to an increase in dendritic arbors with no significant effect on the presynaptic terminals. Expression using the RN2-Flp, Tub-FRT-CD2-FRT-Gal4, stock initiates in the fully developed larval neuron; hence, the changes in dendritic branches are likely to be a consequence of lack of neuronal activity, rather than a developmental effect. What are the mechanisms by which neuronal activity can alter morphologies of neurons? It has been demonstrated that tetanus toxin expression in motorneurons not only affects its presynaptic release because of cleavage of synaptobrevin, but also alters synaptic input by an as yet unknown mechanism. The finding of altered dendritic morphology supports the possibility that homeostatic alterations occur to compensate for a lack of activity (Singh, 2007).

A large body of data provides evidence for retrograde signaling in the development and consolidation of synapses. The observation of expanded dendritic trees upon expression of a dominant negative form of Glued, while intriguing, is difficult to explain in this light. The changes that were seen are in the dendritic (post-synaptic) field when retrograde transport is blocked cell-autonomously. While this needs further investigation, a possible explanation is that these effects are an indirect consequence of physiological alterations at the presynaptic terminals. Local morphological changes in neurons can be effected by sequestration of proteosomes and other molecules at different regions of the cell in response to activity, which could result in sculpting of cellular architecture due to altered protein composition at different cellular regions (Singh, 2007).

Defects in branching observed by abrogation of vesicle release at the synapse in a serotonergic neuron could implicate this modulator in paracrine or autocrine signaling in regulation of neuronal outgrowth, target selection and synapse formation. Such effects have been demonstrated in the gastropod Helisoma , as well as in Drosophila, where serotonin levels regulate neuronal branching and modulate the development of neuronal varicosities in the central nervous system. In these experiments, no significant changes were detected in the branching pattern of CSDn upon strong reduction of serotonin (and dopamine) using a temperature sensitive allele of dopa decarboxylase. Furthermore, unlike in M. sexta, where afferents are necessary for the formation of glomerular tufts of the serotonergic neuron within the antennal lobe, development of the CSDn occurs normally in the absence of sensory input from the antenna (Singh, 2007).

The olfactory pathway consists of afferent sensory neurons, local integrating neurons and projection neurons. Circuitry for an additional level of integration exists in the atypical projection neurons (aPNs), the antennal posterior superior protocerebral neuron (APSP), the giant symmetric relay interneurons (GSI) and the bilateral ACT relay interneurons (bACT). The architecture as well as the serotonergic nature of the CSDn closely resembles the S1 neuron in M. sexta, which receives input from bilateral projections in the protocerebrum and terminates in the lobe contralateral to the soma to modulate the activity of interneurons. It is proposed that the ipsilateral dendrites receive input from as-yet unidentified neural elements in the antennal lobe, while some axonal arbors are postsynaptic to interneurons in the calyx of the mushroom bodies and the lateral horn. It is speculated that the targets of the terminal arbors are either the PNs or the LNs since their ablation results in a reduction in branching. This architecture, which needs to be confirmed by electron microscopic analysis, provides circuitry for 'top-down' regulation of the primary olfactory center. It seems very likely that the CSDn, like its counterpart in the moth, responds to mechanosensory stimulation, providing an important role in responses to odor stimulation coupled with airflow, as would be expected in insects during flight. The modulatory effects of this large field neuron on its partners in the antennal lobe needs to be investigated by high-resolution functional imaging (Singh, 2007).

This study describes a serotonergic neuron whose anatomy suggests feedback integration within the antennal lobe of insects. The neuron undergoes remodeling during pupal life from a simple larval to a more complex adult pattern. These studies suggest that the morphology of the dendritic arbors that terminate in the lobe ipsilateral to the soma is regulated by neuronal activity. The arborization of terminal arbors depends on vesicle recycling, endocytosis and Dynein-dependant retrograde transport. These findings demonstrate a useful identified-neuronal preparation where developmental mechanisms and remodeling can be studied in the context of olfactory behavior (Singh, 2007).

Drosophila olfactory receptor neurons project to the antennal lobe, the insect analog of the mammalian olfactory bulb. GABAergic synaptic inhibition is thought to play a critical role in olfactory processing in the antennal lobe and olfactory bulb. However, the properties of GABAergic neurons and the cellular effects of GABA have not been described in Drosophila, an important model organism for olfaction research. Whole-cell patch-clamp recording, pharmacology, immunohistochemistry, and genetic markers have been used to investigate how GABAergic inhibition affects olfactory processing in the Drosophila antennal lobe. This study shows that many axonless local neurons (LNs) in the adult antennal lobe are GABAergic. GABA hyperpolarizes antennal lobe projection neurons (PNs) via two distinct conductances, blocked by a GABA_A- and a GABA_B-type antagonist, respectively. Whereas GABA_A receptors shape PN odor responses during the early phase of odor responses, GABA_B receptors mediate odor-evoked inhibition on longer time scales. The patterns of odor-evoked GABA_B-mediated inhibition differ across glomeruli and across odors. LNs display broad but diverse morphologies and odor preferences, suggesting a cellular basis for odor- and glomerulus-dependent patterns of inhibition. Together, these results are consistent with a model in which odors elicit stimulus-specific spatial patterns of GABA release, and as a result, GABAergic inhibition increases the degree of difference between the neural representations of different odors (Wilson, 2005).

Smell begins when odor molecules interact with olfactory receptor neurons (ORNs). ORNs then project to the brain following anatomical rules common to species as evolutionarily distant as flies and rodents. Briefly, the odor sensitivity of a particular ORN is specified by the expression of a single olfactory receptor gene. All the ORNs that express a particular receptor send their axons to the same glomeruli in the brain. There, ORNs make synapses with second-order neurons [mitral cells (in vertebrates) or projection neurons (in insects)] (Wilson, 2005).

What happens when signals reach these second-order olfactory neurons is determined by complex local circuitry. One obstacle to understanding this circuitry is the sheer number of input channels in the mammalian olfactory system. The rat olfactory bulb contains ~1000 glomeruli; in contrast, the Drosophila antennal lobe contains just ~40 glomeruli. This, along with the genetic advantages of Drosophila, makes the fruit fly a useful model for investigating olfactory processing (Wilson, 2005).

A given odor excites many Drosophila antennal lobe projection neurons (PNs) but inhibits others. These odor-evoked inhibitory epochs can last from ~100 ms to several seconds. Similar odor-evoked inhibition has also been observed in other insects and in olfactory bulb mitral cells. Some odor responses of mitral cells and PNs are purely inhibitory. Other responses are multiphasic, in which an inhibitory epoch follows or precedes an excitatory epoch. These temporal patterns are cell and odor dependent and have been proposed to encode information about the stimulus. However, the mechanism of these 'slow' patterns is not fully understood (Wilson, 2005).

One possibility is that inhibitory epochs represent periods when principal neurons are synaptically inhibited by GABAergic local neurons (LNs). GABA-immunoreactive LNs are present in the adult antennal lobe of several species and in the larval Drosophila antennal lobe. Antennal lobe LNs can synaptically inhibit PNs, and the antennal lobe is strongly immunoreactive for GABA_A receptors). However, GABA_A antagonists do not block odor-evoked slow inhibition or slow temporal patterns in PNs. Therefore, these inhibitory epochs have been hypothesized to reflect a metabotropic conductance or the action of a different inhibitory neurotransmitter. Alternatively, inhibition of PNs could be caused by inhibition of ORNs (Wilson, 2005).

This study investigated the mechanisms of odor-evoked inhibition in PNs. It was confirmed that many Drosophila antennal lobe LNs are GABAergic. GABA receptors contribute to odor-evoked inhibition of PNs on both fast and slow time scales, and GABA-mediated slow inhibition increases the diversity of odor-evoked responses among PNs. This is consistent with models that invoke GABAergic inhibition to increase the discriminability of olfactory representations (Wilson, 2005).

As in the olfactory bulb, each glomerulus in the Drosophila antennal lobe contains four main classes of neurons: (1) the axon terminals of ORNs, (2) the dendrites of PNs that convey information from ORNs to higher brain centers, (3) neurites from LNs that interconnect glomeruli, and (4) the centrifugal axonal projections of neurons that relay information to the antennal lobe from higher brain centers. Recent studies have illuminated the development, morphology, and physiology of Drosophila ORNs and PNs. Drosophila LNs, in contrast, have not received much attention. LNs have been noted in Golgi-impregnated antennal lobes, but remarkably little is known about the number, morphology, and connectivity of these cells or about their impact on other antennal lobe neurons. If adult LNs are also GABAergic, and if GABA is inhibitory (as it is in other insects), then LNs could participate in sculpting the inhibitory epochs prominent in many PN odor responses. In the larval Drosophila antennal lobe, many LNs are immunopositive for GABA. In the adult, it has been shown that many somata around the antennal lobes express the GABA biosynthetic enzyme glutamic acid decarboxylase (Wilson, 2005).

This study confirms that many adult Drosophila antennal LNs are GABAergic. Using confocal immunofluorescence microscopy with an anti-GABA antibody, many GABA-positive somata were observed in the vicinity of the antennal lobe neuropil. To identify LNs, flies were used in which a large subpopulation of these cells were genetically labeled. In these flies (GAL4 enhancer trap line GH298), reporter gene activity labels a cluster of somata lateral to the antennal lobe neuropils. The neurites of these neurons collectively fill the antennal lobes, reminiscent of the morphology of LNs identified in Golgi impregnations. When whole-cell patch-clamp recordings were made from the somata of GFP-positive cells in GH298-GAL4, UAS-CD8GFP flies, intrinsic properties characteristic of LNs were observed, namely high input resistances and action potentials with amplitude >40 mV. It was also confirmed with single-cell biocytin fills that these GFP-positive neurons were indeed LNs. When GH298-GAL4,UAS-CD8GFP brains stained for GABA were visualized using dual-channel confocal microscopy, it was found that most GFP-positive somata were also GABA positive. About one-fifth of the GFP-positive somata did not stain for GABA. These neurons may contain a different neurotransmitter, or the staining may not have been sensitive enough to detect low levels of GABA. The possibility cannot be excluded that these GABA-negative neurons are not LNs (Wilson, 2005).

It was then confirmed that GABA hyperpolarizes antennal lobe neurons. In LNs, these results imply that inhibition is mediated entirely by GABA_A receptors. In contrast, GABAergic inhibition of PNs is mediated by both GABA_A and GABA_B receptors. Thus, synaptic inhibition onto PNs and LNs is functionally specialized (Wilson, 2005).

How might GABAergic inhibition contribute to olfactory processing in the Drosophila antennal lobe? Recent studies using optical measurements of neural activity have concluded that ORN and PN odor responses are very similar and that the antennal lobe is merely a relay station that faithfully transmits ORN signals to PNs without alteration. These conclusions imply that synaptic inhibition in the antennal lobe may exist merely to control global excitability and may not play an important role in representing information about the stimulus. However, the optical reporters used in these studies lack temporal resolution, have limited dynamic range, and may not be sensitive to inhibitory events. Whole-cell patch-clamp recordings from Drosophila PNs show prominent inhibitory epochs in many odor responses, generating odor-dependent spatiotemporal response patterns. Such complex temporal patterns are not present in the responses of ORNs, implying that they arise in the antennal lobe and thus represent a transformation of the olfactory code between the first and second layers of olfactory processing. These temporal patterns are reminiscent of those seen in olfactory bulb mitral cells and in other insects (Wilson, 2005).

A common notion in olfaction is that such spatiotemporal patterns represent lateral interactions, the net effect of which is to amplify contrast. This idea has taken two main forms. The first proposes a contrast-enhancement mechanism akin to that seen in the retina. According to this model, specific mutual inhibitory interactions exist between principal neurons in nearby glomeruli with similarly tuned ORN inputs. When a principal neuron is activated strongly by an odor, it will trigger lateral inhibition of its neighbors to suppress weak responses to that odor, sharpening the difference between their tuning curves. A different hypothesis is that lateral interactions exist in a more distributed manner. Odors are represented as stimulus-specific sequences of neuronal ensembles. The stimulus is represented both by the identity of the active neurons and the time when they are active. According to this model, the net effect of interglomerular interactions is not to prune away weak responses. Rather, inhibitory interactions may coexist with excitatory interactions (or relief-of-inhibition mechanisms), such that new principal neuron responses appear as others disappear. Because each stimulus is represented by an evolving neural ensemble, the available coding space is expanded. Again, the outcome of this process is thought to be a progressive decorrelation, such that overlap is reduced between stimulus representations (Wilson, 2005).

Both these models predict that eliminating odor-evoked inhibitory epochs in second-order olfactory neurons will increase the similarity between the spatiotemporal activity patterns produced in these neurons by different odors. This study reports that odor-evoked inhibitory epochs in Drosophila PNs are mostly suppressed by a GABA_B receptor antagonist and that blocking GABA_B receptors decreases the coefficient of variation among PN peristimulus-time histograms. These results are consistent with models in which lateral interactions between principal and local neurons increase the degree of difference between the neural representations of different odors (Wilson, 2005).

It is important to point out that the effect of the GABA_B antagonist on PN odor responses may be mediated partly by presynaptic effects on ORN axon terminals or by indirect effects via other excitatory inputs to PNs. Determining the locus of this effect will require additional experiments using cell type-specific genetic manipulations. However, because GABA_B receptors mediate much of the direct effect of GABA on PNs, it seems likely that the effect of CGP54626, a compound that blocks the late inhibitory epoch in a PN odor response, on odor-evoked PN activity is attributable at least in part to postsynaptic GABA_B receptors (Wilson, 2005).

Finally, it should be noteed that two conceptually distinct kinds of temporal patterns can in principle coexist among second-order olfactory neurons. Slow temporal patterns are punctuated by inhibitory epochs on the timescale of tens to thousands of milliseconds. In this study, it was shown that these slow patterns in the Drosophila antennal lobe are sensitive to a GABA_B antagonist. Distinct from this is fast inhibition, which synchronizes the firing of principal neurons on time scales of several milliseconds and is sensitive to picrotoxin. Fast, odor-evoked synchronous oscillations occur in the olfactory systems of many organisms and are required for fine olfactory discrimination in the honeybee. There is little evidence for such oscillatory synchronization among Drosophila PNs. These observations deserve additional investigation but suggest that different organisms may emphasize different strategies for olfactory processing (Wilson, 2005).

Theoretical models of olfactory processing that invoke synaptic inhibition to increase the contrast between different stimulus representations presume nonuniform connectivity between inhibitory and principal neurons. In insects, a GABAergic LN can arborize across the entire antennal lobe, and so it is not obvious that single LNs will make connections preferentially with particular glomeruli. In this study, it was found that the neurites of single LNs form spatially heterogeneous patterns in the antennal lobe. This finding alone does not prove that individual LNs make connections preferentially in the glomeruli in which their dendrites are most dense; for example, average synaptic strength could be higher in glomeruli with fewer neurites. However, individual LNs also displayed specific odor preferences. This supports the idea that the odor tuning of individual LNs might be correlated with which glomeruli were preferentially innervated by that LN. According to this model, LN odor tuning would be biased toward the tuning of the excitatory neurons innervating those glomeruli. Drosophila LNs receive excitatory input from PNs. In other insect species, LNs are also known to receive direct input from ORNs (Wilson, 2005).

Consistent with these conclusions, a functional imaging study of the Drosophila antennal lobe has found that each odor stimulus evokes GABA release in some glomeruli more than others. Furthermore, these spatial patterns of GABA release are odor dependent (Ng, 2002). That study measured synaptic release from all GABAergic neurons simultaneously. This investigation has now been extended to single LNs, the morphological and functional diversity of which suggests a cellular mechanism for how the pattern of GABA release can be nonuniform and odor dependent. Ultimately, a test of this idea should come from correlating the morphology of single LNs with their odor preferences. Recent studies have reported the odor tuning of a large subset of Drosophila olfactory receptors and the mapping of each receptor to a specific ORN type. Once it is know which ORN type corresponds to each glomerulus, it should be possible to design experiments of this type more systematically (Wilson, 2005).

Neural circuits are often remodeled after initial connections are established. The mechanisms by which remodeling occurs, in particular whether and how synaptically connected neurons coordinate their reorganization, are poorly understood. In Drosophila, olfactory projection neurons (PNs) receive input; their dendrites synapse with olfactory receptor neurons in the antennal lobe and relay information to the mushroom body (MB) calyx and lateral horn. Embryonic-born PNs participate in both the larval and adult olfactory circuits. In the larva, these neurons generally innervate a single glomerulus in the antennal lobe and one or two glomerulus-like substructures in the MB calyx. They persist in the adult olfactory circuit and are prespecified by birth order to receive input from a subset of glomeruli distinct from larval-born PNs. Developmental studies indicate that these neurons undergo stereotyped pruning of their dendrites and axon terminal branches locally during early metamorphosis. Electron microscopy analysis reveals that these PNs synapse with MB gamma neurons in the larval calyx and that these synaptic profiles are engulfed by glia during early metamorphosis. As with MB gamma neurons, PN pruning requires cell-autonomous reception of the nuclear hormone ecdysone. Thus, these synaptic partners are independently programmed to prune their dendrites and axons (Marin, 2005).

One of the best-studied examples of neuronal reorganization in an insect brain is the gamma neuron of Drosophila mushroom bodies (MBs). MB gamma neurons are born during embryonic and early larval stages. They send dendrites into the MB calyx and axons into larval medial and dorsal MB axon lobes. During early metamorphosis, gamma neurons prune their larva-specific dendrites and axon branches before re-extending adult-specific processes. What happens to their synaptic partners while MB gamma neurons reorganize their dendrites and axons? In this study, it was shown that a subset of olfactory projection neurons -- the major presynaptic partners of MB gamma neurons -- are also morphologically differentiated to function in both larva and adult. The reorganization of these neurons during metamorphosis is independently controlled by some of the same molecular mechanisms as that of the MB gamma neurons (Marin, 2005).

In the adult fly, odors are detected by olfactory receptors (ORs) on the dendrites of about 1300 olfactory receptor neurons (ORNs) in the antennae and maxillary palps. In general, each ORN appears to express one of ~45 possible OR types, and the axons of all ORNs expressing a given OR converge to one of ~45 stereotypical glomeruli in the antennal lobe (AL), the equivalent of the mammalian olfactory bulb. From there, 150-200 projection neurons (PNs) relay olfactory activity to higher brain centers, the MB calyx and the lateral horn (LH) of the protocerebrum. Systematic clonal analysis using the MARCM method to label single and clonally related clusters of PNs that express the GAL4 driver GH146 revealed that these PNs are prespecified by lineage and birth order to receive input via their dentrites from particular glomeruli in the adult AL. Moreover, each glomerular class of PNs exhibits a characteristic axon branching pattern in the LH, suggesting stereotyped targets in at least one higher olfactory center (Marin, 2005).

The Drosophila larval olfactory system is much smaller and simpler by comparison, shown to consist of only 21 ORNs in the dorsal organ and believed to include ~50 PNs relaying information to the larval MB and LH. Developmental analysis has shown that the PNs born during larval stages exhibit only a single unbranched process from the cell body to the MB calyx until early metamorphosis, when dendrites and axon terminal branches start to elaborate. Thus, larval-born PNs do not participate in the larval olfactory circuit (Marin, 2005).

What, then, is the origin of the relay interneurons that connect the larval AL to higher olfactory centers? Do they contribute to the adult olfactory system as well? This study shows that, in contrast to the larval-born PNs, PNs generated during embryogenesis exhibit morphologically differentiated dendrites and axons in both larva and adult. These neurons prune their processes locally during the first few hours of metamorphosis and later re-extend them to innervate developing adult structures. This pruning process is regulated by ecdysone and TGFß signaling, as has been demonstrated previously for MB gamma neurons. Thus, developmentally programmed remodeling allows these embryonic-born PNs to participate in two distinct olfactory circuits at two different stages in the Drosophila life cycle (Marin, 2005).

The MARCM method allows the labeling of a single neuron, or all neurons born from the same neuroblast, that express a particular GAL4 driver. These studies focus on the ~90 (out of an estimated total 150-200) PNs that express GAL4-GH146. A heatshock-promoter-driven FLP recombinase was used to control the timing of the mitotic recombination that results in labeled MARCM clones. In a previous study (Jefferis, 2004), at least one GAL4-GH146-positive PN was identified that could only be labeled by heatshock-induced mitotic recombination during embryogenesis. This embryonic-born PN specifically targeted its dendrites to glomerulus VA2, one of many glomeruli innervated when labeling the entire population of GH146 PNs, yet never by PN single-cell or neuroblast clones labeled by heatshock during larval stages. This discrepancy in the number of innervated glomeruli suggested that a fraction of adult AL glomeruli were being targeted by a subset of PNs born during embryogenesis (Marin, 2005).

To study the embryonic-born neurons labeled by the GH146 driver, MARCM clones were systematically generated by heatshock induction at embryonic stages. Large anterodorsal neuroblast clones labeled by heatshock early in embryogenesis innervated at least 15 glomeruli not targeted by either the anterodorsal or lateral neuroblast clones labeled by heatshocking newly hatched larvae. MARCM single-cell clones were used to characterize embryonic-born PNs that innervate eight different landmark glomeruli in the adult AL. The gross morphology of these PNs in the adult brain is quite similar to that of the larval-born PNs previously described: each PN generally innervates a single glomerulus in the antennal lobe (distinct from those innervated by larval-born PNs), then sends its axon via the inner antennocerebral tract to display a characteristic terminal branching pattern in the LH according to its glomerular class, along with a number of collaterals in the MB calyx that end in prominent boutons (Marin, 2005).

By comparing the specific glomeruli innervated in each partial anterodorsal neuroblast clone generated by heatshock at different times during embryogenesis, it was ascertained that: (1) these embryonic-born PNs were generated in the order DP1m, VL2p, VA6, VA2, DL5, DM3, VM3 and finally DL6, and (2) every clone labeled by embryonic heatshock included all of the larval-born anterodorsal PNs analyzed in the previous study, indicating that both PN subsets originate from the same neuroblast. Upon generation of the DL6 PN(s), the anterodorsal neuroblast apparently arrests, producing additional projection neurons later only in larval life (as indicated by heatshock-induced labeling of just a single anterodorsal glomerular class, DL1, until about 36 hours after larval hatching) (Marin, 2005).

In summary, embryonic-born PNs look just like larval-born PNs with regard to both their dendritic and axonal projections in the adult brain. Moreover, since their dendrites target a distinct subset of AL glomeruli and their axons exhibit characteristic terminal branching patterns in the LH according to their glomerular classes, these embryonic-born PNs serve to expand the repertoire of odor representation in adults beyond the larval-born PNs previously characterized (Marin, 2005).

Given their early origin, these GH146-positive embryonic-born PNs may participate in the larval olfactory circuit as well. Indeed, examining third instar larval brains reveals that GH146 is strongly expressed in presumptive projection neurons that appear to innervate the larval AL and to send axons up to the MB calyx and larval equivalent of the adult LH. These projections appear to be contributed by about 16 to 18 clustered neurons that are presumably derived from the anterodorsal neuroblast (Marin, 2005).

To examine the morphology and connectivity of these PNs in the larval olfactory system with greater resolution, the MARCM method was used to specifically label PNs generated prior to larval hatching and brains were dissected from wandering third instar larvae. In contrast to the larval-born PNs analyzed in earlier studies (Jefferis, 2004), all anterodorsal embryonic-born PNs exhibited densely branched dendrites in the larval AL and axons with large synaptic structures targeting glomerulus-like subregions in the MB calyx as well as branches in the presumptive LH. The large majority of the anterodorsal embryonic-born PNs each targeted a single glomerulus in the LAL and/or in the MB. In some cases, individual PNs targeted two glomeruli in one or both structures (Marin, 2005).

Several lines of evidence suggest that the embryonic-born PNs observed in the larval olfactory system are the same cells as the PNs that contribute to the much larger and more complex adult circuit. (1) The frequencies of labeled single-cell clones are comparable between the two stages, arguing against the possibilities that embryonic-born PNs are either dying off during metamorphosis or remaining quiescent and undetected through larval life. (2) The numbers of GH146-positive PNs observed at the time of puparium formation and in the adult are similar. (3) Most importantly, each embryonic-born PN undergoes characteristic morphological changes during metamorphosis. Therefore, the PNs labeled by embryonic heatshock are referred to as persistent projection neurons (PPNs). However, at this point, the methods do not allow correlation of specific glomerular classes in larva with those observed later in adulthood (Marin, 2005).

Prior studies have used MB gamma neurons as a model system to study the molecular mechanisms of axon pruning. γ neuron pruning depends on cell-autonomous reception of the steroid hormone ecdysone; single neurons that are homozygous mutant for the ecdysone co-receptor ultraspiracle (usp) in an otherwise heterozygous brain fail to reorganize their processes and retain both dorsal and medial axon lobes in the adult brain. In addition, gamma neurons must upregulate the expression of ecdysone receptor isoform B1 (EcRB1) prior to axon pruning. This upregulation requires TGFß signaling; MB gamma neurons that are mutant for the TGFß/Activin Type I receptor baboon (babo) or its downstream effector mad do not upregulate EcRB1 expression and consequently fail to prune (Marin, 2005 and references therein).

It was asked whether a similar molecular pathway is utilized during PPN reorganization. To ascertain whether the pruning of PPNs is also regulated by ecdysone, EcRB1 expression patterns were analzyed. At puparium formation, only 20 of the ~90 GH146+ projection neurons present were strongly positive for EcRB1. These strongly stained PNs, ~18 of which belonged to the anterodorsal cluster, also had noticeably larger and brighter cell bodies than surrounding PNs, which were probably immature larval-born PNs. Single-cell MARCM clones generated by embryonic heatshock were also strongly positive for EcRB1 at puparium formation. Thus, it is concluded that EcRB1 expression is highly expressed in PPNs at the onset of metamorphosis (Marin, 2005).

Is TGFβ signaling generally required to regulate expression of EcRB1 for neuronal pruning during metamorphosis? MARCM was used to label cells that were homozygous for the strongest baboon allele, babo^Fd4, in a heterozygous background to test whether PPNs also require TGFß reception for normal pruning. At the wandering third instar stage, PPNs homozygous for babo^Fd4 appeared to have normal dendritic and axonal projections. However, babo^Fd4 PPNs failed to show high-level expression of EcRB1 by the onset of puparium formation. This result indicates that, as for the MB gamma neurons, high-level expression of EcRB1 in remodeling PPNs depends on TGFß signaling (Marin, 2005).

Consistent with the loss of EcRB1 expression, babo^Fd4 PPNs fail to reorganize their processes normally during the first few hours of metamorphosis. For wild-type PPNs at 8 hours APF, approximately 95% of dendrites, 80% of MB calyx processes, and 85% of LH processes are in the final two stages of pruning. However, most of the embryonic-born babo^Fd4 PPNs still retain dendrites and axons with larval morphology at this time. Dense dendritic processes were visible in the larval AL for 100% of PN clones examined, and only 14% of axon branches in the LH appeared to resemble the final two stages of pruning. The degree of pruning in the calyx was more difficult to estimate, due to the concurrent degeneration of gamma MB neuron dendrites and loss of glomerular organization, but disappearance of synaptic boutons still seemed inhibited (Marin, 2005).

To confirm that this failure to prune resulted from loss of ecdysone signal reception, MARCM was used to label PPNs that were homozygous for a well-characterized mutant allele of the ecdysone co-receptor, usp³. At the wandering third instar stage, usp³ PPNs exhibit normal morphology, and, as expected, EcRB1 was expressed at wild-type levels at the time of puparium formation (Marin, 2005).

However, when these brains were examined at 8 hours APF, a significant defect in dendrite and axon pruning was observed. In the majority of cases, both dendritic densities in the location of the larval antennal lobe and axon branches in the MB calyx and LH had been retained. Taken together, these mosaic experiments suggest that PPN dendritic and axonal pruning require cell-autonomous function of EcRB1/USP, as has been shown previously for MB gamma neurons (Marin, 2005).

What are the consequences for the adult olfactory circuit when larval circuits fail to prune? PPNs homozygous for usp³ or babo^Fd4 that failed to prune their dendrites and axons during metamorphosis allowed investigation of this question. When examined in adults, wild-type PPN dendrites were confined to a single glomerulus in the adult AL with the exception of the VL2p+ class. Dendrites of single-cell PPN clones homozygous for usp³ generally appeared to target glomeruli in the adult AL appropriate for PPNs; however, ectopic processes in additional areas of the AL, which could be interpreted as persisting larval dendrites, were often present. In a few cases, usp³ PPN dendrites were sparser and less specifically targeted to particular glomeruli, but still remained somewhat confined to certain regions of the AL. Likewise, whereas wild-type PPNs always exhibit terminal swellings on short side branches, about 40% of usp³ PPNs retained larval-like boutons directly on their main trunks in the MB calyx; however, they always had side branches with terminal swellings as well, implying that re-extension and adult-specific outgrowth were not completely impaired. In addition, the main axon trunk often diverted conspicuously from the inner antennocerebral tract in the MB calyx, presumably to maintain contact with the larval boutons. Nearly all usp³ PPN axons exhibited grossly wild-type morphologies in adult LH; only one usp³ PPN axon in the sample failed to enter the LH. In summary, usp³ PPNs display ectopic processes in AL and MB that appear to be due to defects in pruning during early metamorphosis; these pruning defects do not seem to interfere with the growth or even targeting (in the case of AL) of adult-specific processes (Marin, 2005).

In comparison, babo^Fd4 PPNs exhibited more severe dendritic and axonal phenotypes in the adult brain. In a few cases, these PPNs had targeted an appropriate glomerulus but also featured ectopic processes. More commonly, sparse diffuse processes were observed in the AL that were somewhat localized but did not appear to target any specific glomerulus. Processes also occasionally strayed to arborize outside the ventral AL. In the most severe cases, sparse dendrites were distributed broadly throughout the AL. In the MB calyx, all babo^Fd4 PPN axons appeared to have retained large larval-like boutons directly on their main trunks, rather than exclusively terminal swellings on short side branches as in wild type; in 64% of cases, there were no MB collaterals with wild-type adult appearance at all. The main axon trunk often diverged dramatically from the inner antennocerebral tract in the MB calyx. Finally, in the LH, the majority of babo^Fd4 PPNs featured significant aberrations including unusually profuse swellings along the branches, failure to enter the LH and/or failure to elaborate higher order branches in the LH. These phenotypes imply an axon re-extension, pathfinding and/or targeting defect in addition to the impaired pruning observed at 8 hours APF (Marin, 2005).

In summary, both usp³ and babo^Fd4 PPNs exhibit phenotypes in the adult brain consistent with blockage of pruning during early metamorphosis, including extraglomerular processes in the AL as well as large larval-like boutons on the main trunk and diversion from the inner antennocerebral tract as the axon passes through the MB calyx. However, babo^Fd4 PPNs also feature more severe phenotypes, particularly a complete lack of glomerular innervation and of adult-like axon collaterals with terminal swellings in the MB calyx, as well as failure to enter the LH and/or to elaborate higher order terminal branches. These latter phenotypes appear to be qualitatively different from those attributable to a simple loss of pruning, suggesting that TGFß signaling via baboon may have an additional role in re-extension and/or adult-specific targeting during metamorphosis (Marin, 2005).

In this study, the PPNs of the Drosophila olfactory system have been shown to play analogous functions in two neural circuits at different life stages. They do so by developmentally programmed disassembly and reassembly of synaptic connections during metamorphosis. The implications of these findings for the larval and adult olfactory systems and to neural circuit reorganization are discussed below (Marin, 2005).

Therefore, PPNs serve as relay interneurons connecting the antennal lobe to the MB calyx and the presumptive LH in larvae, just as previously characterized larval-born projection neurons do in adults. Each PPN generally targets its dendrites to one glomerular substructure in the larval AL, probably receiving input from one of the 21 olfactory receptor neurons of the dorsal organ. From there, the PPN's axon extends to higher brain centers, forming one or two large synaptic structures en passant on its way through the MB calyx to the LH. Electron microscopy studies with genetically encoded markers expressed separately in PPNs or in MB gamma neurons established that PPNs form functional synapses in the larval circuit and that MB gamma neurons are among their postsynaptic partners (Marin, 2005).

This analysis of these PPNs in the adult olfactory circuit confirmed and extended the developmental and wiring logic derived from previous analysis of larval-born PNs. Just like larval-born PNs, embryonic-born PPNs are prespecified to target their dendrites to particular glomeruli according to their birth order. Specifically, most PPNs are derived from the same anterodorsal neuroblast that later gives rise to about half the GH146-positive PNs. Like the larval-born PNs, PPNs exhibit stereotyped terminal arborization patterns in the LH. Interestingly, in the adult AL, PPNs innervate a distinct subset of glomeruli from either their larval-born anterodorsal cousins or the projection neurons generated by the lateral neuroblast. This indicates that, in addition to relaying activity from larva-specific olfactory receptor neurons earlier in development, PPNs expand the olfactory repertoire of the adult circuit (Marin, 2005).

In addition to serving larval-specific functions, one proposed function for larval circuits is to provide a foundation upon which adult circuits can be built. In the case of the olfactory circuit, however, previous analysis indicates that the adult-specific antennal lobes form adjacent to, but spatially distinct from, the larval antennal lobe. Analysis of PPN remodeling supports the notion that the adult circuit is constructed de novo rather than upon the larval circuit. A developmental timecourse analysis revealed that PPNs prune their dendrites and axon branches during early metamorphosis, so that only the main unbranched process from the cell body to the distal edge of the calyx remains by 12 hours APF. By contrast, the larval-born PNs begin to elaborate dendrites at the onset of puparium formation and restrict their processes to specific regions of the developing AL between 6 and 12 hours APF. Persistent projection neurons start exhibiting this type of localized dendritic outgrowth in the adjacent but distinct adult AL site only at 18 hours APF, around the time that adult-specific ORN axons arrive but prior to their invasion of the AL. This strongly implies that, far from providing contact-mediated cues for differentiating larval-born PNs, PPNs target glomeruli in the developing AL only after the larval-born PNs have established their dendritic target domains. The finding that PPN-specific glomeruli are intercalated with those targeted by dendrites of larval-born PNs, rather than occupying a spatially segregated domain in the adult antennal lobe, implies complex targeting rules in the establishment of wiring specificity of the adult circuit (Marin, 2005).

The fact that PPNs have clearly identifiable addresses for their dendritic targeting in the adult circuit suggested an interesting question: does assembly of the adult circuit depend on the disassembly of the larval circuit? The data suggest that neuronal reorganization appears to be separable into two at least partially independent events, pruning and re-extension. Even usp³ PPNs whose larva-specific dendrites and axons appear unpruned still exhibit the random fine filopodial extensions characteristic of wild-type neurons at 8-12 hours APF, and moreover target their new dendrites to appropriate adult antennal lobe glomeruli, as well as exhibiting adult-specific axon collaterals in the MB calyx and grossly wild-type terminal branches in the LH (Marin, 2005).

The fact that most usp³ persistent PNs still innervate appropriate glomeruli in the adult antennal lobe and have axons with adult characteristics would suggest that ultraspiracle-mediated execution of ecdysone signaling is required for pruning but not for responding to re-extension and/or targeting cues in the developing brain. However, most babo^Fd4 PPNs failed to target appropriately in the adult olfactory system. This difference in phenotypes may be due to differential perdurance of wild-type Usp versus Babo protein in single-cell MARCM clones and/or to differences in the severity of the alleles examined, consistent with the observation that babo^Fd4 PPNs show slightly more homogeneous pruning phenotypes at 8 hours APF. However, usp³ carries a missense mutation that alters an invariant arginine in the DNA-binding domain and blocks MB gamma neuron pruning completely. Thus, the possibility is favored that baboon is required for additional ultraspiracle-independent functions during metamorphosis, in the initiation of pruning, re-extension and/or targeting of adult olfactory structures (Marin, 2005).

Small poikilotherms such as the fruit fly Drosophila depend on absolute temperature measurements to identify external conditions that are above (hot) or below (cold) their preferred range and to react accordingly. Hot and cold temperatures have a different impact on fly activity and sleep, but the circuits and mechanisms that adjust behavior to specific thermal conditions are not well understood. This study use patch-clamp electrophysiology to show that internal thermosensory neurons located within the fly head capsule (the AC neurons(1)) function as a thermometer active in the hot range. ACs exhibit sustained firing rates that scale with absolute temperature-but only for temperatures above the fly's preferred ~25°C (i.e., "hot" temperature). ACs were identified in the fly brain connectome and demonstrate that they target a single class of circadian neurons, the LPNs.(2) LPNs receive excitatory drive from ACs and respond robustly to hot stimuli, but their responses do not exclusively rely on ACs. Instead, LPNs receive independent drive from thermosensory neurons of the fly antenna via a new class of second-order projection neurons (TPN-IV). Finally, silencing LPNs blocks the restructuring of daytime "siesta" sleep, which normally occurs in response to persistent heat. Previous work described a distinct thermometer circuit for cold temperature.(3) Together, the results demonstrate that the fly nervous system separately encodes and relays absolute hot and cold temperature information, show how patterns of sleep and activity can be adapted to specific temperature conditions, and illustrate how persistent drive from sensory pathways can impact behavior on extended temporal scales (Alpert, 2022).

The detection of environmental temperature and regulation of body temperature are integral determinants of behaviour for all animals. These functions become less efficient in aged animals, particularly during exposure to cold environments, yet the cellular and molecular mechanisms are not well understood. This study identifies an age-related change in the temperature preference of adult fruit flies that results from a shift in the relative contributions of two parallel mushroom body (MB) circuits-the β'- and β-systems. The β'-circuit primarily controls cold avoidance through dopamine signalling in young flies, whereas the β-circuit increasingly contributes to cold avoidance as adult flies age. Elevating dopamine levels in β'-afferent neurons of aged flies restores cold sensitivity, suggesting that the alteration of cold avoidance behaviour with ageing is functionally reversible. These results provide a framework for investigating how molecules and individual neural circuits modulate homeostatic alterations during the course of senescence (Shih, 2015).

Circadian clocks are endogenous timers adjusting behaviour and physiology with the solar day. Visual and non-visual photoreceptors are responsible for synchronizing circadian clocks to light, but clock-resetting is also achieved by alternating day and night temperatures with only 2°-4°C difference. This study shows that Drosophila Ionotropic Receptor 25a (IR25a) is required for behavioural synchronization to low-amplitude temperature cycles. This channel is expressed in sensory neurons of internal stretch receptors previously implicated in temperature synchronization of the circadian clock. IR25a is required for temperature-synchronized clock protein oscillations in subsets of central clock neurons. Extracellular leg nerve recordings reveal temperature- and IR25a-dependent sensory responses, and IR25a misexpression confers temperature-dependent firing of heterologous neurons. It is proposed that IR25a is part of an input pathway to the circadian clock that detects small temperature differences. This pathway operates in the absence of known 'hot' and 'cold' sensors in the Drosophila antenna, revealing the existence of novel periphery-to-brain temperature signalling channels (Chen, 2015).

In Drosophila, daily activity rhythms are controlled by a network of ~150 clock neurons expressing the clock genes period (per) and timeless (tim). These encode repressor proteins that negatively feedback on their own promoters resulting in 24h oscillations of clock molecules. Temperature cycles (TC) synchronize molecular clocks present in peripheral appendages in a tissue-autonomous manner, whereas synchronization of clock neurons in the brain mainly depends on peripheral temperature receptors located in the chordotonal organs (ChO) and the ChO-expressed gene no circadian temperature entrainment (nocte) (Chen, 2015).

To discover novel factors involved in temperature entrainment, Nocte-interacting proteins were identified by co-immunoprecipitation and mass-spectrometry. Focus was placed on IR25a, a member of a divergent subfamily of ionotropic glutamate receptors, and the interaction by co-immunoprecipitation was varfied after overexpressing IR25a and Nocte in all clock cells using tim-gal4. IR25a is expressed in different populations of sensory neurons, including those in the antenna and labellum. In the olfactory system IR25a acts as a co-receptor with different odour-sensing IRs (Abuin, 2011; Chen, 2015 and references therein).

To investigate if IR25a is co-expressed with nocte in ChO, IR25a expression in femur and antennal ChO was analyzed using an IR25a-gal4 line (Abuin, 2011). IR25a-gal4-driven mCD8-GFP labelled subsets of ChO neurons in the femur, overlapping substantially with nompC-QF driven QUAS-Tomato signals (using the QF binary transcriptional activation system). nompC-QF is expressed in larval ChO18 and in the adult femur ChO. Comparison of IR25a-driven mCD8-GFP and nuclear DsRed signals with those of other ChO neuron drivers suggests that IR25a is expressed in a subset of femur ChO neurons and Johnston's Organ (JO) neurons. To determine if IR25a-gal4 ChO signals reflect endogenous IR25a expression, the presence of IR25a mRNA in the femur and leg was confirmed, and the co-localization was confirmed of anti-IR25a immunofluorescence signals in femur ChO neurons. IR25a was detected in ChO neuron cell bodies and ciliated dendrites, as was an mCherry-IR25a fusion protein expressed in these cells (Chen, 2015).

As nocte¹ mutants do not synchronize to 12h-12-h 16°C:25°C temperature cycles in constant light (LL), IR25a^-/- mutants were analyzed under these conditions. Unlike nocte¹, the IR25a^-/- flies synchronized well to this regime, and similar results were obtained at warmer temperature cycles. To test whether IR25a is specifically required for synchronization to small temperature intervals, IR25^-/- flies were subjected to various temperature cycles with an amplitude of only 2°C. Surprisingly, and in contrast to wild-type, IR25^-/- mutants did not synchronize to any of the shallow temperature cycles in LL or constant darkness (DD). In LL, wild-type and IR25a rescue flies showed a clear activity peak in the second part of the warm period before and after the 6h shift of the temperature cycle. By contrast, IR25a^-/- mutants were constantly active throughout the temperature cycle, apart from a short period of reduced activity at the beginning of the warm phase of TC1. In DD, control flies slowly advanced (or delayed) their evening activity peak during phase-advanced (or delayed) temperature cycles. The phase of this activity peak was maintained in the subsequent free-running conditions (DD, constant 25°C) indicating stable re-entrainment of the circadian clock. By contrast, IR25a mutants did not shift their evening peak during the temperature cycle, keeping their original phase throughout the experiment (Chen, 2015).

To quantify entrainment in LL, the 'entrainment index' (EI) was determined, whereas for most DD experiments the phase difference of the main activity peak upon release into constant conditions between IR25a mutants and controls was calculated. In all 2°C amplitude temperature cycles tested the entrainment index of IR25a^-/- flies was significantly lower and phase calculation indicated no phase shift or a significantly reduced phase shift compared to controls. The same non-synchronization phenotype was observed in IR25-/Df(IR25a) flies, and temperature synchronization was fully restored in IR25^-/- rescue flies IR25a^-/- mutants synchronize to light and have normal free-running and temperature compensated periods. These results suggest that IR25a enables the circadian clock to sense subtle temperature changes across the entire physiological range, rather than mediating synchronization to a specific range. Increasing the temperature cycle amplitude to 4°C consistently restored temperature entrainment in IR25a^-/- flies (Chen, 2015).

Temperature receptors located in fly antennae and arista are not required for temperature-synchronized behaviour. As expected, it was found that antennal IR25a function is not required for temperature entrainment. To reveal the importance of IR25a expression in ChO neurons, tissue-specific IR25a RNA interference (RNAi) was performed using validated transgenes. IR25a RNAi in all or subsets of ChO neurons resulted in a lack of entrainment. By contrast, IR25a RNAi in multidendritic, TRPA1-expressing or clock neurons did not impair temperature entrainment. These findings are consistent with the absence of IR25a expression in clock neurons and the brain and show that IR25a functions in ChO neurons for temperature entrainment to 25°C:27°C temperature cycles in LL (Chen, 2015).

To identify the neural substrates underlying the lack of behavioural synchronization, clock protein levels were quantified in wild-type, IR25a^-/-, and IR25a^-/- rescue flies exposed to a shallow temperature cycle in LL. Although TIM expression was robustly rhythmic and synchronized in all clock neuronal groups in controls, TIM was barely detectable in the Dorsal Neuron 1 (DN1) and DN2 of IR25a^-/- flies. Moreover, in the small and large ventral lateral neurons (s-LNv and l-LNv), TIM expression exhibited an additional peak during the warm phase. In the DN3, TIM declined earlier compared to controls and there was no effect on the dorsal lateral neurons (LNd). In temperature cycles and DD, TIM levels in DN1 were also blunted but oscillations in the DN2 and DN3 were similar to controls. In contrast to LL, TIM did not oscillate in any of the LN groups and was at constantly low levels, consistent with the behavioural results obtained under these conditions. The alterations of TIM expression are temperature specific, as normal oscillations were observed in LD cycles at 25°C. An increase of the temperature cycle amplitude to 4°C also restored normal TIM expression in IR25a^-/- flies, in agreement with the behavioural rescue. In summary, in low-amplitude temperature cycles, IR25a is required for normally synchronized TIM oscillations in DN1-3 and LNv in LL and in DN1 and LN clock neurons in DD (Chen, 2015).

Tests were performed to see if the clock neurons affected by the lack of IR25a are indeed involved in regulating behavioural synchronization to shallow temperature cycles by blocking synaptic transmission using tetanus-toxin (TNT). Indeed, TNT-expression in DN1 and DN2 blocked synchronization in LL, whereas in DD only DN1 blockage interfered with temperature entrainment. Consistent with the differential effect on TIM oscillations in LL and DD these results strongly suggest that IR25a is required for the synchronized output of the DN1 (LL and DD) and DN2 (LL) to control temperature-entrained behaviour (Chen, 2015).

Next, it was asked if ChO might directly sense temperature in an IR25a-dependent manner. Leg nerve activity was recorded in restrained preparations, and ChO units were identified in the compound signal. In both wild-type and IR25a^-/- flies, spontaneous leg movement changed as a function of temperature along with motor and sensory activity. Additionally, presumed ChO activity of wild-type flies also increased during periods without movement. This temperature-induced but movement-independent, ChO activity was absent in IR25^-/- flies, showing that temperature is sensed in the legs in an IR25a-dependent manner. To test if IR25a contributes directly to temperature-sensing, this channel was ectopically expressed in the physiologically well-characterized, IR25a-negative, l-LNv. As a positive control, the temperature-sensitive Drosophila TRPA1 channel was also expressed in the l-LNv. Isolated brains were exposed to a temperature ramp, and spike frequency of individual l-LNv was recorded. Control l-LNv did not show a significant temperature-dependent change in neural activity. As expected, the firing rate of TRPA1 expressing neurons drastically increased linearly with temperature, as did other cellular parameters. IR25a expression resulted in a linear and reversible temperature-dependent increase in action potential firing frequency, whereas other cellular parameters showed no difference. Increasing the temperature by only 2°-3°C also lead to a reversible increase in firing frequency in IR25a expressing l-LNv. By contrast, expression of the related, but olfactory-specific co-receptor IR8a (which is not required for temperature entrainment) did not confer temperature-sensitivity. These observations suggest that IR25a is at least part of a thermosensory receptor required for temperature entrainment (Chen, 2015).

These data indicating that IR25a contributes to temperature sensing within ChO extend the roles of IR's beyond chemoreception, reminiscent of the requirement for the 'gustatory receptor' Gr28b in warmth-avoidance (Ni, 2013). Although this study shows that IR25a-expressing leg neurons are capable of sensing temperature and mediating temperature entrainment, it is possible that this receptor has a similar role elsewhere in the peripheral nervous system. IR25a responds to small temperature changes and it is proposed that the fly continuously integrates temperature signals received from multiple ChO across the whole body for synchronization of the clock. This potential reliance on weakly responding temperature receptors might explain why the Drosophila circadian clock is insensitive to brief temperature pulses, which could help maintain synchronized clock function in natural conditions of rapid and large temperature fluctuations (Chen, 2015).

The ability to adjust the speed of locomotion is essential for survival. In limbed animals, the frequency of locomotion is modulated primarily by changing the duration of the stance phase. The underlying neural mechanisms of this selective modulation remain an open question. This study reports a neural circuit controlling a similarly selective adjustment of locomotion frequency in Drosophila larvae. Drosophila larvae crawl using peristaltic waves of muscle contractions. This study found that larvae adjust the frequency of locomotion mostly by varying the time between consecutive contraction waves, reminiscent of limbed locomotion. A specific set of muscles, the lateral transverse (LT) muscles, co-contract in all segments during this phase, the duration of which sets the duration of the interwave phase. Two types of GABAergic interneurons were identified in the LT neural network, premotor neuron A26f and its presynaptic partner A31c, which exhibit segmentally synchronized activity and control locomotor frequency by setting the amplitude and duration of LT muscle contractions. Altogether, these results reveal an inhibitory central circuit that sets the frequency of locomotion by controlling the duration of the period in between peristaltic waves. Further analysis of the descending inputs onto this circuit will help understand the higher control of this selective modulation (Liu, 2023).

Neural circuit function underlies cognition, sensation, and behavior. Proper circuit assembly depends on the identity of the neurons in the circuit (gene expression, morphology, synapse targeting, and biophysical properties). Neuronal identity is established by spatial and temporal patterning mechanisms, but little is known about how these mechanisms drive circuit formation in postmitotic neurons. Temporal patterning involves the sequential expression of transcription factors (TFs) in neural progenitors to diversify neuronal identity, in part through the initial expression of homeodomain TF combinations. This study addresses the role of the Drosophila temporal TF Hunchback and the homeodomain TF Bicoid in the assembly of the Pair1 (SEZ_DN1) descending neuron locomotor circuit, which promotes larval pausing and head casting. Both Hunchback and Bicoid are expressed in larval Pair1 neurons, Hunchback activates Bicoid in Pair1 (opposite of their embryonic relationship), and the loss of Hunchback function or Bicoid function from Pair1 leads to ectopic presynapse numbers in Pair1 axons and an increase in Pair1-induced pausing behavior. These phenotypes are highly specific, as the loss of Bicoid or Hunchback has no effect on Pair1 neurotransmitter identity, dendrite morphology, or axonal morphology. Importantly, the loss of Hunchback or Bicoid in Pair1 leads to the addition of new circuit partners that may underlie the exaggerated locomotor pausing behavior. These data are the first to show a role for Bicoid outside of embryonic patterning and the first to demonstrate a cell-autonomous role for Hunchback and Bicoid in interneuron synapse targeting and locomotor behavior (Lee, 2022).

Neural circuit formation underlies the generation of behavior, and aberrant neural circuit development has been associated with many neural disorders, such as autism and attention deficit hyperactivity disorder. It is widely accepted that circuit formation requires the assembly of precise interconnectivity between diverse neuron subtypes. Although the mechanisms for generating molecularly and morphologically distinct neurons are well studied, little is known about how these developmental mechanisms regulate 'higher-order' neuronal properties such as pre- and post-synapse numbers or circuit partner choice (Lee, 2022).

In Drosophila, neuronal identity is specified by the combination of spatial and temporal transcription factors (TFs) acting on neuronal stem cells (neuroblasts in Drosophila). Spatial patterning creates molecularly distinct neuroblasts, followed by each neuroblast sequentially expressing a series of temporal TFs: Hunchback > Kruppel Temporal TFs are known to specify axon and dendrite morphology and targeting as well as behavior. For example, in neuroblast 7-1, the best characterized lineage in the embryo, the zinc-finger temporal TF Hunchback promotes expression of the homeodomain TF even-skipped that is required for proper motor neuron morphology and connectivity; and the combination of Kruppel and Pdm temporal TFs promotes expression of the homeodomain TF Nkx6 (FlyBase: HGTX) that is required for proper ventral projecting motor neuron morphology and connectivity. In both cases, transient temporal TF expression activates a homeodomain TF that persists in the postmitotic neuron to determine neuron morphology and neuromuscular connectivity. Similarly, work from the Hobert lab in C. elegans supports a model in which each of the 302 neurons is specified by a unique combination of homeodomain TFs. Overall, from worms to flies to mammals, temporal TFs activate homeodomain TFs to specify molecular and morphological neuronal identity (Lee, 2022).

Although homeodomain TFs are well known to specify these early aspects of motor neuron identity, their role in specifying later aspects of neuronal identity such as synapse number, position, and connectivity remains poorly understood. To address this question, the Pair1 (SEZ_DN1) locomotor circuit in Drosophila was used. Pair1 is a GABAergic interneuron with ipsilateral dendrites and contralateral descending axonal projections. The moonwalker descending neurons (MDN) provide inputs to Pair1, and Pair1 sends outputs to A27h neurons in the ventral nerve cord (VNC). When optogenetically activated, the Pair1 neurons induce a pause in forward locomotion and increase in head casting, in part by inhibiting the A27h neurons, which drive forward locomotion. Importantly, it was previously reported that the temporal TF Hunchback and the homeodomain TF Bicoid are expressed in Pair1 neurons throughout life, providing candidates to study the transcriptional regulation of Pair1 neuronal identity and connectivity (Lee, 2022).

Hunchback is the first temporal TF to be expressed in the Drosophila embryo and acts transiently to generate early born neurons. In the embryonic CNS, Hunchback is not required to maintain neuronal identity, although it is required to maintain proper dendrite morphology of the mAL interneuron in adult males. Bicoid is a homeodomain TF; however, its expression and function outside the early embryo had not been reported until recent work from this lab. Bicoid is well known to form an anterior-posterior morphogen gradient that directly activates hunchback to properly pattern the anterior-posterior body axis.26 Although the role of Hunchback in temporal patterning is conserved in mammals, Bicoid is found only in higher dipteran insects, making it an interesting contributor to insect evolution. This study tested the model that the temporal TF Hunchback activates the homeodomain TF Bicoid (opposite of their early embryo relationship) and whether Hunchback and Bicoid play a role in Pair1 neurotransmitter expression, neuron morphology, synapse number, circuit function, and behavior. The data support the emerging model that temporal TFs drive expression of homeodomain TFs that maintain distinct aspects of neuronal identity including synapse number/position, connectivity, and behavior (Lee, 2022).

The results show that Hunchback activates Bicoid in postmitotic Pair1 neurons, where it regulates specific and important aspects of neuronal identity-synapse number, synapse density, and connectivity. When Hunchback or Bicoid levels are decreased, synapse density is increased, with a corresponding disruption of the function of the Pair1 locomotor neural circuit. This work demonstrates a novel role for Hunchback and Bicoid-functioning postmitotically to regulate synapse number and to ensure proper circuit function. Importantly, this work also reproduces a phenotype previously seen in C. elegans-a single homeobox gene (unc-4) specifically regulates synaptic connectivity but not other aspects of neuronal identity. Interestingly, unc-4 expression is also regulated by a nonhomeodomain TF, suggesting that this regulatory pathway may be conserved between species to specify highly specific aspects of neuronal identity (Lee, 2022).

Unlike most early born neurons in the VNC that only transiently express Hunchback, and Bicoid which is only expressed in the first few hours of embryogenesis, the Pair1 neuron maintains both Hunchback and Bicoid expression into the adult. This suggests that a Pair1-specific regulatory mechanism may be leading to the persistent Hunchback and Bicoid expression and function. Given that the Pair1 neuron persists into adulthood, still expresses Hunchback and functions within a similar locomotor neural circuit, it is hypothesized that Hunchback and Bicoid expressions may be required in Pair1 neurons throughout life for the maintenance of the Pair1 locomotor neural circuit (Lee, 2022).

Surprisingly, Bicoid protein expression in larval Pair1 neurons was often detected in one or more spherical puncta located in the cytoplasm; this was observed with two independent Bicoid antibodies and a third FLAG-tagged Bicoid protein and was abolished by Bicoid RNAi. Given that Bicoid contains highly disordered regions with an abundance of glutamine and glycine, the spherical puncta may represent a phase-separation condensate, perhaps to keep nuclear Bicoid levels low. Interesting, Bicoid does not form spherical puncta outside of the larvae. Further investigation is needed to understand nature of the Bicoid cytoplasmic puncta, but these studies have the potential to elucidate a novel role for phase-separation in mature neurons (Lee, 2022).

Previous work showed that Bicoid activates hunchback in the early embryo. This study is the first to demonstrate the reverse that Hunchback can promotes Bicoid expression in vivo. Hunchback may regulate Bicoid directly or indirectly; supporting the former possibility are the findings that Hunchback protein binds two distinct regions at the 3' and 5' end of the bicoid locus. Alternatively, Hunchback may act indirectly by promoting Bicoid phase separation in larval neurons. Regardless, this finding supports the initial hypothesis that temporal TFs, like Hunchback, can activate homeodomain TFs, like Bicoid, to specify some or all aspects of neuronal identity. Other morphogens have been previously associated with establishing properties of neuronal identity, further suggesting that early developmental TFs may be important regulators of neuronal identity, connectivity, and circuit function in general (Lee, 2022).

Hunchback and Bicoid had no detectable role in regulating dendrite morphology, axon morphology, nor GABA expression, key aspects of Pair1 neuronal identity. However, both Hunchback and Bicoid are required for maintaining synapse number and functional connectivity of the Pair1 neuron. Trans-Tango experiments show that reduced Hunchback levels resulted in the addition of new synaptic partners of Pair1, although it cannot be excluded that these may be normal partners that are too weak to see in controls. Although the novel neuronal partners were not formally identified, the Drosophila larvae TEM volume was used to speculate that Pair1 could be synapsing with the A27h neurons located in the thoracic region. Given that A27h neurons are involved in forward locomotion, additional thoracic A27h neurons synapsing onto, and therefore being inhibited by Pair1 activation, could explain the increased pausing phenotype observed when Hunchback in knocked down in Pair1. Alternatively, abdominal A27h neurons could be forming more synapses with Pair1 in the posterior axonal regions (Lee, 2022).

Interestingly, it appears that Bicoid is not the only homeodomain TF functioning downstream of Hunchback in Pair1. When Hunchback is knocked down in Pair1, pausing speed is increased, head casting is increased, and recovery speeds are decreased. However, Bicoid knockdown only replicated the decreased recovery speed phenotype; this suggests that another homeodomain TF may be functioning downstream of Hunchback to regulate pausing speed and head casting. The data presented in this study begin to support this hypothesis, but additional work is needed to identify other homeodomain TFs functioning downstream of Hunchback (Lee, 2022).

This work is the first to demonstrate a role for Hunchback and Bicoid in postmitotic neurons to regulate synapse number, connectivity, and circuit function. These results raise the question of which is the more ancestral function of these two TFs: in segmentation, temporal patterning in neuroblasts, or postmitotic neuronal circuit maintenance (Lee, 2022)?

Deciphering how the brain regulates motor circuits to control complex behaviors is an important, long-standing challenge in neuroscience. In the fly, Drosophila melanogaster, this is coordinated by a population of ~ 1100 descending neurons (DNs). Activating only a few DNs is known to be sufficient to drive complex behaviors like walking and grooming. However, what additional role the larger population of DNs plays during natural behaviors remains largely unknown. For example, they may modulate core behavioral commands or comprise parallel pathways that are engaged depending on sensory context. This study evaluated these possibilities by recording populations of nearly 100 DNs in individual tethered flies while they generated limb-dependent behaviors, including walking and grooming. The largest fraction of recorded DNs encode walking while fewer are active during head grooming and resting. A large fraction of walk-encoding DNs encode turning and far fewer weakly encode speed. Although odor context does not determine which behavior-encoding DNs are recruited, a few DNs encode odors rather than behaviors. Lastly, this study illustrated how one can identify individual neurons from DN population recordings by using their spatial, functional, and morphological properties. These results set the stage for a comprehensive, population-level understanding of how the brain's descending signals regulate complex motor actions (Aymanns, 2022).

How circuits self-assemble starting from neuronal stem cells is a fundamental question in developmental neurobiology. This study addressed how neurons from different stem cell lineages wire with each other to form a specific circuit motif. In Drosophila larvae, developmental genetics (Twin spot MARCM, Multi-color Flip Out, permanent labeling) was combined with circuit analysis (calcium imaging, connectomics, network science). For many lineages, neuronal progeny are organized into subunits called temporal cohorts. Temporal cohorts are subsets of neurons born within a tight time window that have shared circuit level function. Sharp transitions in patterns of input connectivity were found at temporal cohort boundaries. In addition, a feed-forward circuit was identified that encodes the onset of vibration stimuli. This feed-forward circuit is assembled by preferential connectivity between temporal cohorts from different lineages. Connectivity does not follow the often-cited early-to-early, late-to-late model. Instead, the circuit is formed by sequential addition of temporal cohorts from different lineages, with circuit output neurons born before circuit input neurons. Further, this study generated new tools for the fly community. These data raise the possibility that sequential addition of neurons (with outputs oldest and inputs youngest) could be one fundamental strategy for assembling feed-forward circuits (Wang, 2022).

This study addressed two questions about the stem cell-based assembly of neuronal circuits. First, what is the relationship between neuronal birth order within a lineage and patterns of synaptic connectivity at the single-neuron level? Second, how do neurons from different lineages wire with each other? This study characterized the birth order, morphology, and input connectivity of all neurons in the NB3-3A1L/R lineage at single-neuron and single-synapse resolution. A feed-forward circuit was identified that encodes the onset of vibrational stimuli. And, for a majority of nerve cord interneurons within this circuit, their stem cell parent, birth order within their lineage, and birth timing relative to each other were identified. Together, this identifies four temporal cohorts, all of which have sharp connectivity boundaries. For most, but not all, there is inter-segmental connectivity between segmentally homologous temporal cohorts (e.g., early-born ELs in other segments connect to early-born ELs in A1). Further, neurons of different temporal cohorts from different lineages assemble sequentially, with circuit output neurons born before circuit input neurons (Wang, 2022).

The Drosophila larval connectome is a resource that can be used to understand circuit assembly. However, because the connectome is an anatomical dataset, a major challenge is to develop approaches that connect anatomy to development. In this article, several approaches were developed. For example, ts-MARCM was optimized for use in Drosophila embryos. ts-MARCM is the gold standard for determining birth order in Drosophila. However, for technical reasons, ts-MARCM has been used only in adults. This study discovered ts-MARCM clones can be robustly generated in early stage larvae with the addition of an amplifying and immortalizing gene cassette. In addition, several recently developed methods for inferring developmental origin were independently validated based on anatomical features in the Drosophila larval connectome. Specifically, the validations include use of neurite bundles as a proxy for lineage-relatedness and use of cortex neurite length as a rough proxy for birth order within a lineage. Finally, network science methods (distance analysis) were adapted to characterize the patterns of connectivity in connectome data. These approaches should be useful for Drosophila neurobiologists and beyond (Wang, 2022).

This study also developed NB3-3A1L/R as the first entire lineage for which birth order, morphology, and input connectivity is known at single neuron and synapse precision. One reason this is important is because NB3-3 has been extensively studied in embryos and much is known about molecular marker expression of NB3-3 progeny at the single-neuron level. Because the current dataset achieves cellular resolution, good guesses about the embryonic molecular-larval morphological pairings can be made using single-neuron birth order as a point of cross-reference. Such integrated data generates detailed and testable predictions. For example, the data predict that Castor expression in the late-born ELs promotes projection neuron morphology. Additionally, this study generated a new NB3-3-GAL4 line that can be used to manipulate gene expression in NB3-3. Thus, NB3-3A1L/R is a model lineage in which transcription factor expression in neuronal stem cells can be linked to circuit assembly and tested for function, and this study has generated tools that will enable hypothesis testing in the future (Wang, 2022).

A circuit motif is a pattern of synaptic connections between a set of specific neuron types that can be found across brain areas and across species. Circuit motifs have been suggested to represent the physical substrates of 'computational primitives'. There a are small number of fundamental, recurring circuit motifs (e.g., feed-forward, feedback, lateral inhibition, etc.). And so, a new conceptual approach in this article is to ask how are specific circuit motifs assembled during development. This study identified a new feed-forward circuit motif. Generally, feed-forward motifs are characterized as a pattern of connectivity in which one neuron (or neuron type) provides both direct and indirect input onto a second neuron (or neuron type). Feed-forward circuit motifs are common, found in many animals (e.g., nematodes, insects, mouse) and in many brain regions (e.g., somatosensory systems, olfactory systems, neocortex). Thus, feed-forward motifs are fundamental to neural signal processing. Feed-forward motifs can be further subdivided into feed-forward inhibitory or feed-forward excitatory circuit motifs, depending on the transmitter types of the neurons involved. Notably, anatomically the pattern of synaptic connectivity between neurons is the same in either motif subtype, and so this study does not distinguish between the two (Wang, 2022).

This study found that, in general, early-born ELs get direct excitatory synapses from CHOs and indirect (excitatory or inhibitory) input from chordotonals via Basins and Ladder interneurons (see Early-born ELs are embedded in a feed-forward motif and encode the onset of vibrational stimuli). Notably, there are differences in connectivity patterns between early-born ELs. For example, although all early-born ELs get direct input from chordotonals and Ladders, A08j1-3s get left-right symmetrical inputs, whereas A08x and A08m get asymmetrical inputs. This could correspond to functional differences. For example, A08x and A08m may be involved in left-right asymmetrical signal detection. Teasing apart the diversity of computations performed by each individual EL interneurons is the domain of future studies (Wang, 2022).

Because this study characterized the development of a feed-forward circuit in Drosophila, information is available that allows searching for similar circuits in other insect species. For example, in Drosophila, ELs express the transcription factor, Even-skipped (Eve). In Locust and other insects, lateral interneurons also express Eve. In Drosophila, MNB generates H-shaped, inhibitory interneurons (Ladders), which get input from sound/vibration sensitive, CHOs. In Locust, MNB generates H-shaped, inhibitory interneurons, which encode sound stimuli. Further, in both Drosophila and Locusts, MNB interneuron progeny express the transcription factor Engrailed. Thus, the neuronal components of the Ladder to early-born EL circuit are conserved, which raises the possibility of circuit-level conservation (Wang, 2022).

A major unanswered question in developmental neuroscience is how the mechanisms that generate neuronal diversity contribute to the formation of functional neuronal circuits. Part of the answer lies in the observation that temporal cohorts are subunits of lineages, linked both to larval circuit anatomy/function as well as to the embryonic gene expression programs that generate neuronal diversity. Thus far, temporal cohorts had been looked for and found in 8 out of 30 lineages in the nerve cord. This study identified three additional temporal cohorts-Basins, Ladders, and Notch OFF NB7-1 interneurons-in three lineages-NB3-5, MNB, and NB7-1, bringing the number to 11. This underscores the idea that temporal cohorts are common (Wang, 2022).

One open question about temporal cohorts was to what extent are temporal cohort borders associated with sharp changes in connectivity. Previous studies had identified temporal cohorts and linked them with function and connectivity. However, these studies lacked the resolution to distinguish between a 'graded' or 'sharp' wiring transition models. This study identified four temporal cohorts (early-born ELs, late-born ELs, Basins, Ladders) in three lineages (NB3-3, NB3-5, MNB), all of which have sharp changes in connectivity correlated with temporal cohort borders. For the Basin temporal cohort, it is noted that Basins have similar connectivity patterns with two additional NB3-5 progeny, 'Down and Back' and 'Crescent' neurons. Crescent is adjacent to Basins in birth order, whereas Down and Back is born much earlier. From this, two things were learned: (1) there can be significantly similar connectivity between neurons in one hemilineage that have nonadjacent birth times. (2) There can be significantly similar connectivity between neurons with adjacent birth times, but of different morphological classes. Another example of morphological variants with similar connectivity are late-born ELs, which can be either local or projection neurons. This underscores the idea that temporal cohorts are subunits of hemilineages defined by birth within a tight time window, rather than defined by similar neuronal morphology per se. Further, it is noted that within a temporal cohort, sequentially born neurons are often, but not always, the most similar in terms of connectivity. For example, within the early-born EL temporal cohort, the fifth-born neuron (A08m) is more similar to the first-born neuron (A08x) than its temporal neighbor (A08j2), the fourth-born EL. Thus, the data, combined with previous studies, show how temporal cohorts are developmental units related to circuits both at the functional and anatomical levels (Wang, 2022).

A second open question about temporal cohorts was the extent to which they are copies of each other. One reason this is an interesting question relates to circuit evolution. For the evolution of gene function, a popular model is a 'duplicate and diverge' model. Similarly, temporal cohorts could be duplications within a lineage whose function could then diverge, thereby driving circuit evolution. Such an idea motivated asking the question to what extent are temporal cohort copies. For example, do early-born ELs process chordotonal stimuli in a manner identical to how late-born ELs process proprioceptive stimuli? These data suggest a more complex picture. Early-born and late-born ELs differ in their connectivity patterns-including left-right and following pair similarities and inter-and intra-segmental connectivity patterns, suggesting that early-born and late-born ELs are likely to process information differently. Independent of the evolutionary implications, the data reveal previously unknown diversity in the structure of connectivity between neurons within adjacent temporal cohorts, which may indicate a diversity in underlying circuit assembly mechanisms (Wang, 2022).

A third open question is what sets up the borders of temporal cohorts. Previous work used motor neuron temporal cohorts of NB7-1 and NB3-1 as a model to address this question. In these stem cells, mis-expression of temporal transcription factors, Hunchback, Pdm, and Castor, modulates the number of motor neurons in a temporal cohort, without changing the size of the lineage. Thus, temporal transcription factors are able to regulate motor neuron temporal cohort borders. But it remains unclear how temporal transcription factors do so. Two, not mutually exclusive possibilities are that they act as transcriptional co-factors to induce differential gene expression programs and/or that they act as pioneer factors to alter the chromatin landscape. For NB3-3, it is noted that during the 7th to 11th divisions, which generate late-born ELs, NB3-3 expresses Grainyhead, which raises the possibility that Grainyhead may define the late-born EL cohort. Testing this idea is an important future direction (Wang, 2022).

It has been hypothesized that nerve cord circuits are assembled by preferential connectivity between distinct temporal cohorts. The current data provide experimental support for this hypothesis. Specifically, this study found that interneurons from three temporal cohorts wire together to form a feed-forward circuit-early-born ELs from NB3-3, mid-to-late-born Basins from NB3-5, and late-born Ladders from MNB. Although a vast majority of both the total synaptic input and the strongly connected individual neurons come from these lineages, there are also neurons from other lineages that synapse on early-born ELs (Wang, 2022).

Notably, one other study provided limited supported for the hypothesis that Drosophila nerve cord circuits are assembled by preferential connectivity between distinct temporal cohorts. This study focused on the Jaam-to-late-born EL-to-Saaghi circuit . Jaams are later-born interneurons in the NB5-2, Notch OFF hemilineage, and Saaghis are later-born interneurons in the NB5-2 Notch ON hemilineage. These data raised several possibilities: (1) there could be global alignment between lineages (e.g., all NB3-3 neurons get synapses from NB5-2 neurons). (2) Notch ON/OFF pairs of neurons might be pre-/postsynaptic partners of neurons within a temporal cohort. (3) Birth order-matched temporal cohorts from different lineages might selectively wire together (e.g., early-to-early and late-to-late connectivity). However, the current data demonstrate that none of these possibilities are generally true. Instead, a diversity was found in the manner in which temporal cohorts associate. The one consistent theme is that a limited number of temporal cohorts highly interconnect (Wang, 2022).

For most of this discussion, neurons were labelled as 'early-born' and 'late-born.' These labels refer to the birth order of neurons within a lineage. However, these labels do not refer to the absolute time at which a neuron is born. This is because in the Drosophila nerve cord neuroblasts are generated over a large span of embryogenesis. Thus, early-born neurons from one lineage can be generated at the same time as later-born neurons from a different lineage. This study is unique in that it determined both neuronal birth order and birth time (Wang, 2022).

One unexpected findings is that circuit outputs in one lineage are most often born before circuit inputs from other lineages. Broadly speaking, nerve cord and spinal cord contain many local circuits, which output to the brain (e.g., circuits processing somatosensory stimuli) or which output to muscles (e.g., circuits that generate motor patterns). Here, early-born ELs are the outputs of a somatosensory processing circuit and transmit information to the brain. Early-born ELs are born before their local, nerve cord inputs-Ladders, Basins, NB7-1 Notch OFF interneurons. However, from a first principles perspective, the opposite would be expected. This is because 22 of 30 neuroblasts in the nerve cord are born before NB3-3, and many of them divide multiple times to produce neurons before NB3-3 begins to generate early-born neurons. Therefore, there should be more neurons born before early-born ELs compared to those born after early-born ELs. And so, by chance alone one would expect early-born ELs to get more input from neurons born earlier or at the same time (Wang, 2022).

What are the hypotheses about the importance of birth order of output versus input neurons? In Drosophila, birth order is linked to two things: (1) lineage-intrinsic factors such as dynamically changing programs of gene expression in the stem cell, and (2) lineage-extrinsic factors or the dynamic environmental context into which neurons are born. Intrinsic factors, or extrinsic factors, or both may be playing a role in the assembly of this feed-forward circuit. A potential intrinsic mechanism is that of a temporal transcription factor 'matching code,' in which early-born ELs, Ladders, and Basins would all be derived from the same temporal transcription factor window. For NB3-5 and MNB, temporal transcription factor expression is only partially characterized. But tantalizingly both early-born ELs and a subset of Ladders are born during a period in which their respective neuroblasts express the temporal transcription factor, Castor. A potential extrinsic mechanism is that early-born EL dendrites may provide some type of signal that promotes later born neurons to synapse. There is evidence for such communication among Drosophila nerve cord neurons. Finally, it will be interesting to understand if sequential assembly is an absolute requirement, or if instead it facilitates rapid, efficient, or robust circuit assembly (Wang, 2022).

Studies of circuit assembly are still in their infancy. It is known that lineage-circuit relationships differ depending on circuit anatomy. But the converse is not known. That is, do all circuits of the same anatomy have a common lineage-circuit relationship? In the case of this study, focus was place on a single feed-forward circuit in the Drosophila larval nerve cord. And the specific question it raises is: To what extent is sequential addition of temporal cohorts from different lineages the only mechanism used to assembly all feed-forward motifs? Currently, the current answers are only partial and speculative. It is noted that there are so many connections made among neurons, it is unlikely that just one simple phenomenon that can explain the full complexity. The goals of the current research must be to identify rules and the circumstances where those rules apply. For example, these data rule out a 'strict' early-to-early, late-to-late model, meaning that this model alone cannot explain the wiring observe in this circuit. And yet, an early-to-early model does apply to inter-segmental wiring among temporal cohorts of the same lineage. Further, this early-to-early wiring phenomenon occurs alongside a sequential addition phenomenon. It speculate there could be additional phenomena underlying assembly of this simple motif. For example, beyond local interneurons, the birth date of sensory neurons was not determined, which provide the initial input to the circuit, nor did this study investigate the developmental origins of central brain neurons, which receive the output from the circuit. It is noted there is some evidence to support the generality that for assembly of feed-forward circuit motifs, presynaptic interneurons are born after their postsynaptic partners. Specifically, motor neurons are always circuit outputs (to muscle), and, in general, they are among the first neurons to be born during neurogenesis. Moreover, this pattern holds true in both Drosophila nerve cord and spinal cord. Therefore, the current data raise the possibility that one of many fundamental rules for circuit assembly is that feed-forward circuits are assembled sequentially from circuit output to circuit input (Wang, 2022).

Neuromodulators, such as neuropeptides, can regulate and reconfigure neural circuits to alter their output, affecting in this way animal physiology and behavior. This study presents a quantitative framework to study the relationships between the temporal pattern of activity of peptidergic neurons and of motoneurons during Drosophila ecdysis behavior, a highly stereotyped motor sequence that is critical for insect growth. This study analyzed, in the time and frequency domains, simultaneous intracellular calcium recordings of peptidergic CCAP (crustacean cardioactive peptide) neurons and motoneurons obtained from isolated central nervous systems throughout fictive ecdysis behavior induced ex vivo by Ecdysis triggering hormone. The activity of both neuronal populations was found to be tightly coupled in a cross-frequency manner, suggesting that CCAP neurons modulate the frequency of motoneuron firing. To explore this idea further, a probabilistic logistic model was used to show that calcium dynamics in CCAP neurons can predict the oscillation of motoneurons, both in a simple model and in a conductance-based model capable of simulating many features of the observed neural dynamics. Finally, an algorithm was developed to quantify the motor behavior observed in videos of pupal ecdysis, and their features were compared to the patterns of neuronal calcium activity recorded ex vivo. The motor activity of the intact animal was found to be more regular than the motoneuronal activity recorded from ex vivo preparations during fictive ecdysis behavior; the analysis of the patterns of movement also allowed to identification of a new post-ecdysis phase (Pineiro, 2021).

The mechanisms specifying neuronal diversity are well-characterized, yet it remains unclear how or if these mechanisms regulate neural circuit assembly. To address this, the developmental origin was mapped of 160 interneurons from seven bilateral neural progenitors (neuroblasts), and they were identified in a synapse-scale TEM reconstruction of the Drosophila larval CNS. Lineages were found to concurrently build the sensory and motor neuropils by generating sensory and motor hemilineages in a Notch-dependent manner. Neurons in a hemilineage share common synaptic targeting within the neuropil, which is further refined based on neuronal temporal identity. Connectome analysis shows that hemilineage-temporal cohorts share common connectivity. Finally, this study showed that proximity alone cannot explain the observed connectivity structure, suggesting hemilineage/temporal identity confers an added layer of specificity. Thus, this study demonstrated that the mechanisms specifying neuronal diversity also govern circuit formation and function, and that these principles are broadly applicable throughout the nervous system (Mark, 2021).

Tremendous progress has been made in understanding the molecular mechanisms generating neuronal diversity in both vertebrate and invertebrate model systems. In mammals, spatial cues generate distinct pools of progenitors, which generate neuronal diversity in each spatial domain. The same process occurs in invertebrates like Drosophila, but with a smaller number of cells, and this process is particularly well understood. The first step occurs when spatial patterning genes act combinatorially to establish single, unique progenitor (neuroblast) identities. These patterning genes endow each neuroblast with a unique spatial identity (Mark, 2021).

The second step is temporal patterning -- the specification of neuronal identity based on birth-order, an evolutionarily conserved mechanism for generating neuronal diversity. This study focused on Drosophila embryonic neuroblasts, which undergo a cascade of temporal transcription factors: Hunchback (Hb), Krüppel (Kr), Pdm, and Castor (Cas). Each temporal transcription factor is inherited by ganglion mother cells (GMCs) born during each expression window. The combination of spatial and temporal factors endows each GMC with a unique identity (Mark, 2021).

The third step is hemilineage specification, which was initially characterized in Drosophila larval and adult neurogenesis, and may also be used in vertebrate neurogenesis. Hemilineages are formed by GMC asymmetric division into a pair of post-mitotic neurons; during this division, the Notch inhibitor Numb (Nb) is partitioned into one neuron (NotchOFF neuron), whereas the other sibling neuron receives active Notch signaling (NotchON neuron), thereby establishing NotchON and NotchOFF hemilineages. In summary, three mechanisms generate neuronal diversity within the embryonic central nervous system (CNS): neuroblast spatial identity, GMC temporal identity, and neuronal hemilineage identity (Mark, 2021).

A great deal of progress has also been made in understanding neural circuit formation in both vertebrates and invertebrate model systems, revealing a multi-step mechanism. Neurons initially target their axons to broad regions (e.g., thalamus/cortex), followed by targeting to a neuropil domain (glomeruli/layer), and finally forming highly specific synapses within the targeted domain (Mark, 2021).

Despite the progress in understanding the generation of neuronal diversity and the mechanisms governing axon guidance and neuropil targeting, how these two developmental processes are coordinated remains largely unknown. While it is accepted that the identity of a neuron is linked to its connectivity, the developmental mechanisms involved are unclear. For example, do clonally related neurons target similar regions of the neuropil due to the expression of similar guidance cues? Do temporal cohorts born at similar times show preferential connectivity? This study addressed the question of whether any of the three developmental mechanisms (spatial, temporal, hemilineage identity) are correlated with any of the three circuit-wiring mechanisms (neurite targeting, synapse localization, connectivity). This study mapped the developmental origin for 80 bilateral pairs of interneurons in abdominal segment 1 (A1) by identifying and reconstructing these neurons within a full CNS TEM volume -- this is over a quarter of the ~300 neurons per hemisegment. The unexpected observation was made that hemilineage identity determines neuronal projection to sensory or motor neuropils; thus, neuroblast lineages coordinately produce sensory and motor circuitry. In addition, it was shown that neurons with shared hemilineage-temporal identity target pre- and post-synapse localization to similar positions in the neuropil, and that hemilineage-temporal cohorts share more common synaptic partners than that produced by neuropil proximity alone. Thus, temporal and hemilineage identity plays essential roles in establishing neuronal connectivity (Mark, 2021).

This study determined the relationship between developmental mechanisms (spatial, temporal, and hemilineage identity) and circuit assembly mechanisms (projections, synapse localization, and connectivity). To do this, both developmental and circuit features were mapped for 160 neuronal progeny of 14 neuroblast lineages in a serial section TEM reconstruction - this allows characterization neurons that share a developmental feature at single synapse resolution. It is important to note that the seven neuroblasts in this study were chosen based on successful clone generation and availability of single neuroblast Gal4 lines, and thus there should be no bias towards a particular pattern of neurite projections, synapse localization, or connectivity. The results show that individual neuroblast lineages have unique but broad axon and dendrite projections to both motor and sensory neuropil; hemilineages restrict projections and synapse localization to either motor or sensory neuropil; and distinct temporal identities within hemilineages provide additional specificity in synapse localization and connectivity. Thus, all three developmental mechanisms act combinatorially to progressively refine neurite projections, synapse localization, and connectivity (Mark, 2021).

In mammals, clonally related neurons often have a similar location, morphology, and connectivity. In contrast, this study found that clonally related neurons project widely in the neuropil, to both sensory and motor domains, and thus lack shared morphology. Perhaps as brain size expands to contain an increasing number of progenitors, each clone takes on a more uniform structure and function. Yet the observation that each neuroblast clone had highly stereotyped projections suggests that neuroblast identity (determined by the spatial position of the neuroblast) determines neuroblast-specific projection patterns. Testing this functionally would require manipulating spatial patterning cues to duplicate a neuroblast and assay both duplicate lineages for similar projections and connectivity (Mark, 2021).

This study found that hemilineages produce sensory and motor processing units via a Notch-dependent mechanism. Pioneering work on Drosophila third instar larval neuroblast lineages showed that each neuroblast lineage is composed of two hemilineages with different projection patterns and neurotransmitter expression. These studies were extended to embryonic neuroblasts and showed that Notch signaling determines motor versus sensory neuropil projections in all lineages examined. Surprisingly, the NotchON hemilineage always projected to the dorsal/motor neuropil, whereas the NotchOFF hemilineage always projected to the ventral/sensory neuropil. The relationship between the NotchON hemilineage projecting to the motor neuropil may be a common feature of all 30 segmental neuroblasts or it could be that the NotchON/NotchOFF provides a switch to allow each hemilineage to respond differently to dorsoventral guidance cues, with some projecting dorsally and some projecting ventrally. Analysis of additional neuroblast lineages will resolve this question. Another point to consider is the potential role of Notch in post-mitotic neurons as these experiments generated Notchintra misexpression in both newborn sibling neurons as well as mature post-mitotic neurons. Future work manipulating Notch levels specifically in mature post-mitotic neurons undergoing process outgrowth will be needed to identify the role of Notch in mature neurons, if any (Mark, 2021).

Elegant work has identified neuropil gradients of Slit and Netrin along the mediolateral axis, Semaphorins along the dorsoventral axis, and Wnt5 along the anteroposterior axis. The finding that neurons in a hemilineage project to a common region of the neuropil strongly suggests that all neurons within a hemilineage respond in the same way to these global pathfinding cues. Conversely, the finding that neurons in different hemilineages target distinct regions of the neuropil suggests that each hemilineage expresses a different palette of guidance receptors, which enable them to respond differentially to the same global cues. For example, neurons in ventral hemilineages may express Plexin receptors to repel them from high Semaphorins in the dorsal neuropil (Mark, 2021).

Hemilineages have not been well described in vertebrate neurogenesis. Notch signaling within the Vsx1 + V2 progenitor lineage generates NotchOFF V2a excitatory interneurons and NotchON V2b inhibitory interneurons, which may be distinct hemilineages. Interestingly, both V2a and V2b putative hemilineages contain molecularly distinct subclasses; this study raises the possibility that these subtypes arise from temporal patterning within the V2 lineage. In addition, NotchON/NotchOFF hemilineages may exist in the pineal photoreceptor lineage, where NotchON and NotchOFF populations specify cell-type identity (Mark, 2021).

Only recently have the role of hemilineages been tested for their functional properties. In adults, activation of each larval hemilineage from NB5-2 showed similar behavioral output, whereas each hemilineage from NB6-1 elicited different behaviors. Previous work showed that the Eve+, Saaghi, and Jaam neurons are part of a proprioceptive circuit; this study shows that each class of neurons represents a hemilineage-temporal cohort. Note that the Jaam neurons process sensory input and are in a NotchOFF hemilineage, supporting the conclusion that NotchOFF hemilineages are devoted to sensory processing; the Saaghi premotor neurons are in a NotchON hemilineage consistent with their role in motor processing. Interestingly, both input and output neurons in this circuit arise from a common progenitor (NB5-2), which may generate late-born Jaam/Saaghi sibling neurons. In the future, it would be interesting to determine if other sibling hemilineages are in a common circuit to generate a specific behavior (Mark, 2021).

The hemilineage results have several implications. First, the results reveal that sensory and motor processing components of the neuropil are being built in parallel, with one half of every GMC division contributing to either sensory or motor networks. This would be an efficient mechanism to maintain sensory/motor balance as lineage lengths are modified over evolutionary time. Second, the results suggest that looking for molecular or morphological similarities in full neuroblast clones may be misleading due to the full neuroblast clone comprising two different hemilineages. For example, performing bulk RNAseq on all neurons in a neuroblast lineage is unlikely to reveal key regulators of pathfinding or synaptic connectivity due to the mixture of disparate neurons from the two hemilineages (Mark, 2021).

The cortex neurite length of neurons was used as a proxy for birth-order and shared temporal identity. This is thought to be a good approximation, but it clearly does not precisely identify neurons born during each of the Hb, Kr, Pdm, Cas temporal transcription factor windows. Nevertheless, there was sufficient resolution to observe that neurons with the same temporal identity clustered their pre- or postsynapses, rather than localizing them uniformly through the hemilineage neuropil domain. Interestingly, the three-dimensional location of each hemilineage temporal cohort synaptic cluster is identical on the left and right side of A1, ruling out the mechanism of stochastic self-avoidance. Other possible mechanisms include hemilineage-temporal cohorts expressing different levels of the presynapse spacing cue Sequoia or hemilineage-temporal cohorts exhibiting different responses to global patterning cues. Testing the function of temporal identity factors in synaptic tiling will require hemilineage-specific alteration of temporal identity, followed by assaying synapse localization within the neuropil (Mark, 2021).

The results strongly suggest that hemilineage identity and temporal identity act combinatorially to allow small pools of neurons to target pre- and postsynapses to highly precise regions of the neuropil, thereby restricting synaptic partner choice. Yet precise neuropil targeting is not sufficient to explain connectivity as many similarly positioned axons and dendrites fail to form connections. The model is favored that hemilineages direct gross neurite targeting to motor or sensory neuropil, whereas temporal identity acts combinatorially with each hemilineage to direct more precise neurite targeting and synaptic connectivity. Thus, the same temporal cue (e.g., Hb) could promote targeting of one pool of neurons in one hemilineage and another pool of neurons in an adjacent hemilineage. This limits the number of regulatory mechanisms needed to generate precise neuropil targeting and connectivity for all ~600 neurons in a segment of the larval CNS (Mark, 2021).

In conclusion, this study demonstrates how developmental information can be integrated with connectomic data. Lineage information, hemilineage identity, and temporal identity can all be accurately predicted using morphological features (e.g., number of fascicles entering the neuropil for neuroblast clones and radial position for temporal cohorts). This both greatly accelerates the ability to identify neurons in a large EM volume as well as sets up a framework in which to study development using data typically intended for studying connectivity and function. This framework is used to relate developmental mechanism to neuronal projections, synapse localization, and connectivity. Lineage, hemilineage, and temporal identity were found act sequentially to progressively refine neuronal projections, synapse localization, and connectivity, and the data supports a model where hemilineage-temporal cohorts are units of connectivity for assembling motor circuits (Mark, 2021).

To investigate circuit mechanisms underlying locomotor behavior, this study used serial-section electron microscopy (EM) to acquire a synapse-resolution dataset containing the ventral nerve cord (VNC) of an adult female Drosophila melanogaster. To generate this dataset, GridTape, a technology that combines automated serial-section collection with automated high-throughput transmission EM, was developed. Using this dataset, neuronal networks were studied that control leg and wing movements by reconstructing all 507 motor neurons that control the limbs. A specific class of leg sensory neurons was shown to synapse directly onto motor neurons with the largest-caliber axons on both sides of the body, representing a unique pathway for fast limb control. Open access is provided to the dataset and reconstructions registered to a standard atlas to permit matching of cells between EM and light microscopy data. GridTape instrumentation designs and software are provided to make large-scale EM more accessible and affordable to the scientific community (Phelps, 2021).

Large-scale neuronal wiring diagrams at synapse resolution will be a crucial element of future progress in neuroscience. This paper presents GridTape, a technology for accelerating large-scale electron microscopy (EM) data acquisition. The power of this approach is demonstrated by acquiring a dataset encompassing an adult female Drosophila ventral nerve cord (VNC). This dataset was used to identify a monosynaptic circuit that directly links a specialized proprioceptive cell type, the bilateral campaniform sensillum (bCS) neurons, with specific motor neurons (MNs). The results highlight how EM datasets can be used to characterize cell types and guide development of cell type-specific driver lines. The public release of this dataset provides a resource for studying the circuit connectivity underlying motor control and demonstrates the rapid advances that can be powered by the GridTape approach (Phelps, 2021).

Data acquisition remains a rate-limiting step in generating EM connectomics datasets. Manual sectioning for TEM is slow, imprecise, and unreliable. Meanwhile, SEM approaches that circumvent the need for manual sectioning have slow imaging speeds or require massive parallelization of expensive electron optics to acquire comparable datasets. GridTape builds on previous efforts toward TEM parallelization and automation, but overcomes the need for manual sectioning, allowing faster and more consistent section collection and imaging. Because imaging is nondestructive, GridTape is compatible with enhancement by post-section labeling and allows for re-imaging. By eliminating the need to separately handle thousands of fragile sections, GridTape reduces data loss and artifact frequency. This results in better alignment of sections into a coherent, high signal-to-noise image volume, leading to efficient and accurate reconstructions (Phelps, 2021).

GridTape is also less expensive than high-throughput SEM platforms. For the current price of one commercial multi-beam SEM system, ten TEMCA-GTs can be built, and samples collected on GridTape can be distributed across microscopes for simultaneous imaging. The fixed microscope hardware costs are accompanied by consumable costs associated with support film coating (∼USD$4 per slot, or ∼$18,000 for this study), but this cost is expected to decrease due to technological improvements and economies of scale (Phelps, 2021).

In the future, GridTape acquisition rates will increase as cameras and imaging sensors improve. Because TEM imaging is a widefield technique, imaging throughput can be increased by using larger camera arrays and brighter electron sources. Moreover, sections larger than current slot dimensions could be accommodated with wider tape and larger slots, although custom microscopes may be necessary for very large samples and slot size will depend on material properties of the support film (Phelps, 2021).

The EM dataset presented in this study provides a public resource for understanding how the Drosophila nervous system generates behavior. An adult Drosophila VNC was chosen because it is an ideal test case for generating and validating a connectomic dataset. The circuit is genetically and electrophysiologically accessible and neurons are identifiable across individuals. The VNC is compact, containing approximately a third of the neurons in the adult CNS, but contains neuronal networks for executing complex motor behaviors. Because the brain controls behavior via descending projections to the VNC, it is critical to be able to study neuronal circuits in both the brain and the VNC at synaptic resolution. Notably, this VNC dataset complements the recent release of an EM dataset comprising the complete adult female Drosophila brain (Phelps, 2021).

The VNC dataset was validated by automatically mapping its synapses with high accuracy, successfully registering the predicted synapse density map to a standard atlas and finding a high degree of similarity between EM and LM reconstructed neurons. A pipeline is demonstrated for identifying cells of interest in the dataset by comparing EM reconstructions to LM data. Finally, as a foundation for future work, >1,000 neuron reconstructions and their connectivity are made publicly available. Although these reconstructions were generated manually, advances in automated segmentation approaches are dramatically accelerating analysis of serial-section TEM data (Phelps, 2021).

Flexible motor control relies heavily on feedback from proprioceptors, a class of sensory neurons that measure body position, velocity, and load. In both vertebrates and invertebrates, proprioceptive feedback is processed by the central nervous system to tune motor output. In insects, morphologically distinct subclasses of chordotonal neurons encode different features of leg movement such as position, velocity, and vibration. Campaniform sensilla encode load signals similar to mammalian Golgi tendon organs. Althoughthe main proprioceptor types are known and the signals they encode, it is now an opportune time to understand how motor circuits integrate proprioceptive inputs to control the body by mapping the complete wiring diagram of an adult Drosophila VNC (Phelps, 2021).

EM datasets also enable the discovery of cell types and synaptic connections that may be overlooked by other methods. For instance, targeted reconstruction of sensory afferents revealed that the leg sensory neurons with the largest-caliber axons are the bCS neurons, which make direct synapses onto large-caliber leg MNs. This connection is monosynaptic and bCS inputs are specifically located near the putative MN spike initiation zone, suggesting that speed and reliability are essential for the function of these connections (Phelps, 2021).

The unique bilateral and intersegmental projections of bCS neurons suggests that they directly influence multiple limbs on both sides of the body. This leads to several hypotheses about their function. Prior work suggested that campaniform sensilla encode information about step timing that could drive the transition between stance and swing phases of walking. However, this study observed that bCS neurons synapse onto the same MNs on both sides of the bod, suggesting they drive symmetric movements of left and right legs. This makes it unlikely that bCS neurons contribute to walking, which involves antiphase movement of contralateral legs. Instead, bCS neurons may underlie a fast reflex where multiple legs flex in response to bCS activation. CS can signal either increases or decreases in load, depending on the sensillum's placement and orientation on the leg. Therefore, bCS neuron activation could forcefully stabilize posture in response to additional weight (e.g., to prevent the body from being crushed) or to grip a surface in response to a loss of load (e.g., to prevent being blown away by a gust of wind). The genetic tools that were created to target bCS neurons will enable future analyses of their function (Phelps, 2021).

Monosynaptic sensory-to-motor neuron connectivity is infrequent in larval Drosophila, but has been observed in other adult insects. Direct sensory feedback may be key in adults for precise control of their segmented limbs. The absence of such connections in larvae may indicate that controlling a limbless body relies less on sensory feedback and more on feedforward processing. As adult flies move much faster than larvae, another possibility is that fast monosynaptic sensory feedback is crucial for fast-moving animals. Indeed, research on escape responses has demonstrated that high-velocity movements are often controlled by the fastest neuronal pathways (Phelps, 2021).

MNs have diverse but stereotyped functions, reflecting the array of muscles and muscle fibers they innervate. Some MNs have unique and reproducible transcription factor signatures that underlie their physiological properties and axonal morphology. These unique transcription factor patterns specify morphologies that are fairly stereotyped across animals. The results extend these findings by quantitatively demonstrating that most dendritic arborizations of leg MNs are sufficiently stereotyped to be individually identifiable by structure alone. Because the complete population of MNs controlling the two front legs was reconstructed, it was possible to show that mirror symmetry in primary neurite number and position is a systematic principle of MN populations. In contrast, sensory neurons have more redundant copies and variable copy numbers (Phelps, 2021).

Previously, comprehensive neuronal connectivity maps were acquired for the nerve cords of other organisms including C. elegans, leeches, lampreys, and Drosophila larvae. These maps enabled a more complete understanding of how the nervous system controls locomotor rhythms underlying swimming and crawling. Less is known about the connectivity underlying motor control in limbed animals. The EM dataset presented in this study as a public resource will enable complete connectivity mapping for the circuits that control the legs and wings of an adult Drosophila. Combined with recent advances in recording activity from genetically identified VNC neurons during behavior, adult Drosophila is emerging as a powerful system for studying motor control. With these tools, it is expected that a deeper understanding of the circuit basis for complex motor control is within reach (Phelps, 2021).

Typical patterned movements in animals are achieved through combinations of contraction and delayed relaxation of groups of muscles. However, how intersegmentally coordinated patterns of muscular relaxation are regulated by the neural circuits remains poorly understood. This study identified Canon, a class of higher-order premotor interneurons, that regulates muscular relaxation during backward locomotion of Drosophila larvae. Canon neurons are cholinergic interneurons present in each abdominal neuromere and show wave-like activity during fictive backward locomotion. Optogenetic activation of Canon neurons induces relaxation of body wall muscles, whereas inhibition of these neurons disrupts timely muscle relaxation. Canon neurons provide excitatory outputs to inhibitory premotor interneurons. Canon neurons also connect with each other to form an intersegmental circuit and regulate their own wave-like activities. Thus, these results demonstrate how coordinated muscle relaxation can be realized by an intersegmental circuit that regulates its own patterned activity and sequentially terminates motor activities along the anterior-posterior axis (Hiramoto, 2021).

The Drosophila ventral nerve cord (VNC) is composed of thousands of neurons born from a set of individually identifiable stem cells. The VNC harbors neuronal circuits required to execute key behaviors, such as flying and walking. Leveraging the lineage-based functional organization of the VNC, this study investigated the developmental and molecular basis of behavior by focusing on lineage-specific functions of the homeodomain transcription factor, Unc-4. Unc-4 was found to function in lineage 11A to promote cholinergic neurotransmitter identity and suppress the GABA fate. In lineage 7B, Unc-4 promotes proper neuronal projections to the leg neuropil and a specific flight-related take-off behavior. It was also uncovered that Unc-4 acts peripherally to promote proprioceptive sensory organ development and the execution of specific leg-related behaviors. Through time-dependent conditional knock-out of Unc-4, it was found that its function is required during development, but not in the adult, to regulate the above events (Lacin, 2020).

How does a complex nervous system arise during development? Millions to billions of neurons, each one essentially unique, precisely interconnect to create a functional central nervous system (CNS) that drives animal behavior. Work over several decades shows that developmentally established layers of spatial and temporal organization underlie the genesis of a complex CNS. For example, during spinal cord development in vertebrates, different types of progenitor cells arise across the dorso-ventral axis and generate distinct neuronal lineages in a precise spatial and temporal order. The pMN progenitors are located in a narrow layer in the ventral spinal cord and generate all motor neurons. Similarly, twelve distinct pools of progenitors that arise in distinct dorso-ventral domains generate at least 22 distinct interneuronal lineages. Within each lineage, neurons appear to acquire similar identities: they express similar sets of transcription factors, use the same neurotransmitter, extend processes in a similar manner and participate in circuits executing a specific behavior (Lacin, 2020).

The adult Drosophila ventral nerve cord (VNC), like the vertebrate spinal cord, also manifests a lineage-based organization. The cellular complexity of the VNC arises from a set of segmentally repeated set of 30 paired and one unpaired neural stem cells (Neuroblasts [NBs]), which arise at stereotypic locations during early development. These individually identifiable NBs undergo two major phases of proliferation: the embryonic phase generates the functional neurons of the larval CNS, some of which are remodeled to function in the adult, and the post-embryonic phase generates most of the adult neurons. The division mode within NB lineages adds another layer to the lineage-based organization of the VNC. Each NB generates a secondary precursor cell, which divides via Notch-mediated asymmetric cell division to generate two neurons with distinct identities. After many rounds of such cell divisions, each NB ends up producing two distinct hemilineages of neurons, termed Notch-ON or the 'A' and Notch-OFF or the 'B' hemilineage. This paper focuses only on postembryonic hemilineages, which from this point on in the paper are refered to as hemilineages for simplicity. Within a hemilineage, neurons acquire similar fates based on transcription factor expression, neurotransmitter usage, and axonal projection. Moreover, neurons of each hemilineage appear dedicated for specific behaviors. For example, artificial neuronal activation of the glutamatergic hemilineage 2A neurons elicit specifically high frequency wing beating, while the same treatment of the cholinergic hemilineage 7B neurons leads to a specific take-off behavior. Thus, hemilineages represent the fundamental developmental and functional unit of the VNC (Lacin, 2020).

Previous work has mapped the embryonic origin, axonal projection pattern, transcription factor expression, and neurotransmitter usage of essentially all hemilineages in the adult Drosophila VNC (see Lacin, 2019; Shepherd, 2019). This study leveraged this information to elucidate how a specific transcription factor, Unc-4, acts within individual hemilineages during adult nervous system development to regulate neuronal connectivity and function, and animal behavior. Unc-4, an evolutionarily conserved transcriptional repressor, is expressed post-mitotically in seven of the 14 cholinergic hemilineages in the VNC: three 'A' -Notch-ON- hemilineages (11A, 12A, and 17A) and four 'B' -Notch-OFF- hemilineages (7B, 18B, 19B, and 23B). For four of the Unc-4+ hemilineages (7B, 17A, 18B, and 23B), the neurons of the sibling hemilineage undergo cell death. For the remaining three (11A, 12A, and 19B), the neurons of the sibling hemilineage are GABAergic (Lacin, 2019). Unc-4 expression in these hemilineages is restricted to postmitotic neurons and it appears to mark uniformly all neurons within a hemilineage during development and adult life (Lacin, 2014; Lacin, 2016; Lacin, 2020 and references therein).

This study generated a set of precise genetic tools that allowed uncovering of lineage-specific functions for Unc-4: in the 11A hemilineage, Unc-4 drives the cholinergic identity and suppresses the GABAergic fate; in the 7B hemilineage, Unc-4 promotes correct axonal projection patterns and the ability of flies to execute a stereotyped flight take-off behavior. It was also found that Unc-4 is expressed in the precursors of chordotonal sensory neurons and required for the development of these sensory organs, with functional data indicating Unc-4 functions in this lineage to promote climbing, walking, and grooming activities (Lacin, 2020).

Using precise genetic tools, this study dissected the function of the Unc-4 transcription factor in a lineage-specific manner. Within the PNS, Unc-4 function is needed for the proper development of the leg chordotonal organ and walking behavior; whereas in the CNS, Unc-4 dictates neurotransmitter usage within lineage 11A and regulates axonal projection and flight take-off behavior in lineage 7B. Below, are discussed three themes arising from this work: lineage-specific functions of individual transcription factors, an association of Unc4+ lineages with flight, and the lineage-based functional organization of the CNS in flies and vertebrates (Lacin, 2020).

Seven neuronal hemilineages express Unc-4 in the adult VNC, but the phenotypic studies revealed a function for Unc-4 in only two of them: in the 11A hemilineage, Unc-4 promotes the cholinergic fate and inhibits the GABAergic fate, while in the 7B hemilineage, Unc-4 ensures proper flight take-off behavior likely by promoting the proper projection patterns of the 7B interneurons into the leg neuropil. Why was no loss-of-function phenotype detected for Unc4 in most of the hemilineages in which it is expressed? A few reasons may explain this failure. First, the phenotypic analysis was limited: Neuronal projection patterns and neurotransmitter fate were detected, but not other molecular, cellular, or functional phenotypes. Unc-4 may function in other lineages to regulate other neuronal properties that were not assayed, such as neurotransmitter receptor expression, channel composition, synaptic partner choice, and/or neuronal activity. In addition, as this analysis assayed all cells within the lineage, it would have missed defects that occur in single cells or small groups of cells within the entire hemilineage. Second, Unc-4 may act redundantly with other transcription factors to regulate the differentiation of distinct sets of neurons. Genetic redundancy among transcription factors regulating neuronal differentiation is commonly observed in the fly VNC. Thus, while the research clearly identifies a role for Unc-4 in two hemilineages, it does not exclude Unc-4 regulating more subtle cellular and molecular phenotypes in the other hemilineages in which it is expressed. Similarly, pan-neuronal deletion of Unc-4 specifically in the adult did not lead to any apparent behavioral defect even though Unc-4 expression is maintained in all Unc-4+ lineages throughout adult life, suggesting that Unc-4 function is dispensable in mature neurons after eclosion under standard lab conditions. Future work will be required to ascertain whether Unc-4 functions during adult life or in more than two of its expressing hemilineages during development. Nonetheless, this work shows that Unc-4 executes distinct functions in the 7B and 11A lineages. The Hox transcription factors, Ubx, Dfd, Scr, and Antp, have also been shown to execute distinct functions in different lineages in the fly CNS, suggesting transcription factors may commonly drive distinct cellular outcomes in the context of different lineages. What underlies this ability of one transcription factor to regulate distinct cellular events in different neuronal lineages? The ancient nature of the lineage-specific mode of CNS development likely holds clues to this question. The CNS of all insects arises via the repeated divisions of a segmentally repeated array of neural stem cells whose number, ~30 pairs per hemisegment, has changed little over the course of insect evolution. Within this pattern, each stem cell possessing a unique identity based on its position and time of formation. Each stem cell lineage has then evolved independently of the others since at least the last common ancestor of insects, approximately 500 million years ago. Thus, if during evolution an individual transcription factor became expressed in multiple neuronal lineages after this time, it would not be surprising that it would execute distinct functions in different neuronal lineages. The lineage-specific evolution of the CNS development in flies, worms, and vertebrates may explain why neurons of different lineages that share specific properties, for example, neurotransmitter expression, may employ distinct transcriptional programs to promote this trait (Lacin, 2020).

Although Unc-4 appears to have distinct functions in different lineages, this study found that an association with flight is a unifying feature among most Unc4+ interneuron lineages and motor neurons. All Unc-4+ hemilineages in the adult VNC except the 23B hemilineage heavily innervate the dorsal neuropils of the VNC, which are responsible for flight motor control and wing/haltere related behaviors, including wing-leg coordination. For example, hemilineages 7B, 11A, and 18B regulate flight take-off behavior and 12A neurons control wing-based courtship singing. In addition, most Unc-4+ motor neurons are also involved with flight - these include MN1-5, which innervate the indirect flight muscles, as well as motor neurons that innervate the haltere and neck muscles, which provide flight stabilization. Since Unc-4 is conserved from worms to humans, it is likely that Ametabolous insects, like silverfish, which are primitively wingless, also have unc-4. It has yet to be determined, though, whether in such ametabolous insects the same hemilineages express Unc-4, and hence this pattern was in place prior to the evolution of flight. This would suggest that there was some underlying association amongst this set of hemilineages that may have been exploited in the evolution of flight. Alternatively, Unc-4 may be lacking in these hemilineages prior to the evolution of flight but then its expression may have been acquired by these hemilineages as they were co-opted into a unified set of wing-related behaviors (Lacin, 2020).

The adult fly VNC is composed of 34 segmentally repeated hemilineages, which are groups of lineally related neurons with similar features for example, axonal projection and neurotransmitter expression. These hemilineages also appear to function as modular units, each unit appears responsible for regulating particular behaviors, indicating the VNC is assembled via a lineage-based functional organization. The vertebrate spinal cord exhibits similar organization: lineally-related neurons acquire similar fates ('cardinal classes') and function in the same or parallel circuits. The similarity of the lineage-based organization in insect and vertebrate nerve/spinal cords raises the question whether they evolved from a common ground plan or are an example of convergent evolution. Molecular similarities in CNS development between flies and vertebrates support both CNS's arise from a common ground plan. For example, motor neuron identity in both flies and vertebrates, use the same set of transcription factors: Nkx6, Isl, and Lim3. Moreover, homologs of many transcription factors expressed in fly VNC interneurons, such as Eve and Lim1, also function in interneurons of the vertebrate spinal cord. Whether any functional/molecular homology is present between fly and vertebrate neuronal classes awaits comparative genome-wide transcriptome analysis and functional characterization of neuronal classes in the insect VNC and vertebrate spinal cord (Lacin, 2020).

This approach was validated using the A27h neuron, a well-characterised cholinergic premotor interneuron that the connectome indicates is synaptically connected to aCC. Two additional cholinergic pre-motor interneurons termed A18a and A18b3. The connectome identifies A23a as a GABAergic interneuron directly presynaptic to aCC, involved in locomotion and activated in both forward and backward peristaltic waves. The connectome also identified A31k as a GABAergic interneuron, synaptically connected to aCC, which delivers proprioceptive feedback to motoneurons (Giachello, 2020).

Neural circuit assembly occurs with subcellular precision, yet the mechanisms underlying this precision remain largely unknown. Subcellular synaptic specificity could be achieved by molecularly distinct subcellular domains that locally regulate synapse formation, or by axon guidance cues restricting access to one of several acceptable targets. These models have been addressed using two Drosophila neurons: the dbd sensory neuron and the A08a interneuron. In wild-type larvae, dbd synapses with the A08a medial dendrite but not the A08a lateral dendrite. dbd-specific overexpression of the guidance receptors Unc-5 or Robo-2 results in lateralization of the dbd axon, which forms anatomical and functional monosynaptic connections with the A08a lateral dendrite. It is concluded that axon guidance cues, not molecularly distinct dendritic arbors, are a major determinant of dbd-A08a subcellular synapse specificity.

Interneurons (INs) coordinate motoneuron activity to generate appropriate patterns of muscle contractions, providing animals with the ability to adjust their body posture and to move over a range of speeds. In Drosophila larvae several IN subtypes have been morphologically described and their function well documented. However, the general lack of molecular characterization of those INs prevents the identification of evolutionary counterparts in other animals, limiting the understanding of the principles underlying neuronal circuit organization and function. This study characterized a restricted subset of neurons in the nerve cord expressing the Maf transcription factor Traffic Jam (TJ). TJ(+) neurons were found to be highly diverse and selective activation of these different subtypes disrupts larval body posture and induces specific locomotor behaviors. Finally, this study shows that a small subset of TJ(+) GABAergic INs, singled out by the expression of a unique transcription factors code, controls larval crawling speed (Babski, 2019).

Animal locomotion requires spatiotemporally coordinated contraction of muscles throughout the body. This study investigate how contractions of antagonistic groups of muscles are intersegmentally coordinated during bidirectional crawling of Drosophila larvae. Two pairs of higher-order premotor excitatory interneurons present in each abdominal neuromere were identified that intersegmentally provide feedback to the adjacent neuromere during motor propagation. The two feedback neuron pairs are differentially active during either forward or backward locomotion but commonly target a group of premotor interneurons that together provide excitatory inputs to transverse muscles and inhibitory inputs to the antagonistic longitudinal muscles. Inhibition of either feedback neuron pair compromises contraction of transverse muscles in a direction-specific manner. These results suggest that the intersegmental feedback neurons coordinate contraction of synergistic muscles by acting as delay circuits representing the phase lag between segments. The identified circuit architecture also shows how bidirectional motor networks could be economically embedded in the nervous system.

Animals generate diverse motor behaviors, yet how the same motor neurons (MNs) generate two distinct or antagonistic behaviors remains an open question. This study characterized Drosophila larval muscle activity patterns and premotor/motor circuits to understand how they generate forward and backward locomotion. All body wall MNs are activated during both behaviors, but a subset of MNs change recruitment timing for each behavior. TEM was used to reconstruct a full segment of all 60 MNs and 236 premotor neurons (PMNs), including differentially-recruited MNs. Analysis of this comprehensive connectome identified PMN-MN ‘labeled line’ connectivity; PMN-MN combinatorial connectivity; asymmetric neuronal morphology; and PMN-MN circuit motifs that could all contribute to generating distinct behaviors. A recurrent network model was generated that reproduced the observed behaviors, and used functional optogenetics to validate selected model predictions. This PMN-MN connectome will provide a foundation for analyzing the full suite of larval behaviors (Zarin, 2019a and b).

This study reports a comprehensive larval proprioceptor-PMN-MN connectome and describes individual muscle/MN phase activity during both forward and backward locomotor behaviors. PMN-MN connectivity motifs were identified that could generate muscle activity phase relationships, and selected experimental validation was performed. Proprioceptor-PMN connectivity was identified that provides an anatomical explanation for the role of proprioception in promoting locomotor velocity, and it identifies a new candidate escape motor circuit. Finally, a recurrent network model was generated that produces the observed sequence of motor activity, showing that the identified pool of premotor neurons is sufficient to generate two distinct larval behaviors. It is concluded that different locomotor behaviors can be generated by a specific group of premotor neurons generating behavior-specific motor rhythms (Zarin, 2019a and b).

Locomotion is a rhythmic and flexible motor behavior that enables animals to explore and interact with their environment. Birds and insects fly, fish swim, limbed animals walk and run, and soft-body invertebrates crawl. In all cases, locomotion results from coordinated activity of muscles with different biomechanical output. This precisely regulated task is mediated by neural circuits composed of motor neurons (MNs), premotor interneurons (PMNs), proprioceptors, and descending command-like neurons. A partial map of neurons and circuits regulating rhythmic locomotion have been made in mouse, cat, fish, tadpole, lamprey, leech, crayfish, and worm. These pioneering studies have provided a wealth of information on motor circuits, but with the exception of C. elegans, there has been no system where all MNs and PMNs have been identified and characterized. Thus, a comprehensive picture of how an ensemble of interconnected neurons generate diverse locomotor behaviors is missing (Zarin, 2019a and b).

How does the Drosophila larva executes multiple behaviors, in particular forward versus backward locomotion. Are there different motor neurons used in each behavior? Are the same motor neurons used but with distinct patterns of activity determined by premotor inputs? How does the ensemble of premotor and motor neurons generate additional behaviors, such as escape behavior via lateral rolling? A rigorous answer to these questions requires both comprehensive anatomical information -- i.e., a premotor/motor neuron connectome -- and the ability to measure rhythmic neuronal activity and perform functional experiments. All of these tools are currently available in Drosophila, and this study used them to characterize the neuronal circuitry used to generate forward and backward locomotion, and how proprioception is integrated by the PMN ensemble (Zarin, 2019a and b).

The Drosophila larva is composed of 3 thoracic (T1-T3) and 9 abdominal segments (A1-A9), with sensory neurons extending from the periphery into the CNS, and motor neurons extending out of the CNS to innervate body wall muscles. Most segments contain 30 bilateral body wall muscles that are grouped by spatial location and orientation: dorsal longitudinal (DL; includes previously described DA and some DO muscles), dorsal oblique (DO), ventral longitudinal (VL), ventral oblique (VO), ventral acute (VA) and lateral transverse (LT). Using these muscles, the larval nervous system can generate forward locomotion, backward locomotion, turning, hunching, digging, self-righting, and escape. This study focused on forward and backward locomotion. Forward crawling behavior in larvae involves a peristaltic contraction wave from posterior to anterior segments; backward crawling entails a posterior propagation of the contraction wave (Zarin, 2019a and b).

Body wall muscles are innervated by approximately 60 MNs per segment, consisting of 28 left/right pairs that typically each innervate one muscle, and whose neuromuscular junctions have big boutons, therefore also called type-Ib MNs; two pairs of type-Is (small bouton) MNs that innervate large groups of dorsal or ventral muscles; three type II ventral unpaired median MNs that provide octopaminergic innervation to most muscles; and one or two type III insulinergic MNs innervating muscle 12. All MNs in segment A1 have been identified by backfills from their target muscles, and several have been shown to be rhythmically active during larval locomotion, but only a few of their premotor inputs have been described. Some excitatory PMNs are involved in initiating activity in their target MNs, while some inhibitory PMNs limit the duration of MN activity or produce intrasegmental activity offsets. Interestingly, some PMNs are active specifically during forward locomotion or backward locomotion. In addition, there are six pair of proprioceptor neurons in each abdominal segment (ddaE, ddaD, vpda, dmd1, dbd and vbd). They are important for promoting locomotor velocity and posture, and some of their CNS targets have been identified, but to date little is known about how or if they are directly connected to the PMN/MN circuits (Zarin, 2019a and b).

It is a major goal of neuroscience to comprehensively reconstruct neuronal circuits that generate specific behaviors, but to date this has been done only in C. elegans. Recent studies in mice and zebrafish have shed light on the overall distribution of PMNs and their connections to several well-defined MN pools. However, it remains unknown if there are additional PMNs that have yet to be characterized, nor are their any insights into potential connections between PMNs themselves, which would be important for understanding the network properties that produce coordinated motor output. In the locomotor central pattern generator circuitry of leech, lamprey, and crayfish, the synaptic connectivity between PMNs or between PMNs and other interneurons are known to play critical roles in regulating the swimming behavior. However, it is difficult to be certain that all the neural components and connections of these circuits have been identified. Thus, the comprehensive anatomical circuitry reconstructed in this study provides an anatomical constraint on the functional connectivity used to drive larval locomotion; all synaptically-connected neurons may not be relevant, but at least no highly connected local PMN is absent from this analysis (Zarin, 2019a and b).

The current results confirm and significantly extend previous studies of Drosophila larval locomotion. For example, a recent study has shown that the GABAergic A14a inhibitory PMN (also called iIN1) selectively inhibits MNs innervating muscle 22/LT2 (CMuG F4; CMuG refers to Co-active Muscle Group), thereby delaying muscle contraction relative to muscle 5/LO1 (CMuG F2). This study was extended by showing that A14a also disinhibits MNs in early CMuGs F1/2 via the inhibitory PMN A02e. Thus, A14a both inhibits late CMuGs and disinhibits early CMuGs. In addition, previous work has suggested that all MNs receive simultaneous excitatory inputs from different cholinergic PMNs. However, dual calcium imaging data of the A27h excitatory PMN shows that it is active during CMuG F4 and not earlier. Therefore, MNs may receive temporally distinct excitatory inputs, in addition to the previously reported temporally distinct inhibitory inputs. This study has identified dozens of new PMNs that are candidates for regulating motor rhythms; functional analysis of all of these PMNs is beyond the scope of this paper, particularly due to the additional work required to screen and identify Gal4/LexA lines selectively targeting these PMNs, but the predictions of this paper are clear and testable when reagents become available (Zarin, 2019a and b).

MNs innervating a single Spatial Muscle Group (SMuG) belong to more than one CMuG, therefore SMuGs do not generally match CMuGs. This could be due to the several reasons: (1) MNs in each SMuGs receive inputs from overlapping but not identical array of PMNs. (2) Different MNs in the same SMuG receive a different number of synapses from the same PMN. (3) MNs in the same SMuG vary in overall dendritic size and total number of post-synapses, thereby resulting in MNs of the same SMuGs fall into different CMuGs (Zarin, 2019a and b).

This study demonstrates that during both forward and backward crawling, most of longitudinal and transverse muscles of a given segment contract as early and late groups, respectively. In contrast, muscles with oblique or acute orientation often show different phase relationships during forward and backward crawling. Future studies will be needed to provide a biomechanical explanation for why oblique muscles -- but not longitudinal or transverse muscles -- need to be recruited differentially during forward or backward crawling. Also, it will be interesting to determine which spatial muscle groups (e.g., either or both VOs and VLs) are responsible for elevating cuticular denticles during propagation of the peristaltic wave in forward and backward crawling; if the VOs, it would mean that lifting the denticles occurs at different phases of the crawl cycle in forward and backward locomotion. Finally, understanding how the premotor-motor circuits described in this study are used to generate diverse larval motor behaviors will shed light on mechanisms underlying the multi-functionality of neuronal circuits (Zarin, 2019a and b).

A recent study has reported that proprioceptive sensory neurons (dbd, vbd, vpda, dmd1, ddaE, and ddaD) show sequential activity during forward crawling. dbd responds to stretching and whereas the other five classes are activated by muscle contraction (Vaadia, 2019). All proprioceptors show connectivity to the tier of PMNs described in this study, and this study has identified circuit motifs that are consistent with the observed timing and excitatory function of each proprioceptor neuron. It will be of great interest perform functional experiments to test these anatomical circuit motifs for functional relevance (Zarin, 2019a and b).

A recurrent network model accurately predicts the order of activation of specific PMNs, yet many of its parameters remain unconstrained, and some PMNs may have biological activity inconsistent with activity predicted by this model. Sources of uncertainty in the model include incomplete reconstruction of inter-segmental connectivity and descending command inputs, the potential role of gap junctions (which are not resolved in the TEM reconstruction), as well as incomplete characterization of PMN and MN biophysical properties. Recent studies have suggested that models constrained by TEM reconstructions of neuronal connectivity are capable of predicting features of neuronal activity and function in the Drosophila olfactory and visual systems, despite the unavoidable uncertainty in some model parameters. Similarly, for the locomotor circuit described in this study, it is anticipated that the addition of model constraints from future experiments will lead to progressively more accurate models of PMN and MN dynamics. Despite it's limitations, the ability for the PMN network to generate appropriate muscle timing for two distinct behaviors in the absence of any third-layer or command-like interneurons suggests that a single layer of recurrent circuitry is sufficient to generate multiple behavioral outputs, and provides insight into the network architecture of multifunctional pattern generating circuits (Zarin, 2019a and b).

Previous work in other animal models have identified multifunctional muscles involved in more than one motor behavior: swimming and crawling in C. elegans and leech; walking and flight in locust; respiratory and non-respiratory functions of mammalian diaphragm muscle unifunctional muscles which are only active in one specific behavior in the lobster Homarus americanus; swimming in the marine mollusk Tritonia diomedea; and muscles in different regions of crab and lobster stomach. Single-muscle calcium imaging data indicates that all imaged larval body wall muscles are bifunctional and are activated during both forward and backward locomotion. It will be interesting to determine if all imaged muscles are also involved in other larval behaviors, such as escape rolling, self-righting, turning, or digging. It is likely that there are different CMuGs for each behavior, as this study has \ seen for forward and backward locomotion, raising the question of how different CMuGs are generated for each distinct behavior (Zarin, 2019a and b).

The Drosophila giant fiber (GF) escape circuit is an extensively studied model for neuron connectivity and function. Researchers have long taken advantage of the simple linear neuronal pathway, which begins at peripheral sensory modalities, travels through the central GF interneuron (GFI) to motor neurons, and terminates on wing/leg muscles. This circuit is more complex than it seems, however, as there exists a complex web of coupled neurons connected to the GFI that widely innervates the thoracic ganglion. This study defines four new neuron clusters dye coupled to the central GFI, which were named GF coupled (GFC) 1-4. New transgenic Gal4 drivers were identified that express specifically in these neurons, and both neuronal architecture and synaptic polarity were mapped. GFC1-4 share a central site of GFI connectivity, the inframedial bridge, where the neurons each form electrical synapses. Targeted apoptotic ablation of GFC1 reveals a key role for the proper development of the GF circuit, including the maintenance of GFI connectivity with upstream and downstream synaptic partners. GFC1 ablation frequently results in the loss of one GFI, which is always compensated for by contralateral innervation from a branch of the persisting GFI axon. Overall, this work reveals extensively coupled interconnectivity within the GF circuit, and the requirement of coupled neurons for circuit development. Identification of this large population of electrically coupled neurons in this classic model, and the ability to genetically manipulate these electrically synapsed neurons, expands the GF system capabilities for the nuanced, sophisticated circuit dissection necessary for deeper investigations into brain formation (Kennedy, 2018).

Rapid and efficient escape behaviors in response to noxious sensory stimuli are essential for protection and survival. Yet, how noxious stimuli are transformed to coordinated escape behaviors remains poorly understood. In Drosophila larvae, noxious stimuli trigger sequential body bending and corkscrew-like rolling behavior. A population of interneurons in the nerve cord of Drosophila, termed Down-and-Back (DnB) neurons, was identified that are activated by noxious heat, promote nociceptive behavior, and are required for robust escape responses to noxious stimuli. Electron microscopic circuit reconstruction shows that DnBs are targets of nociceptive and mechanosensory neurons, are directly presynaptic to pre-motor circuits, and link indirectly to Goro rolling command-like neurons. DnB activation promotes activity in Goro neurons, and coincident inactivation of Goro neurons prevents the rolling sequence but leaves intact body bending motor responses. Thus, activity from nociceptors to DnB interneurons coordinates modular elements of nociceptive escape behavior (Burgos, 2018).

Nocifensive escape behavior in Drosophila larvae consists of C-shaped body bending and rolling, followed by rapid forward crawling. Recent studies have begun to identify circuits that mediate nocifensive behaviors. Prior work identified Basin neurons as multisensory interneurons that drive rolling behavior in response to vibration and noxious stimuli, and identified downstream Goro as command-like neurons for rolling. This study has identified and characterized DnB interneurons that are essential for nocifensive behavior in Drosophila larvae (see Summary model for DnB neurons controlling nocifensive escape). DnB neurons are direct targets of nociceptive cIV neurons and multiple mechanosensory cell types, including cII and cIII gentle touch da neurons and es neurons. Thus, DnBs provide a potential node for multisensory integration of tactile and noxious stimuli. The convergence of input from cIII gentle-touch receptors and cIV nociceptors onto DnB neurons is reminiscent of vertebrate interneurons that receive direct excitatory input from C-fiber/A∂ nociceptors and Aβ mechanoreceptors. Based on these studies nociceptive inputs appear to be integrated with multiple mechanosensory submodalities by Basin and DnB interneurons (Burgos, 2018).

EM reconstruction of DnB targets supported divergent major downstream circuitry. Output synapses on DnB axons converge on premotor neurons, at least some of which promote peristaltic wave propagation during locomotion. Other downstream neurons receive input from presynaptic sites on the DnB dendrite, and lead to Goro rolling command-like neurons. The spatial segregation of DnB output sites may mirror a functional segregation of downstream circuitry into bending and rolling modules. It is still unclear which muscle groups are recruited and how segments coordinate during body bending and rolling. This study provides evidence that silencing the period-positive median segmental interneuron (PMSI) cohort, which includes direct DnB targets A02g and A02e, reduces rolling behavior. PMSIs are glutamatergic inhibitory premotor neurons that terminate motor neuron bursting to regulate crawling speed. Future work to selectively silence groups of premotor neurons will help to elucidate their role in nocifensive escape downstream of DnBs. Although silencing DnB neurons slightly increased the speed of forward locomotion, overall, forward crawling remained intact. Given that peristaltic waves also consist of segmental contractions, links to premotor neurons provide candidate neurons for dual control of crawling and C-shape bending behavior. Notably, DnB neurons target motor neurons innervating LT1 muscles, which have been implicated in larval self-righting behaviors. Self-righting consists of a C-shape type body bend, and 180° turn, so it is possible that LT1 muscles facilitate curved body bends that underlie both self-righting and rolling behavior. It is noted that the impact of DnB neurons on nociceptive circuits is likely to be more broad than indicated by synaptic connections, since EM and marker expression suggest that DnB neurons are peptidergic. Identification of the putative neuropeptide expressed by DnB neurons, and physiological effects, will be an important future direction, particularly given the important role of neuropeptides in vertebrate pain pathways, and recent evidence that mechanical nociception in larvae is under peptidergic control (Burgos, 2018).

Prior data showed that rolling is directional and is advantageous for dislodging attacking parasitoid wasps. Efficient rolling occurs coincident with deep C-shaped body bends, but the significance of these body bends for escape behavior has not been determined. DnB neural circuitry appears to be critically important for evoking body bend behavior prior to and during nocifensive rolling. Bending may provide the initial, most rapid, form of withdrawal from a noxious stimulus, and may subsequently support rolling locomotion by orienting and focusing the energy of muscle contraction into lateral thrusts. Re-orientation of denticle belts, triangle-shaped extensions of the cuticle, may also aid rapid lateral locomotion by providing substrate traction. Compromised escape rolling upon DnB inactivation may therefore arise both from weakened Goro activation and decreases in body bend angle. Understanding the circuit mechanisms that promote bending downstream of DnB neurons, and the muscle activities and physical mechanisms that underlie rolling behavior are important future aims (Burgos, 2018).

Analysis of DnB function revealed modular control of nocifensive escape behavior, consistent with EM reconstruction data. When DnB neurons were ectopically activated C-shaped body bending was observed that was often, but not always, associated with rolling. Other, non-rolling, animals bent with minimal crawling, or bent persistently while attempting to crawl forward. These observations provided initial evidence that C-shaped bending and rolling control circuits are separable, and that nocifensive bending could be combined with other behaviors, like pausing or crawling. Loss of function data supported bending as a primary motor output of DnB activity, with probabilistic activation of rolling motor programs. These behaviors could conceivably be linked, such that reduction in bending compromises rolling ability, or could arise from parallel influence of DnB activity on bending and rolling as suggested by EM reconstruction. Consistent with an important role for DnBs in promoting rolling, silencing Goro while activating DnB neurons promoted persistent bending without rolling, and uncoordinated snake-like forward crawling. This result further implicates a separate premotor circuitry in nocifensive body bending. These data further suggest that the bend-roll sequence must be tightly regulated by interactions between the parallel bend-roll premotor circuits, such that bending occurs first to facilitate rolling, which occurs second. However, bending can occur without being followed by rolling, indicating C-shaped bending itself is not sufficient to trigger rolling. Such independent, but sequentially regulated behavioral modules are consistent with hierarchical models of sequence generation as in fly grooming, human speech, roll-crawl sequence, and hunch-bend sequence. It is noted however, that although bending and rolling are sequential, they co-occur for much of the defensive behavior sequence, in contrast to such sequential and non-overlapping behavioral sequences. Elucidating the mechanisms of timing and interaction between the different circuit modules (bend vs roll) identified therefore promises to shed light on the general mechanisms of circuit implementation of sequence generation and co-ordination between different motor modules (Burgos, 2018).

In most animals, the brain controls the body via a set of descending neurons (DNs) that traverse the neck. DN activity activates, maintains or modulates locomotion and other behaviors. Individual DNs have been well-studied in species from insects to primates, but little is known about overall connectivity patterns across the DN population. This study systematically investigated DN anatomy in Drosophila melanogaster and created over 100 transgenic lines targeting individual cell types. Roughly half of all Drosophila DNs were identified and connectivity between sensory and motor neuropils in the brain and nerve cord, respectively, were comprehensively mapped. The nerve cord was found to be a layered system of neuropils reflecting the fly's capability for two largely independent means of locomotion -- walking and flight -- using distinct sets of appendages. These results reveal the basic functional map of descending pathways in flies and provide tools for systematic interrogation of neural circuits (Namiki, 2018).

This study systematically characterized the organization of DNs, a population of interneurons that conduct information from the brain of a fly to motor centers in the VNC. This analysis was based on the morphologies of 98 DN cell types, covering 190 bilateral pairs of neurons. To discern DN morphologies, individual neurons from driver lines targeting many cells were segmented, and also a library of 133 split-GAL4 lines were generated that sparsely target 54 DN types. By registering the morphology of all the DNs with standardized maps of the brain and VNC, three major sensory-motor pathways. One pathway links two neuropils on the posterior slope of the brain (IPS and SPS) to dorsal neuropils associated with the neck, wing, and haltere motor systems, and a second carries neurons with dendrites in the gnathal ganglion (GNG) to the leg neuromeres. The third pathway consists of DNs originating from an array of brain neuropils that converge to innervate the tectulum, a long thin region of the VNC sandwiched between the wing and leg motor neuropils (Namiki, 2018).

The simple, tripartite anatomical pattern that was observed may reflect both the functional organization of the DNs as well as the evolutionary history of Drosophila. With the notable exception of insects (and the mythical horse, Pegasus), all flying animals use a modified foreleg as a wing. That is, an appendage originally evolved for walking was coopted for flight in pterosaurs, birds, and bats - a fact supported by the fossil record, comparative morphology, and the organization of the underlying motor circuitry. The evolution of flight was quite different in insects, because their wings and associated muscles, did not arise via sacrifice of an entire ancestral leg, and thus the novel aerial mode of locomotion did not strongly compromise the more ancient, terrestrial mode. As a result, insects are unique in possessing two somewhat independent motor systems, a fact that is elegantly manifest in the organization of the VNC and the pattern of DN innervation that was observed: the ventral leg neuromeres of flies resemble those of apterygote hexapods from which they derived, whereas the more recent wing neuropil sits atop the VNC like icing on a cake. It is speculated that the GNG-to-leg neuromere descending pathway represents a very ancient pathway and some of its member DNs may have deep homologies with other arthropod taxa, whereas the pathway linking the posterior slope neuropils to the dorsal motor neuropils of the neck, wing, and haltere are more recently evolved within insects (Namiki, 2018).

Many behaviors such as grooming, courtship, take-off, and landing require the simultaneous use of both legs and wings. Thus, insects must have a means of coordinating activity across the two motor systems, a need that arose during or after the evolution of flight. As described more fully below, it is speculated that the teculum, and possibly the lower teculum, are neuropils that mediate this functional integration of motor actions between the two systems. The convergence of DNs into the tectulum from such a broad array of brain nuclei may reflect the high degree of sensory integration required to trigger and regulate these more complex, multi-appendage behaviors (Namiki, 2018).

Based on PA-GFP labeling of neurons in the neck connective, ~350 DN pairs were counted. This is within the range of 200-500 DN pairs estimated in other insect species, but smaller than a value of ~550 pairs estimated in Drosophila based on backfills using a dextran dye. Part of this discrepancy can be explained by the fact that the current count excluded several specialized cell populations that were included in a previous study. These include a set of ~19 pairs of neck motor neurons, whose axons exit the neck connective posterior to the region illuminated for PA-GFP photoconversion, as well as 16 neurons selectively innervating the retrocerebral complex. One of these cells (DNd01), which innervates both the VNC and retrocerebral complex, was included. The current analysis is also likely an underestimate of the total because the nsyb-LexA driver line used to pan-neuronally express PA-GFP, may not label all neurons. For example, this line does not label the Giant Fiber. It is also possible that certain cells are harder to label using the PA-GFP approach as opposed to dextran backfills. The estimates from the two studies agree quite closely for DNs with cell bodies in the cerebral ganglia (172 in this study). Most of the discrepancy with previous studies concerns DNs in the GNG group; this study counted 180 pairs, only 51% of the number by a previous study. Taking the estimate of 350 as a lower bound and 550 an upper bound, it is estimated that the DNs described in this study represent between one third and one half of the entire population (Namiki, 2018).

Identification of particular DN types in this study relied on the existence of a GAL4-line in the Rubin or Vienna (BrainBase) collection with sparse enough expression to recognize individual DN morphology. Additionally, most of the expression patterns that were screened were from female flies, thus the analysis would not include any potential male-specific DNs. As a result, some DNs were not found that have been reported in other studies, including the Moonwalker Descending Neuron (MDN), which controls backwards walking in flies, and pIP10/p2b, which are involved in the male courtship sequence (Namiki, 2018).

A direct pathway was found linking the posterior slope of the brain to dorsal VNC neuropils. The posterior slope is innervated by lobula plate tangential cells (LPTCs) projecting from the optic lobe, which are excited by patterns of optic flow resulting from self-rotation. These optic flow patterns are especially relevant during flight, when the fly is able to move freely about all six degrees of freedom, and it has been suggested that LPTCs mediate both corrective steering maneuvers of the wings as well as gaze stabilization of the head. Most of the DNs in this pathway targeted all three segmental dorsal VNC neuropils, which contain neck (T1), wing (T2), or haltere (T3) motor neurons, sensory neuron projections from associated mechanoreceptors, and premotor interneurons. DN innervation of all three segmental dorsal neuropils is consistent with recent studies showing that neck and wing movements are highly correlated and suggests that the DNs of this major posterior slope-to-dorsal neuropil pathway are involved in flight control. This notion is confirmed by recent whole cell recordings from tethered flying flies showing that three members of this population are strongly correlated with compensatory visual responses, and another is involved with spontaneous turns and collision avoidance (Namiki, 2018).

A similar pathway, in which DNs receiving inputs in the posterior slope target flight neuropil, has been observed in blowflies and flesh flies. These have been contrasted with other DNs in the protocerebrum that have anterior dendrites near the outputs of the lobula that project to ventral leg neuropils. They suggested that the posterior and anterior DN protocerebral pathways are parallel systems linked to separate photoreceptor channels that process different features of the visual scene (e.g. color vs. motion) and may be loosely analogous to the dorsal and ventral streams of the mammalian visual system. The dataset allowed evaluation of this hypothesis in Drosophila by examining the subset of 42 DNs with dendrites in the protocerebrum. In keeping with the observations from large fly species, examples were found in which a DN with more posterior dendrites (e.g. DNg02) projected to the dorsal part of the VNC, whereas a DN with anterior dendrites (e.g. DNg13) projected to the ventral leg neuropils. It was also found that the median location of a DN's dendrites along anterior-posterior axis largely predicted whether its axons targeted dorsal or ventral leg neuropil (although see exceptions DNb01, DNb06, DNp07, and DNp18). However, the dendritic locations of DNs projecting to the dorsal and leg neuropils of the VNC were not segregated into separable, parallel groups, but instead form a continuous pattern of innervation in the protocerebrum. That is, the DN representation is graded in the protocerebrum, at least at the level of resolution of the analysis. Furthermore, the dendritic arbors of many DNs are broad enough that they sample from both anterior and posterior regions of the protocerebrum, suggesting that many DNs integrate information from both the lobula and lobula plate. Rather than the two separate parallel pathways suggested previously - one carrying visual information from the lobula plate to the wing neuropil and the other carrying information from the lobula to the leg neuropil - it is proposed that there is a mixing of this visual information in the protocerebrum, possibly in a graded manner along the anterior-posterior axis. A similar divergence and convergence of connectivity has been described in the brainstem of mice. Brainstem nuclei differentially address spinal circuits, forming exclusive connections either with forelimbs, hindlimbs, or both with differing connection strength (Namiki, 2018).

Among all DNs targeting the wing neuropil, evidence was found for at least two distinct control systems, one entering the neuropil from a dorsal tract and targeting the dorsal and medial portion of the wing neuropil layer, where power muscle motor neuron dendrites reside, and one entering the neuropil from a more ventral tract and invading primarily the ventral and medial wing neuropil, where many steering muscle motor neurons dendrites reside. In Drosophila, the power muscles comprise two sets of stretch activated muscles attached across the thorax orthogonally. Alternate deformation of the thoracic cavity by these muscles drives the wing stroke indirectly, powering both flight and courtship song. In contrast, the smaller steering muscles attach to the base of the wing hinge and act directly to coordinate the wing movements that change flight course, or actuate finer movement, such as the timing of song pulses. The results suggest separate descending control of the power and steering muscle systems. Outside of flight and song, flies perform a wide range of different behaviors with their wings, including grooming, aggressive displays, and preparation for takeoff. Although this study found that the posterior slope had the largest number of DNs innervating wing neuropil, a wide range of other brain neuropils, including the gnathal ganglia (GNG), VES, lobula plate (PLP), anterior mechanosensory motor center (AMMC), SAD, superior medial protocerebrum (SMP), and lateral accessory lobe (LAL), are also connected to the wing neuropil, albeit via a smaller number of DNs (see Anatomical compartments of the brain and VNC in Drosophila). These sparser pathways may be important for coordinating wing motion when the flies are not flying (Namiki, 2018).

Despite the trend described in the previous section, in which DNs with more anterior dendrites in the protocerebrum tend to target leg neuropil, this analysis found that a different brain region, the GNG, had the strongest DN connectivity to the six ventral neuromeres of the VNC. This was true even after excluding the many DNs whose neurites are presynaptic in the GNG. Indeed, 90% (88/98) of the DN types have processes in the GNG, most of which are varicose terminals containing synaptogagmin, and thus likely output terminals. Only one-third (29/88) of DNs with processes in the GNG had dendrites in that region, two-thirds of which (18/29) target leg neuropil without any terminals in the dorsal wing, neck, or haltere neuropils (Namiki, 2018).

Given the GNG's evolutionary history as a separate anterior segmental ganglion, it is perhaps not surprising that this neuropil is strongly connected to more posterior motor centers. The suboesophageal ganglion, which includes the GNG, is involved in a variety of behaviors, including walking, stridulation, flight initiation, head movement, and respiration. However, the GNG has been most specifically implicated in the temporal patterning of walking. For example, both supra- and subesophageal DNs are recruited in the preparatory phase before walking, whereas the activity of subesophageal DNs become predominant during the walking phase (Namiki, 2018).

The terminals of DNs targeting the same layers of the VNC clustered together within the GNG. One intriguing possibility is that these foci represent regions in which efferent copies of descending commands to leg and wing motor centers are available to cephalic sensory circuits. This information could then be integrated directly with other descending commands within the GNG, or reciprocal connections could feed the information back to the cerebral ganglia. The GNG also receives ascending inputs from the leg neuropil, allowing further integration within this region of information regarding locomotor state or mechanosensory input. Given that the cerebral ganglia are known to have a strong inhibitory effect on walking in insects, another possibility is that some DN terminals in the GNG are inhibitory. Indeed, a recent study found that 37% of DNs express the inhibitory neurotransmitter GABA, compared to 38% that are cholinergic, and just such an inhibitory pathway from the cerebral ganglia to the GNG has been suggested based on prior behavioral experiments. For example, lesion studies have shown that walking persists when the cerebral ganglia are removed and spontaneous bouts are prolonged. In contrast, removal of the GNG reduces spontaneous walking, but prolongs flight duration. Thus it is possible that the DN pathway identified, linking the posterior slope to wing neuropil, maintains flight and inhibits walking, whereas the pathway linking the GNG to the leg neuropils maintains walking and inhibits flight. Thus, the connections within the GNG may play a critical role in action selection, at least at a coarse level (Namiki, 2018).

DN terminals in the leg neuropils could be sorted into two major types: DNs projecting to the dorso-medial part of each neuromere (type-I) and DNs penetrating through the neuromeres via the oblique tract (type-II). Their terminal locations suggest that type-I and type-II leg DNs may have different access to leg motor neurons because the dendrites are known to form a rough myotopic map across the leg neuromere, with more proximal leg muscles having more proximal dendrites. Based on this arrangement, one possible function of the type-I leg DNs is to coordinate the direction of walking, which depends critically on the control of coxal muscles that protract and retract of the entire leg. Indeed, inverse activation of the thoraco-coxal muscle is required for switching from forward to backward walking in stick insects. In Drosophila, moonwalker DNs (MDNs) innervate the dorso-medial part of the leg neuropil and thus are classified as type-I. Activation of bilateral MDNs cause backward locomotion, whereas the unilateral activation cause backward turning toward the contralateral side. Type-II DNs running through the oblique tract have the opportunity to contact with the entire array of proximal and distal motor neurons and thus may be important for coordinated action of all leg segments. For example, the jumping part of escape takeoffs may require tension in all leg segments, even though the extrinsic muscle extending the trochanter is the primary actor for the fast takeoff mode. Consistent with this idea, type-II DNs are abundant in mesothoracic leg neuropil (DNp02, p05, p06 and p11), and it is the middle legs that flies extend during a jump. Similarly, in locust, the descending contralateral movement detector (DCMD), which is important for escape behavior, has terminals that resemble type-II and synapses directly on the motor neurons in the neuropil associated with the jumping legs (Namiki, 2018).

A small population of nine DNs specifically project to an intermediate zone of the VNC, the lower tectulum, which occupies a volume distinct from wing and leg neuropils and which this study suggests can be distinguished from the other intermediate neuropil, the tectulum, that sits above it. Neuronal connectivity is not well described in this region, and its function is unknown. However, the observations suggest that, like the tectulum, it is an integrative area involved in both leg and wing control. For example, this region includes dendrites from both the tergotrochanteral leg motor neuron (TTMn) and a branch of a wing motor neuron that has been tentatively identified as III1. The lower teculum also contains the peripheral synapsing interneuron (PSI), which is presynaptic to motor neurons for the wing depressor muscles. The giant fiber (GF) descending neurons that drive a looming-evoked escape takeoff terminate with unbranched axons within the lower tectulum and form gap junctions with the TTMn and PSI. It was surmising that the lower tectulum may play a role during takeoff, which requires coordinated actions of the wings and legs. It is known that there are parallel pathways for take-off behavior in Drosophila, although the anatomical source has not yet been identified. A group of eight unique type DNs, in addition to the GF, were identified whose dendrites overlap with the terminals of visual projection neurons that detect looming. Most of these invade the lower tectulum and their axon terminals share some anatomical features with the GF. This population are candidates for parallel pathways for takeoff, as well as other looming-evoked evasive behaviors, and could represent circuits for wing-leg coordination (Namiki, 2018).

No DNs were found that originate in the central complex (CX), consistent with studies in other insect species. Thus, information from the CX must be relayed to motor centers via other brain regions. A prime candidate is the the lateral accessory lobe (LAL), which has dense mutual connections with the CX and, together with the bulb (BU), is considered the CX primary output. However, many fewer DNs from the LAL were found than from other regions such as PS, PVLP or AMMC. In other insects such as silk moths, connections between the LAL and the PS are well documented. In Drosophila, connectivity between the LAL and PS is suggested by connectomics studies and the morphology of individual neurons connecting these regions has been recently described. Thus, it is suggested that information processed in the CX may descend to the VNC via a CX-LAL-PS pathway (Namiki, 2018).

No DNs were found originating from the mushroom bodies (MB), important processing areas for olfactory and visual memory. However, there are 11 DN types innervating the superior medial protocerebrum (SMP), a major target of MB output neurons. The SMP is also well connected with the LAL, which suggests MB output also uses the major descending pathway from the posterior slope via the LAL (Namiki, 2018).

Prior studies in insects have focused on DN function at the single neuron level. Thus, how DNs operate as a population is still unclear. Evidence in insects and other species suggests that motor directives are likely encoded across the DN population rather than in the activity of individual command neurons. For example, many DNs are active, albeit with different firing patterns, at the same time during walking in locusts, and there are multiple brain locations where electrical stimulation can trigger walking behavior in cockroaches. Also, population vector coding for object direction has been observed in the DNs of dragonflies. Zebrafish have also been shown to utilize population coding in the control of locomotion, despite having only ~220 DNs -- even fewer than Drosophila. In fact, there are very few neurons that fit the rigorous requirements of command neuron (i.e. necessary and sufficient). Even the giant fibers (a.k.a. DNp01), whose activation drives a stereotyped escape jump in response to looming stimuli, are necessary only for a particular 'fast mode' of takeoff, and the behavioral effect of their activation to naturalistic looming stimuli has been shown to depend on the timing of their spike relative to activity in other descending neurons (Namiki, 2018).

This study found that the VNC areas receiving the largest number of DNs are the dorsal neuropils associated with flight control (neck, wing, haltere neuropils and tectulum). It has been suggested that the number of DNs engaged during a behavior might relate to the precision of the control. In mammals, for example, the number of activated corticospinal tract neurons corresponds to the degree of digital dexterity. It is possible a large DN population target flight neuropils because flight requires a high level of precise control. For example, flies can execute sophisticated rapid aerial turns to evade a looming predator, movements that are controlled by a combination of adjustments in firing phase of tonically active motor neurons and recruitment of phasically active cells (Namiki, 2018).

In addition to the number of DNs putatively assigned to wing control, this study found that the organization of wing DNs is different than that of the DNs targeting leg neuropil. Several distinct clusters of DNs were identified with nearly identical morphologies and highly overlapped input and output projections, which are referred to as population type DNs because their similar morphology suggests they may function as a group (e.g. DNg01, g02, g03, g05 and g06). In most cases, these population DNs project to the wing neuropil or tectulum and are thus likely involved in flight. In contrast, only unique type DNs (identifiable single bilateral pairs) projecting to leg neuropil were found. This suggests that the strategy for controlling flight and walking may be fundamentally different. Because of the physics involved, even very small changes in wing motion during flight can result in large aerodynamic forces and moments. The necessity for fine control might account for the greater dependence on population coding in flight as compared to walking. Another difference between flight and walking is the temporal scale required for control. For example, wingbeat frequency is much faster than leg stepping frequency. The control of force generation by wing steering muscles depends on the precise timing of motor neuron spikes. The descending input during flight must have the capacity to regulate motor neuron firing phase on a precise temporal scale, a functionality that might be achieved via population coding. Another possibility is that the number of active DNs encodes the magnitude of a command signal to regulate continuous locomotor parameters such as speed. In larval zebrafish and lamprey, for example, more reticulospinal DNs are recruited with increasing swimming frequency. Further functional studies will be required to test whether DN encoding of flight and walking commands operates by different principles (Namiki, 2018).

This study has analyzed the neuronal organization of descending motor pathways in Drosophila, with single-cell resolution. The wiring diagram revealed, in a genetically accessible model system, creates a framework for understanding of how the brain controls behavior. In combination with the Drosophila genetic toolkit, the driver lines created in the present study open up the possibility to directly probe the function of individual DNs during natural behavior (Namiki, 2018).

The extensive genetic regulatory flows underlying specification of different neuronal subtypes are not well understood at the molecular level. The Nplp1 neuropeptide neurons in the developing Drosophila nerve cord belong to two sub-classes; Tv1 and dAp neurons, generated by two distinct progenitors. Nplp1 neurons are specified by spatial cues; the Hox homeotic network and GATA factor grn, and temporal cues; the hb -> Kr -> Pdm -> cas -> grh temporal cascade. These spatio-temporal cues combine into two distinct codes; one for Tv1 and one for dAp neurons that activate a common terminal selector feedforward cascade of col -> ap/eya -> dimm -> Nplp1. This study molecularly decodes the specification of Nplp1 neurons, and finds that the cis-regulatory organization of col functions as an integratory node for the different spatio-temporal combinatorial codes. These findings may provide a logical framework for addressing spatio-temporal control of neuronal sub-type specification in other systems (Stratmann, 2017).

The Drosophila ventral nerve cord (VNC; defined in this study as thoracic segments T1-T3 and abdominal A1-A10) contains ~10,000 cells at the end of embryogenesis, which are generated by a defined set of ~800 neuroblasts (NBs). The Apterous neurons constitute a small sub-group of interneurons, identifiable by the selective expression of the Apterous (Ap) LIM-homeodomain factor, as well as the Eyes absent (Eya) transcriptional co-factor and nuclear phosphatase. A subset of Ap neurons express the Nplp1 neuropeptide, but can be sub-divided into the lateral thoracic Tv1 neurons, part of the thoracic Ap cluster of four cells, and the dorsal medial row of dAp neurons. In line with the distinct location of the Tv1 and dAp neurons, studies have revealed that they are generated by distinct NBs; NB5-6T and NB4-3, respectively. A number of studies have addressed the genetic mechanisms underlying the specification of the Tv1 and dAp neurons, and the regulation of the Nplp1 neuropeptide. These have revealed that two distinct spatio-temporal combinatorial transcription factor codes, one acting in NB5-6T and the other in NB4-3, converge on a common initiator terminal selector gene; collier, encoding a COE/EBF transcription factor. Col in turn is necessary and sufficient to trigger a feed forward loop (FFL) consisting of Ap, Eya and the Dimmed (Dimm) bHLH transcription factor, which ultimately activates the Nplp1 gene. Strikingly, the combinatorial coding selectivity of the spatio-temporal cues combined with the information-coding capacity of the FFL results in the selective activation of Nplp1 in only 28 out of the ~10,000 cells within the VNC. While these genetic studies have helped resolve the regulatory logic of this cell specification event, they have not addressed the molecular mechanisms by which the two different spatio-temporal combinatorial codes intersect upon the col initiator terminal selector, to trigger a common terminal FFL, or the molecular nature of the FFL (Stratmann, 2017).

To address this issue, this study has identified enhancers for Tv and dAp neuron expression for the genes in the common Tv1/dAp FFL: col, ap, eya, dimm and Nplp1. Transgenic reporters were generated for these enhancers, both wildtype and mutant for specific transcription factor binding sites, to test their regulation in mutant and misexpression backgrounds. CRISPR/Cas9 technology was used to delete these enhancers in their normal genomic location to test their necessity for gene regulation. Strikingly, this study found that the distinct upstream spatio-temporal combinatorial codes, which trigger col expression in Tv1 versus dAp neurons, converge onto different enhancer elements in the col gene. Hence, the col Tv1 neuron enhancer is triggered by Antp, hth, exd, lbe and cas, while the dAp enhancer is triggered by Kr, pdm and grn. In contrast to this subset-specific enhancer set-up for col activation, the subsequent, col-driven Nplp1 FFL feeds onto common enhancers in each downstream gene. These findings reveal that distinct spatio-temporal cues, acting in different neural progenitors, can trigger the same FFL by converging on discrete enhancer elements in an initiator terminal selector, to thereby dictate the same ultimate neuronal subtype cell fate (Stratmann, 2017).

This study has been able to molecularly decode the Tv1/dAp genetic FFL cascades, bolstering evidence for a complex molecular FFL, based upon sequential transcription factor binding to the downstream genes. The NB4-3 and NB5-6T neuroblasts are born in different regions of the VNC, and express different spatial determinants i.e., Antp, Lbe, Hth, Exd and Gr. As lineage progression commences, they undergo a programmed cascade of transcription factor expression; the temporal cascade. Early temporal factors Kr and Pdm integrate with Grn in NB4-3, while the late temporal factor Cas integrates with Antp, Lbe, Hth and Exd in NB5-6T, to create two distinct combinatorial spatio-temporal codes. These two codes converge on two different enhancers in the col gene, triggering Col expression, and hence the Nplp1 FFL. The FFL, in this case a so-called coherent FFL, where regulators act positively at one or several steps of a cascade, was first identified in E.coli and yeast regulatory networks, but have also been identified in C.elegans and Drosophila. Coherent FFLs can act as regulatory timing devices, exemplified by the action of col in NB5-6T: The initial expression of col in Ap cluster cells triggers a generic Ap/Eya interneuron fate in all four cells, while its downregulation in Tv2-4 and maintenance in Tv1 helps propagate the FFL leading to Nplp1 expression (Stratmann, 2017).

This study has found that the two different spatio-temporal programs converge on col, but on different enhancer elements. However, neither enhancer element gave complete null effects when deleted. Specifically, the 6.3kb col-Tv-CRM shows robust reporter expression, overlaps with endogenous col expression, responds to the upstream mutants, and is affected by TFBS mutations. However, when deleted (generating the col^ΔTv-CRM mutant), it had weak effects upon endogenous col expression in NB5-6T, and no effect upon Eya and Nplp1 expression. Deletion of the col-dAp-CRM (generating the col^ΔdAp-CRM mutant), gave more robust effects with reduction of Col, Eya and Nplp1 in dAp cells, although the expression was not lost completely (Stratmann, 2017).

Early developmental genes, which often are dynamically expressed, may be controlled by multiple enhancer modules, to thereby ensure robust onset of gene expression. This has been reported previously in studies of early mesodermal and neuro-ectodermal development, in which several genes i.e., twist, sog, snail are controlled by multiple distal enhancer fragments, so called 'shadow enhancers', in order to ensure reliable onset of gene expression. The shadow enhancer principle is also supported by recent findings on the Kr gene. Moreover, extensive CRM transgenic analysis, scoring thousands of fragments in transgenic flies, has also supported the shadow enhancer idea, revealing that a number of early regulators, several of which encode for transcription factors, indeed have shadow enhancers. The dichotomy between the col transgenic reporter results and the partial impact on col expression upon deletion of its Tv1 and dAp enhancers, gives reason to speculate that col may be under control of additional enhancers, some of which may be referred to as shadow enhancers (Stratmann, 2017).

The results on the eya, ap, dimm and Nplp1 enhancer mutants stand in stark contrast to the col CRMs findings. For these four genes, the enhancer deletion resulted in robust, near null effects, on expression. It is tempting to speculate that the current findings, combined with previous studies, points to a different logic for early regulators, with highly dynamic patterns, requiring several functionally overlapping enhancers for fidelity, and late regulators and terminal differentiation genes, which may operate with one enhancer that is inactive until the pertinent combinatorial TF codes have been established (Stratmann, 2017).

Analysis of the ap and eya enhancers indicates that Col directly interacts with these enhancers. Both of these enhancer-reporter transgenes are affected in col mutants, and can be activated by ectopic col. Moreover, mutation of one Col binding site in the ap enhancer and two sites in the eya enhancer, was enough to dramatically reduce enhancer activity. Direct action of Col on ap and eya is furthermore supported by recent data on Col genome-wide binding, using ChIP, which demonstrated direct binding of Col to these regions of ap and eya in the embryo. The regulation of ap is an excellent example of the complexity of gene regulation, and studies have identified additional enhancers controlling ap expression in the wing, muscle and brain (Stratmann, 2017).

In contrast to regulation of ap and eya, a direct action of Col on dimm and Nplp1 is less clear. Analysis of the dimm and Nplp1 enhancers did not reveal perfectly conserved Col binding sites. Mutation of multiple non-perfect Col binding sites in the dimm enhancer did not affect reporter expression in the Ap cluster, but did however reduce levels in the dorsal Ap cells. Mutation of non-perfect Col binding sites in the Nplp1 enhancer had no impact on enhancer activity, neither in Tv1 nor dAp. These findings support a model where Col is crucial for directly activating ap and eya, which in turn directly activate dimm and Nplp1, with some involvement of Col on dimm. However, support for a direct role for Col on Nplp1 comes from RNAi studies in larvae or adult flies, showing that knockdown of col resulted in loss of Nplp1, while Ap, Eya and Dimm expression was unaffected (Stratmann, 2017).

It is tempting to speculate that Col regulates Nplp1 not via direct interaction with its enhancer, but rather as a chromatin state modulator, keeping the chromatin around the Nplp1 locus in an accessible state, in order for Dimm, Ap and Eya to be able to access the Nplp1 gene. Support for this notion comes from studies on the mammalian Col orthologue EBF, which is connected to the chromatin remodeling complex SWI/SNF during EBF-mediated gene regulation in lymphocytes. Moreover, the central SWI/SNF component Brahma was recently identified in a genetic screen for Ap cluster neurons, and found to affect FMRFa neuropeptide expression in Tv4 without affecting Eya expression, indicating a late role in Ap cluster differentiation. Alternatively, Col may activate Nplp1 via unidentified, low affinity sites, similar to the mechanism by which Ubx regulates some of its embryonic target genes (Stratmann, 2017).

ap encodes a LIM-HD protein, a family of transcription factors well known to control multiple aspects of terminal neuronal subtype fate, including neurotransmitter identity, axon pathfinding and ion channel expression. The current results indicate that Ap in turn acts upon dimm, and subsequently with Dimm on Nplp1. eya encodes an evolutionary well-conserved phosphatase and does not bind DNA directly, instead acting as a transcriptional co-factor. Eya (and its orthologues) have been found to interact with several transcription factors in different systems, but whether it forms complexes with Col and Ap is not known (Stratmann, 2017).

The final transcription factor in the FFL is Dimm, a bHLH protein. Dimm is selectively expressed by the majority of neuropeptide neurons in Drosophila, and is important for expression of many neuropeptides. Intriguingly, Dimm is also both necessary and sufficient to establish the dense-core secretory machinery, found in neuropeptide neurons. Based upon these findings Dimm has been viewed as a cell type selector gene, acting to up-regulate the secretory machinery. This study found evidence for that Dimm acts directly on the Nplp1 enhancer, and this raises the possibility that Dimm is both a selector gene for the dense-core secretory machinery, and can act in some neuropeptide neurons to directly regulate specific neuropeptide gene expression (Stratmann, 2017).

An important, but unaddressed question is whether temporal information that diversifies neuronal progeny within a single lineage also impacts circuit assembly. Circuits in the sensorimotor system (e.g., spinal cord) are thought to be assembled sequentially, making this an ideal brain region for investigating the circuit-level impact of temporal patterning within a lineage. This study used intersectional genetics, optogenetics, high-throughput behavioral analysis, single-neuron labeling, connectomics, and calcium imaging to determine how a set of bona fide lineage-related interneurons in the ventral cord contribute to sensorimotor circuitry in the Drosophila larva. Even-skipped lateral interneurons (ELs) are sensory processing interneurons. Late-born ELs contribute to a proprioceptive body posture circuit, whereas early-born ELs contribute to a mechanosensitive escape circuit. These data support a model in which a single neuronal stem cell can produce a large number of interneurons with similar functional capacity that are distributed into different circuits based on birth timing. In summary, these data establish a link between temporal specification of neuronal identity and circuit assembly at the single-cell level (Wreden, 2017).

This study took advantage of the extremely well-characterized neuronal stem cells (neuroblasts) and their lineages in the Drosophila larval nerve cord to study lineage-circuitry relationships in a sensorimotor system. The Drosophila larval nerve cord is subdivided into a series of bilaterally symmetric segments, each of which contains 30 pairs of neuroblasts that give rise to all nerve cord neurons. This study focused on one class of bona fide sibling neurons-Even-skipped (Eve)-expressing interneurons with lateral cell body positions (ELs), a morphologically diverse set of excitatory interneurons from Neuroblast 3-3 (NB3-3) (Wreden, 2017).

The first abdominal segment consists of left/right clusters of ten ELs that can be subdivided into two groups based on the expression of the enhancer 'R11F02'. R11F02 expresses in the lateral-most ELs and a few other cells. During neurogenesis, newly born neurons displace their older siblings away from the parent neuroblast-generating an early-to-late, medial-to-lateral spatial pattern. Thus, it was hypothesized that R11F02(+) ELs were late born. Using a panel of transcription factors to assess birth order, R11F02(+) ELs were found to express the late-born marker Nab, but not early-born markers Kruppel/Pdm2. Thus, expression of R11F02 subdivides ELs into early-born and late-born temporal cohorts. However, the functional significance of this subdivision is unknown (Wreden, 2017).

To investigate early-born and late-born ELs, lines were used that specifically target each temporal cohort of neurons. For this study, R11F02-GAL80 were generated, that, when used with EL-GAL4, allows GAL4 to remain functional only in early-born, medial ELs. In addition, the split GAL4 lines R11F02-DBD and EL-AD was used to generate functional GAL4 in late-born, lateral ELs. Thus, the activity of each temporal cohort can be selectively manipulated (Wreden, 2017).

The behavioral response to acute stimulation of late-born ELs was examined. Previous work (Heckscher, 2015) showed that chronic stimulation of R11F02(+) ELs caused larvae to crawl with abnormal left/right body posture, but that study did not monitor initial responses of larvae to activation. Thus, it was unknown to what extent acute activation of late-born ELs slows larval motion, as would be expected if late-born ELs process proprioceptive information. In this study a behavior rig was build to monitor behavior before, during, and after optogenetic stimulation. Larval speed was measured by calculating the distance traveled by the larval centroid over time without regard to whether the direction of movement aligned with the body axis. Immediately upon stimulation of late-born ELs, body movements became left/right uncoordinated and speed was significantly reduced. Thus, the normal activity of late-born ELs is required for normal crawling, consistent with the idea that late-born ELs process proprioceptive information (Wreden, 2017).

It was asked whether stimulation of early-born and late-born ELs elicit similar or distinctive behavioral responses. Surprisingly, during optogenetic stimulation of early-born ELs speed transiently increased. Furthermore, all ELs were simultaneously stimulated and a transient increase was found followed by a sustained reduction in speed, which extended a previous finding that measured the later, but not initial, responses of larvae to activation of all ELs. Thus, it is likely that late-born and early-born ELs operate in distinct circuits (Wreden, 2017).

Increases in speed upon stimulation of early-born ELs could be due either to faster crawling or to larvae initiating a distinct movement-escape rolling. Escape rolling is the fastest larval movement and can be identified both because trachea on the dorsal side of the larva disappear beneath the body and because the direction of movement is lateral to the body axis. Higher resolution imaging showed that stimulation of early-born ELs frequently elicited multiple rolls, whereas stimulation of late-born ELs rarely elicited rolling. It was asked whether stimulation of early-born ELs triggered other escape-related behaviors -- hunching, fast crawling, reversals, body bending -- and an increase in body bending was found. Thus, activation of early-born ELs robustly triggers some, but not all, escape-related behaviors (Wreden, 2017).

Next, it was asked whether any early-born EL could be part of an escape circuit. Recently, an escape circuit has been characterized, which contains a set of roll-inducing 'Basin' interneurons (Ohyama, 2015). Furthermore, the neurons downstream of Basins have been identified in a transmission electron microscopic (TEM) volume that contains the entire larval CNS. In the current study it was asked whether any neurons that receive synapses from Basins are early-born ELs. Single-cell clones were generated of early-born ELs, and single-neuron morphology was imaged with fluorescent microscopy. Then, collection of early-born EL morphologies, as determined by light microscopy, was compared to the morphologies of neurons downstream of Basins, as determined by TEM. Three early-born ELs were found receive inputs from Basins. Thus, early-born ELs contribute functionally, and anatomically, to an escape circuit (Wreden, 2017).

This is the first time that single-cell morphology and connectivity have been identified for a majority of lineage-related interneurons within a Drosophila larval segment. This was accomplished by determining the spatiotemporal origin of neurons that were recently annotated in a Drosophila larval brain TEM volume. Within each segment, each neuroblast gives rise to a unique set of neurons, so this study asked what features are shared among ELs because these features are excellent candidates to be encoded at the stem cell level. First, late-born ELs contribute to a proprioceptive processing circuit. All TEM-annotated, late-born ELs in segment A1 receive direct synaptic input from proprioceptors, and some also receive direct synaptic inputs from Jaam interneurons, which themselves receive a large amount of direct proprioceptive input (Heckscher, 2015). Late-born ELs and Jaams get little input from other sensory neurons (Heckscher, 2015). Second, early-born ELs contribute to a mechanosensitive circuit. The TEM-annotated, early-born ELs in segment A1 receive direct synaptic input from mechanosensitive chordotonal sensory neurons and receive direct synaptic input from Basins 1 and 3, which themselves receive a large amount of direct mechanosensory chordotonal input. These early-born ELs, Basin1, and Basin 3 receive little known input from other sensory neurons. Currently, the inputs on to the remaining early-born ELs in segment A1 are unknown. Nonetheless, a majority of ELs in segment A1 are first-order sensory processing interneurons, directly receiving sensory neuron input, and many ELs are second-order sensory processing interneurons, indirectly receiving sensory neuron input. Notably, ELs are largely silent in the absence of sensory input and are therefore likely to encode sensory information. Taken together, these functional and anatomical data suggest that NB3-3 produces many sensory processing neurons (Wreden, 2017).

Anatomically, early-born ELs receive synapses from mechanosensitive, chordotonal sensory neurons (Mechano CHOs), whereas late-born ELs receive synapses from other sensory neurons. Thus, early-born versus late-born ELs are likely to process different stimuli. This study tested this idea by monitoring EL responses to a sound that activates chordotonals (Wreden, 2017).

First, it was shown that sound/vibration stimulus specifically activates chordotonals. In response to sound/vibration, Drosophila larvae perform an avoidance hunch. Hunching can be identified because larvae rapidly reduce crawling speed and shorten their body. A new sound stimulus was generated using a composite of known stimuli. To validate the stimulus, the behavioral rig was adapted by adding a speaker and amplifier, the stimulus was played to larvae, and speed and body perimeter was measured over time. In response to stimulation, control larvae robustly hunched, whereas larvae lacking chordotonals did not hunch. Thus, the sound/vibration stimulus can be sensed by larvae, and the response depends on mechanosensitive chordotonal sensory neurons (Wreden, 2017).

Next, it was asked to what extent do chordotonals and early-born and late-born ELs respond to sound/vibration. A previously described, head-fixed preparation was adapted, in which the anterior portion of the larva that contains the CNS is flattened and nearly immobilized, but the posterior is untouched. Calcium imaging monitored neuronal activity before, during, and after stimulation, and ΔF/F measured fluorescence intensity. As expected, chordotonals robustly responded to stimulation. Early-born ELs responded to stimulation with a smaller amplitude, but with a similar percentage responding in comparison to chordotonals. In contrast, late-born ELs showed little to no response. Thus, early-born versus late-born ELs differentially respond to sensory input. Furthermore, these data strongly suggest that the chordotonal-to-early-born EL connections seen in the TEM are functional, present in multiple larvae, and present in multiple segments along the anterior-posterior axis of the nerve cord (Wreden, 2017).

This work contributes an additional concept, showing that within a temporal cohort interneurons are similar. Furthermore, in the Drosophila motor system there may be many temporal cohorts-for example, Basins, Jaams, as well as another group of neurons that contact ELs, Saaghis, may be temporal cohorts. These interneurons are morphologically similar to each other, and Basins have been explicitly hypothesized to be lineage related. Thus, the observed link between temporal patterning and functional circuit assembly may be representative of a widely occurring phenomenon (Wreden, 2017).

How lineages contribute to neuronal circuits has been investigated in a few brain regions, none of which are sensorimotor. These studies have demonstrated that different brain regions have different lineage-circuitry relationships, which are likely to be critical for establishing region-specific functional differences. Sensorimotor systems perform a unique series of computations-sensing multiple kinds of stimuli, such as self-movement or pain, and producing adaptive motor outputs, such as locomotion or escape. This study used the Drosophila sensorimotor system to show that late-born, lineage-related ELs contribute to a proprioceptive circuit and that early-born, lineage-related ELs contribute to a mechanosensitive circuit. In both circuits, ELs are sensory processing interneurons. Thus, it appears that the NB3-3 lineage endows ELs with the capacity to processes sensory information regardless of circuit identity, and birth-timing segregates ELs into different circuits. Assembling circuitry according to these rules elucidates the developmental mechanisms that generate sensorimotor systems with the ability to process different types of sensory information in parallel (Wreden, 2017).

Depending on context, Basin interneurons can promote multiple types of escape responses, such as rolling, hunching, and bending. Some, but not all, of these escape behaviors occur upon stimulation of early-born ELs, which are downstream of Basins. These findings raise the questions: do other members of the NB3-3 lineage promote other escape responses? Do early-born neurons from other lineages promote other escape responses? Addressing these questions will be important for the field (Wreden, 2017).

The data support the idea that a developmental strategy for assembling sensorimotor circuits is as follows: a given neuronal stem cell can produce many neurons with similar functional capacity, which are distributed into different circuits based on birth timing. This developmental strategy may be used in other sensorimotor systems. Although the exact lineage-circuit relationship is unclear, the mammalian spinal cord provides additional examples of temporal cohorts of developmentally related neurons performing analogous functions in different circuits. For example, Renshaw cells and Ia interneurons are sequentially produced by p1 progenitors. Renshaw cells contribute to a motor neuron feedback circuit, whereas Ia interneurons contribute to a stretch reflex circuit. Despite participation in distinct circuits, Renshaw cells and Ia interneurons perform analogous functions-directly synapsing onto motor neurons and terminating firing. In addition, for extensor and flexor premotor interneurons, many of which originate from the same progenitor domain, time of neurogenesis is correlated with spatial, and inferred functional, segregation. Thus, temporal segregation of lineage-related neurons with similar functional capacities is likely to occur in evolutionarily distant species, suggesting the fundamental importance of this developmental strategy (Wreden, 2017).

Furthermore, the data reveal a correspondence between vertebrate and Drosophila sensorimotor development. In zebrafish, early-born neurons contribute to circuits for fast escape, whereas later-born neurons contribute to circuits for refined movements. These observations led the hypothesis-circuits for fast/gross movements and neurons in these circuits develop early, whereas circuits for slow/refined movements and neurons in these circuits develop later. However, it is unclear how broadly this hypothesis applies. This study shows that similar to zebrafish, in Drosophila, early-born neurons contribute to circuits for fast escape, whereas later-born neurons contribute to circuits for proprioceptive refinement of movements. Thus, this developmental principle guiding sensorimotor circuit assembly may be conserved across species despite separation by hundreds of millions of years of evolution (Wreden, 2017).

Takagi, S., Cocanougher, B. T., Niki, S., Miyamoto, D., Kohsaka, H., Kazama, H., Fetter, R. D., Truman, J. W., Zlatic, M., Cardona, A. and Nose, A. (2017). Divergent Connectivity of Homologous Command-like Neurons Mediates Segment-Specific Touch Responses in Drosophila. Neuron 96(6): 1373-1387 e1376. PubMed ID: 29198754 Animals adaptively respond to a tactile stimulus by choosing an ethologically relevant behavior depending on the location of the stimuli. This study investigated how somatosensory inputs on different body segments are linked to distinct motor outputs in Drosophila larvae. Larvae escape by backward locomotion when touched on the head, while they crawl forward when touched on the tail. A class of segmentally repeated second-order somatosensory interneurons, that was named Wave, was identified whose activation in anterior and posterior segments elicit backward and forward locomotion, respectively. Anterior and posterior Wave neurons extend their dendrites in opposite directions to receive somatosensory inputs from the head and tail, respectively. Downstream of anterior Wave neurons, premotor circuits were identified including the neuron A03a5, which together with Wave, is necessary for the backward locomotion touch response. Thus, Wave neurons match their receptive field to appropriate motor programs by participating in different circuits in different segments (Takagi, 2017).

Drosophila larval locomotion was used as a model to elucidate the working principles of motor circuits. Larval locomotion is generated by rhythmic and sequential contractions of body-wall muscles from the posterior to anterior segments, which in turn are regulated by motor neurons present in the corresponding neuromeres. Motor neurons are known to receive both excitatory and inhibitory inputs, combined action of which likely regulates patterned motor activity during locomotion. Although recent studies identified candidate inhibitory premotor interneurons, the identity of premotor interneurons that provide excitatory drive to motor neurons during locomotion remains unknown. This study searched for and identified two putative excitatory premotor interneurons in this system, termed CLI1 and CLI2 (cholinergic lateral interneuron 1 and 2). These neurons were segmentally arrayed and activated sequentially from the posterior to anterior segments during peristalsis. Consistent with their being excitatory premotor interneurons, the CLIs formed (GFP Reconstruction Across Synaptic Partners) (GRASP)- and ChAT-positive putative synapses with motoneurons and were active just prior to motoneuronal firing in each segment. Moreover, local activation of CLI1s induced contraction of muscles in the corresponding body segments. Taken together, these results suggest that the CLIs directly activate motoneurons sequentially along the segments during larval locomotion (Hasegawa, 2016).

Animals perform various types of rhythmic movements such as respiration, chewing and locomotion for their survival. These rhythmic movements are thought to be regulated by neuronal circuits termed central pattern generators (CPGs). CPGs consist of interneurons and motoneurons whose rhythmic activities induce coordinated patterns of muscle contraction. Although CPGs are regulated by descending and sensory inputs, rhythms very similar to those seen in the intact animal can be generated without these inputs. Because CPGs of invertebrates and vertebrates share many characteristics, CPGs in one animal could be a model for other animals. Moreover, because CPGs show many characteristics common to other neuronal systems, CPGs could be a general model linking neuronal circuits to behaviour. Despite the efforts to elucidate the function of CPGs, their identities and functional mechanisms are not completely understood, in particular in animals with a large central nervous system (CNS). This is partly because manipulating the function of specific neurons in the neural circuits is often difficult, especially in animals with vast numbers of neurons such as mammals (Hasegawa, 2016).

The Drosophila larva is emerging as an excellent model system for studies of CPGs because one can use sophisticated genetic methods, such as the Gal4-UAS system, to manipulate and visualize the activity of specific component neurons in a moderately sized CNS consisting of ~10,000 neurons. Larval forward locomotion is executed by the sequential contraction of muscles from the posterior to the anterior segments. Motoneurons in the ventral nerve cord (VNC) actualize the sequential muscle contraction by being activated from the posterior to the anterior segments during forward locomotion. CPGs responsible for the locomotion seem to be present in the VNC, since neuronal circuits in the thoracic and abdominal segments have been shown to be sufficient for generating the behavior. Calcium imaging of the entire CNS has visualized neurons that are active during larval locomotion including those in the brain, sub-oesophageal zone (SEZ), and the VNC16. However, the identities of these neurons are only beginning to be characterized (Hasegawa, 2016).

Previous studies showed that motor neurons in the VNC receive both excitatory and inhibitory inputs. It is therefore likely that specific patterns of motoneuron activation are regulated by the balance and the timing of excitatory and inhibitory inputs as shown in other systems. Recently, two types of inhibitory premotor interneurons that regulate larval locomotion have been identified. PMSIs (period-positive median segmental interneurons) are glutamatergic inhibitory premotor interneurons that regulate the speed of larval locomotion. Another glutamatergic interneuron, GVLIs (glutamatergic ventro-lateral interneurons) seem to function as premotor inhibitory neurons to terminate motor bursting. In contrast, premotor interneurons that provide excitatory inputs to motor neurons during locomotion remain to be identified, although they are known to be cholinergic. A recent study identified two cholinergic descending interneurons that form putative synaptic contacts with segmental motoneurons. However, whether they are active and play roles during locomotion remains unknown (Hasegawa, 2016).

This study sought and identified putative excitatory premotor interneurons that activate motoneurons during locomotion. These neurons, termed CLI1 and CLI2 (cholinergic lateral interneuron), are segmental interneurons that show wave-like activity during locomotion concurrent with the activity propagation of motoneurons. Consistent with CLIs being excitatory premotor neurons, these neurons form GRASP- and ChAT-positive synaptic contacts with motor neurons and are activated just before the activation of motoneurons in each segment. In addition, forced activation of these neurons locally induces the contraction of muscles. These results suggest that wave-like activity of CLIs activates motoneurons sequentially along the segments during forward locomotion (Hasegawa, 2016).

What are the circuit mechanisms that regulate Drosophila larval locomotion? To answer this question, it is necessary first to identify the neuronal components of the circuits. Excitatory inputs are critical for the generation of locomotor rhythms in various animals. However, identities and roles of excitatory interneurons that regulate Drosophila larval locomotion are unknown. The present study sought such excitatory interneurons using calcium imaging and identified CLI1s and CLI2s as candidate interneurons that excite motor neurons. Anatomical and behavioural studies suggest that these neurons directly activate motoneurons locally in each segment during larval locomotion (Hasegawa, 2016).

The following four lines of evidence suggest that CLIs are excitatory premotor interneurons: (i) CLIs are activated just before the activation of motoneurons in each segment during fictive locomotion, consistent with their providing excitatory drive to motoneurons. (ii) CLIs express ChAT, which synthesizes acetylcholine, a neurotransmitter known to excite motor neurons in this system. (iii) CLIs form GRASP-positive contacts with motoneurons. (iv) Local activation of CLIs results in the contraction of muscles in the corresponding body segments. Although these data are consistent with direct connection between CLIs and motoneurons, it remains possible that CLIs also excite motor neurons indirectly via other interneurons (Hasegawa, 2016).

CLI1s and CLI2s share many morphological and functional characteristics. i) They are neighboring neurons that send axons along a common path to reach the neuropile. This suggests that they are sibling neurons derived from the same neuroblast. Consistent with this notion, they also share the expression of R47E12-Gal4. ii) They both project axons along the same fascicle in the anterior commissure and locally innervate motor neurons in the contralateral side of the CNS. iii) They both are cholinergic premotor interneurons and are activated simultaneously during forward locomotion. iv) Activation of these neurons elicits muscle contraction. Taken together, these observations suggest that CLI1s and CLI2s belong to a class of interneurons that fulfill common function(s). There are also distinct features between these two neurons. i) CLI1s innervate the medial neuropile while CLI2s innervate a lateral region, suggesting that they target distinct neurons. ii) CLI1s but not CLI2s project to the next anterior segment. iii) CLI2s are active both during forward and backward locomotion, whereas CLI1s are active only during forward locomotion. Thus, CLI1s only participate in forward locomotion and may activate motor neurons not only in the same segment but also in the next anterior segment, and thus contribute to feed-forward propagation of motor excitation. In contrast, CLI2s may act locally to excite motoneurons only in the same segment and do so both during forward and backward locomotion (Hasegawa, 2016).

It is currently unknown what motor neurons are the targets of CLI1/2s. Dendrites of motoneurons that innervate different muscle domains form myotopic map along both antero-posterior and medio-lateral axes. The axon terminals of CLI1s are located in the medial neuropile, a region occupied by the dendrites of motoneurons innervating ventral muscles. Thus, CLI1s may form synaptic contacts with the ventral motoneurons. Similarly, candidate targets of CLI2s are dorsal motoneurons, since axon terminals of CLI2s are located in a lateral region occupied by these motoneurons. Consistent with this, it was observed that lifting of the tail, which is likely caused by dorsal muscle contraction when CLI2s but not CLI1s is activated. Moreover, CLI1s and CLI2s are activated at a similar timing as aCC in the same segment, a motor neuron that innervates a dorsal muscle and is activated simultaneously with other motor neurons innervating dorsal/ventral internal muscles. Future studies such as connectomic analyses using serial EM will determine more precisely the downstream circuits of the CLIs (Hasegawa, 2016).

It is also important to determine in the future the upstream circuits of the CLIs. Since dendritic region of CLI1s and CLI2s partially overlap, these neurons may share common upstream neurons. In particular, because the wave-like activity of CLIs was observed in the isolated CNS that receives no sensory inputs, the activity of CLIs must be regulated by the central circuits that generate a rhythm in an autonomous manner. However, it is also possible that CLIs are activated in response to specific sensory stimulation. Recently, neuronal circuits regulating larval behavior in response to specific sensory stimuli have been identified. It will be interesting to study the link between these circuits and CLIs (Hasegawa, 2016).

The wave-like activity of CLIs that occurs concomitant with motor activation strongly suggests that these neurons contribute to sequential activation of motor neurons along the segments during locomotion. Since these neurons are commissural neurons, they may also play a role in left-right coordination, as has been proposed for Dbx1-positive neurons in vertebrates and recently identified EL neurons in Drosophila. However, loss-of-function analyses thus far failed to reveal roles of CLIs in larval behaviors. Shibire^ts, tetanus toxin light chain, Kir2.1, hid and reaper, and ChAT-RNAi were used to inhibit the function of CLIs but no obvious phenotypes were observed. This could be due to insufficient silencing of these neurons by the activity manipulations. It could also be due to the redundancy in the circuit function. It should be noted in this regard that there are likely more CLIs-like neurons present in each segment. The axon terminals of CLI1 and CLI2 only cover part of the motor dendritic region, suggesting other neurons excite motor neurons not targeted by CLIs. Indeed, preliminary results obtained by the ongoing EM reconstruction of the larval CNS suggest that about 10 neurons, in the same neuroblast lineage as CLIs, send their axons locally and contralaterally to the motor region along the common path as CLI1s and CLI2s. It is likely that a group of CLIs-like neurons function in a similar manner and together excite the entire motor system. Unfortunately, direct testing of this possibility is not currently feasible due to the unavailability of Gal4 lines specific to this lineage (Hasegawa, 2016).

Recently, research has identified two classes of segmental premotor inhibitory interneurons PMSIs and GVLIs. These neurons are activated slightly later than the motor neurons and appear to inhibit the activity of motoneurons at distinct timings during the motor cycle: PMSIs at an early phase and GVLIs at a final phase of motoneuronal activation (Kohsaka, 2014; Itakura, 2015). This study identified CLIs that are activated prior to motor neurons and appear to provide an excitatory drive to the motoneurons. These three classes of premotor interneurons likely help shape the pattern of motor activity by providing excitatory and inhibitory inputs to motoneurons at distinct phases of the motor cycle. Since there are only ~400 interneurons per hemisegment in the larval ventral nerve cord, whose connectivity is being reconstructed, it is hoped that all major classes of premotor interneurons in this system will be identified in the near future. Systematic analyses of CLIs, PMSIs, GVLIs and other premotor neurons will elucidate how the motor patterns generating distinct behaviors are shaped by the combinatorial action of premotor interneurons (Hasegawa, 2016).

Drosophila larval crawling is an attractive system to study patterned motor output at the level of animal behavior. Larval crawling consists of waves of muscle contractions generating forward or reverse locomotion. In addition, larvae undergo additional behaviors including head casts, turning, and feeding. It is likely that some neurons are used in all these behaviors (e.g. motor neurons), but the identity (or even existence) of neurons dedicated to specific aspects of behavior is unclear. To identify neurons that regulate specific aspects of larval locomotion, a genetic screen was performed to identify neurons that, when activated, could elicit distinct motor programs. 165 Janelia CRM-Gal4 lines--chosen for sparse neuronal expression--were used to express the warmth-inducible neuronal activator TrpA1, and a screen was carried out for locomotor defects. The primary screen measured forward locomotion velocity, and 63 lines were identified that had locomotion velocities significantly slower than controls following TrpA1 activation (28 ° C). A secondary screen was performed on these lines, revealing multiple discrete behavioral phenotypes including slow forward locomotion, excessive reverse locomotion, excessive turning, excessive feeding, immobile, rigid paralysis, and delayed paralysis. While many of the Gal4 lines had motor, sensory, or muscle expression that may account for some or all of the phenotype, some lines showed specific expression in a sparse pattern of interneurons. These results show that distinct motor programs utilize distinct subsets of interneurons, and provide an entry point for characterizing interneurons governing different elements of the larval motor program (Clark, 2016).

Animals move by adaptively coordinating the sequential activation of muscles. The circuit mechanisms underlying coordinated locomotion are poorly understood. This study reports on a novel circuit for propagation of waves of muscle contraction, using the peristaltic locomotion of Drosophila larvae as a model system. An intersegmental chain of synaptically connected neurons, alternating excitatory and inhibitory, was found to be necessary for wave propagation and active in phase with the wave. The excitatory neurons (A27h) are premotor and necessary only for forward locomotion, and are modulated by stretch receptors and descending inputs. The inhibitory neurons (GDL) are necessary for both forward and backward locomotion, suggestive of different yet coupled central pattern generators, and its inhibition is necessary for wave propagation. The circuit structure and functional imaging indicated that the commands to contract one segment promote the relaxation of the next segment, revealing a mechanism for wave propagation in peristaltic locomotion (Fushiki, 2016).

This study discovered a circuit whose structure and function provides a mechanism for understanding forward wave propagation in peristaltic locomotion. This circuit consists of a chain of alternating excitatory and inhibitory neurons spanning all abdominal segments. The core elements of the chain include just one excitatory and one inhibitory neuron per hemisegment. The inhibitory neuron (GDL) is demonstrated to be sufficient to halt the peristalsis and to relax muscles in all segments, suggesting it is a point of coordination between forward and backward locomotion. It was further demonstrated that the excitatory neuron (A27h) is active during forward but not backward peristalsis, suggesting the existence of another excitatory circuit component critical for backward peristalsis among the synaptic partners of the GDL inhibitory neuron. This circuit defines a backbone of repeating, connected, modules for excitation and inhibition similar to those postulated in a computational model for peristalsis on the basis of behavioral observations that predicted the existence of central pattern generators (Fushiki, 2016).

This study found that the excitatory neuron (A27h) is premotor, directly synapsing onto motor neurons of its own segment only and that control both dorsal and ventral longitudinal muscles. This suggests an explanation for the observation that in forward crawling, dorsal and ventral longitudinal muscles contract simultaneously. In backward peristalsis, however, a phase gap has been observed in the timing of dorsal and ventral muscle contraction. This decoupling could require a more complex circuit structure for backward wave propagation, and therefore suggests an explanation for the lack of an equivalent excitatory neuron in the circuit chain for backward peristalsis. This study found, however, neurons postsynaptic to the inhibitory neuron (GDL) whose anatomy and position in the circuit suggest a role in backward peristalsis. In contrast, the inhibitory neuron (GDL) itself does not synapse onto motor neurons, and therefore occupies a higher-order position in the circuit that allows its participation in both forward and backward wave propagation in peristalsis. Furthermore, the GDL axon targets the intermediate lateral neuropil, which is neither in the domain of motor neuron dendrites nor in the somatosensory domain, suggestive of a role higher-order motor coordination. Relevant for forward peristalsis, GDL disinhibits the excitation of its anterior homologs, by removing inhibition from a glutamatergic interneurons (A02j) implicated in the regulation of peristaltic speed (one of the PMSIs). A02j is presynaptic to GDLs in anterior segments (Fushiki, 2016).

A model of peristaltic locomotion must consider the coordination of left and right hemisegments. Though this study found that the chain of alternating inhibitory and excitatory neurons runs independently on the left and right sides of the body, the excitatory neuron (A27h) presents a bilateral arbor and drives motor neurons bilaterally. The wiring diagram best supports a model of left-right coordination where excitatory neurons communicate with each other, but with the caveat that this synergy takes place by the simultaneous co-activation of the target motor neurons rather than reciprocal excitation. This model has been shown to support longer contraction episodes at the front of the wave, consistent with observations of muscle contraction in peristalsis. Independently of the timing, the fine-tuning in the intensity of left-right contractions has been shown to be under control of Even-skipped+ evolutionarily conserved neurons, which integrate both proprioceptive inputs and motor commands (Fushiki, 2016).

The dissected larval CNS undergoes spontaneous waves of motor neuron activation at about 1/10th the normal speed. These waves occur in the absence of sensory feedback, indicating the presence of CPGs and also suggesting a role for sensory feedback in speeding up the peristaltic wave. The circuit chain of excitatory and inhibitory neurons described in this study could be a part of the CPG, and this study additionally found these neurons are modulated by proprioceptive inputs (from vpda class I dendritic arborization neuron). Given that the vpda is a stretch receptor, it would be active in the segment ahead of the wave of contraction, which is being stretched by the pull exerted by the contracting segment. Proprioceptive feedback action onto the excitatory neuron of the circuit chain could then have two simultaneous effects: promotion of the contraction in the segment ahead of the wave (via activation of A27h), and relaxation of the segment twice removed (via activation of GDL, which acts on the segment anterior to it). Two somatosensory neurons (vdaA and vdaC) were found to synapse axo-dendritically onto the premotor excitatory neuron (A27h) and axo-axonically onto the inhibitory neuron (GDL) in their own segment. Although the function of these two sensory neurons remains unclear, it is speculated that this axo-axonic, likely depolarizing, connection onto GDL reduces the membrane action potential of its axon, reducing synaptic release of GABA onto A27h in the same segment. This model refines a previous model where the proprioceptive feedback was thought to signal the successful contraction of a segment. It is suggested that, in addition, at least some of the proprioceptive feedback (vpda) facilitates wave propagation and, therefore, may underlie the reduction in speed observed in fictive crawling (Fushiki, 2016).

In addition to the excitatory premotor interneuron A27h, this study found two other interneurons that receive direct synaptic inputs from a GDL (A02d and A08e3) and that, like A27h, also integrate inputs from stretch receptors (vpda, dbd and vbd). One interneuron (A08e3) is an Even-Skipped+ neuron that maintains left-right symmetric muscle contraction amplitude. The other (A02d) is a glutamatergic interneuron that belongs to a lineage of neurons thought to mediate speed of locomotion (one of the PMSIs). While A02d is a segment-local interneuron, proprioceptive axons span multiple segments, suggesting that a GDL can suppresses the effect of proprioceptive feedback specifically within its own segment without affecting the relay of proprioception to adjacent segments. Furthermore, A02d synapses onto a glutamatergic interneuron (A08a) thought to contribute to muscle relaxation in the wake of the peristaltic wave, which could be mediated via putative GABAergic premotor neurons (A31d). Taken together, it is suggested that one of the functions of the inhibitory neuron GDL is to gate proprioceptive feedback within its segment which has implications for the control of both speed and posture (Fushiki, 2016).

Finally, a descending neuron from the SEZ was observed that synapses onto the excitatory neuron (A27h) of the circuit chain in all segments. This motif has been observed and modeled in the leech and crayfish, where it enables the modulation of wave propagation speed. The brain and SEZ have been deemed non-essential for wave propagation. Speed of wave propagation, therefore, may be controlled in at least two ways: by proprioceptive feedback and by descending inputs. The existence of a circuit chain formed by excitatory and inhibitory neurons might be all that remains when both sensory feedback and the brain are absent, explaining the existence of wave propagation in decerebrated animals, and even for a small set of isolated abdominal segments (Fushiki, 2016).

Even a simple sensory stimulus can elicit distinct innate behaviors and sequences. During sensorimotor decisions, competitive interactions among neurons that promote distinct behaviors must ensure the selection and maintenance of one behavior, while suppressing others. The circuit implementation of these competitive interactions is still an open question. By combining comprehensive electron microscopy reconstruction of inhibitory interneuron networks, modeling, electrophysiology, and behavioral studies, this study determined the circuit mechanisms that contribute to the Drosophila larval sensorimotor decision to startle, explore, or perform a sequence of the two in response to a mechanosensory stimulus. Together, these studies reveal that, early in sensory processing, (1) reciprocally connected feedforward inhibitory interneurons implement behavioral choice, (2) local feedback disinhibition provides positive feedback that consolidates and maintains the chosen behavior, and (3) lateral disinhibition promotes sequence transitions. The combination of these interconnected circuit motifs can implement both behavior selection and the serial organization of behaviors into a sequence (Jovanic, 2016).

Animals move by adaptively coordinating the sequential activation of muscles. The circuit mechanisms underlying coordinated locomotion are poorly understood. This paper reports on a novel circuit for the propagation of waves of muscle contraction, using the peristaltic locomotion of Drosophila larvae as a model system. This study found an intersegmental chain of synaptically connected neurons, alternating excitatory and inhibitory, necessary for wave propagation and active in phase with the wave. The excitatory neurons (A27h) are premotor and necessary only for forward locomotion, and are modulated by stretch receptors and descending inputs. The inhibitory interneurons (GDL) are necessary for both forward and backward locomotion, suggestive of different yet coupled central pattern generators, and its inhibition is necessary for wave propagation. The circuit structure and functional imaging indicated that the commands to contract one segment promote the relaxation of the next segment, revealing a mechanism for wave propagation in peristaltic locomotion (Fushiki, 2016).

This study has discovered a circuit whose structure and function provides a mechanism for understanding forward wave propagation in peristaltic locomotion. This circuit consists of a chain of alternating excitatory and inhibitory neurons spanning all abdominal segments. The core elements of the chain include just one excitatory and one inhibitory neuron per hemisegment. The inhibitory neuron (GDL) is sufficient to halt the peristalsis and to relax muscles in all segments, suggesting it is a point of coordination between forward and backward locomotion. It was further demonstrated that the excitatory neuron (A27h) is active during forward but not backward peristalsis, suggesting the existence of another excitatory circuit component critical for backward peristalsis among the synaptic partners of the GDL inhibitory neuron. This circuit defines a backbone of repeating, connected, modules for excitation and inhibition similar to those postulated in a computational model for peristalsis on the basis of behavioral observations that predicted the existence of central pattern generators (Fushiki, 2016).

The excitatory neuron (A27h) is premotor, directly synapsing onto motor neurons of its own segment only and controlling both dorsal and ventral longitudinal muscles. This suggests an explanation for the observation that in forward crawling, dorsal and ventral longitudinal muscles contract simultaneously. In backward peristalsis, however, a phase gap has been observed in the timing of dorsal and ventral muscle contraction. This decoupling could require a more complex circuit structure for backward wave propagation, and therefore suggests an explanation for the lack of an equivalent excitatory neuron in the circuit chain for backward peristalsis. Neurons postsynaptic to the inhibitory neuron (GDL) were found whose anatomy and position in the circuit suggest a role in backward peristalsis. In contrast, the inhibitory neuron (GDL) itself does not synapse onto motor neurons, and therefore occupies a higher-order position in the circuit that allows its participation in both forward and backward wave propagation in peristalsis. Furthermore, the GDL axon targets the intermediate lateral neuropil, which is neither in the domain of motor neuron dendrites nor in the somatosensory domain, suggestive of a role higher-order motor coordination. Relevant for forward peristalsis, GDL disinhibits the excitation of its anterior homologs, by removing inhibition from a glutamatergic interneurons (A02j) implicated in the regulation of peristaltic speed (one of the PMSIs). A02j is presynaptic to GDLs in anterior segments (Fushiki, 2016).

A model of peristaltic locomotion must consider the coordination of left and right hemisegments. Though this study found that the chain of alternating inhibitory and excitatory neurons runs independently on the left and right sides of the body, the excitatory neuron (A27h) presents a bilateral arbor and drives motor neurons bilaterally. The wiring diagram (see Summary of the GDL circuit) best supports a model of left-right coordination where excitatory neurons communicate with each other, but with the caveat that this synergy takes place by the simultaneous co-activation of the target motor neurons rather than reciprocal excitation. This model has been shown to support longer contraction episodes at the front of the wave, consistent with observations of muscle contraction in peristalsis. Independently of the timing, the fine-tuning in the intensity of left-right contractions has been shown to be under control of Even-skipped+ evolutionarily conserved neurons, which integrate both proprioceptive inputs and motor commands (Fushiki, 2016).

The dissected larval CNS undergoes spontaneous waves of motor neuron activation at about 1/10th the normal speed. These waves occur in the absence of sensory feedback, indicating the presence of CPGs and also suggesting a role for sensory feedback in speeding up the peristaltic wave. The circuit chain of excitatory and inhibitory neurons described in this study could be a part of the CPG, and it was additionally found that these neurons are modulated by proprioceptive inputs (from vpda class I dendritic arborization neuron). Given that the vpda is a stretch receptor, it would be active in the segment ahead of the wave of contraction, which is being stretched by the pull exerted by the contracting segment. Proprioceptive feedback action onto the excitatory neuron of the circuit chain could then have two simultaneous effects: promotion of the contraction in the segment ahead of the wave (via activation of A27h), and relaxation of the segment twice removed (via activation of GDL, which acts on the segment anterior to it). Two somatosensory neurons (vdaA and vdaC) synapse axo-dendritically onto the premotor excitatory neuron (A27h) and axo-axonically onto the inhibitory neuron (GDL) in their own segment. Although the function of these two sensory neurons remains unclear, it is speculated that this axo-axonic, likely depolarizing, connection onto GDL reduces the membrane action potential of its axon, reducing synaptic release of GABA onto A27h in the same segment. This model refines a previous model where the proprioceptive feedback was thought to signal the successful contraction of a segment. It is suggested that, in addition, at least some of the proprioceptive feedback (vpda) facilitates wave propagation and, therefore, may underlie the reduction in speed observed in fictive crawling (Fushiki, 2016).

In addition to the excitatory premotor interneuron A27h, this study found two other interneurons that receive direct synaptic inputs from a GDL (A02d and A08e3) and that, like A27h, also integrate inputs from stretch receptors (vpda, dbd and vbd). One interneuron (A08e3) is an Even-Skipped+ neuron that maintains left-right symmetric muscle contraction amplitude. The other (A02d) is a glutamatergic interneuron that belongs to a lineage of neurons thought to mediate speed of locomotion (one of the PMSIs). While A02d is a segment-local interneuron, proprioceptive axons span multiple segments, suggesting that a GDL can suppresses the effect of proprioceptive feedback specifically within its own segment without affecting the relay of proprioception to adjacent segments. Furthermore, A02d synapses onto a glutamatergic interneuron (A08a) thought to contribute to muscle relaxation in the wake of the peristaltic wave, which could be mediated via putative GABAergic premotor neurons. Taken together, it is suggested that one of the functions of the inhibitory neuron GDL is to gate proprioceptive feedback within its segment which has implications for the control of both speed and posture (Fushiki, 2016).

Finally, a descending neuron was observed from the SEZ that synapses onto the excitatory neuron (A27h) of the circuit chain in all segments. This motif has been observed and modeled in the leech and crayfish, where it enables the modulation of wave propagation speed. The brain and SEZ have been deemed non-essential for wave propagation. Speed of wave propagation, therefore, may be controlled in at least two ways: by proprioceptive feedback and by descending inputs. The existence of a circuit chain formed by excitatory and inhibitory neurons might be all that remains when both sensory feedback and the brain are absent, explaining the existence of wave propagation in decerebrated animals, and even for a small set of isolated abdominal segments (Fushiki, 2016).

Locomotor systems generate diverse motor patterns to produce the movements underlying behavior, requiring that motor neurons be recruited at various phases of the locomotor cycle. Reciprocal inhibition produces alternating motor patterns; however, the mechanisms that generate other phasic relationships between intrasegmental motor pools, all of the motor neurons that innervate single muscles, are unknown. This study investigated one such motor pattern in the Drosophila larva, using a multidisciplinary approach including electrophysiology and ssTEM-based circuit reconstruction. It was found that two motor pools that are sequentially recruited during locomotion have identical excitable properties. In contrast, they receive input from divergent premotor circuits. It was also found that this motor pattern is not orchestrated by differential excitatory input but by a GABAergic interneuron acting as a delay line to the later-recruited motor pool. These findings show how a motor pattern is generated as a function of the modular organization of locomotor networks through segregation of inhibition, a potentially general mechanism for sequential motor patterns (Zwart, 2016).

Bilaterally symmetric motor patterns--those in which left-right pairs of muscles contract synchronously and with equal amplitude (such as breathing, smiling, whisking, and locomotion)--are widespread throughout the animal kingdom. Yet, surprisingly little is known about the underlying neural circuits. A thermogenetic screen was performed to identify neurons required for bilaterally symmetric locomotion in Drosophila larvae, and the evolutionarily conserved Even-skipped(+) interneurons (Eve/Evx) were identified. Activation or ablation of Eve(+) interneurons disrupted bilaterally symmetric muscle contraction amplitude, without affecting the timing of motor output. Eve(+) interneurons are not rhythmically active and thus function independently of the locomotor CPG. GCaMP6 calcium imaging of Eve(+) interneurons in freely moving larvae showed left-right asymmetric activation that correlated with larval behavior. TEM reconstruction of Eve(+) interneuron inputs and outputs showed that the Eve(+) interneurons are at the core of a sensorimotor circuit capable of detecting and modifying body wall muscle contraction (Heckscher, 2015).

This study identified an anatomical sensorimotor circuit containing an evolutionarily-conserved population of Eve/Evx+ interneurons that is required to maintain left-right symmetric muscle contraction amplitude both during active muscle contraction and at rest. These interneurons are the first known to regulate bilaterally symmetric muscle contraction amplitude. In mouse, Sim1+ V3 interneurons have a related function during alternating gait. In the future, it will be interesting to directly examine muscle contraction amplitude in 'V3 defective' mice to determine whether this class of interneuron is responsible for balancing amplitude of left-right muscle contraction during alternating motor patterns. Similarly, it will be interesting to determine the role of Drosophila interneurons expressing the Sim1 homolog, Single-minded, during left-right symmetric motor output (Heckscher, 2015).

EL interneurons act in a sensorimotor circuit independent of the central pattern generator that generates locomotion. First, in the absence of sensory input ELs do not show locomotion-like patterns of activity. Second, EL perturbation does not alter left-right timing of muscle contraction. Third, EL perturbation alters muscle contraction amplitude during locomotion and at rest (Heckscher, 2015).

The data suggest that EL interneurons receive sensory input that is primarily proprioceptive. Because proprioceptive neurons can detect muscle length and movement, they are well suited to convey muscle amplitude information to the ELs. Closer inspection of the proprioceptor to EL connectivity generates interesting hypotheses. First, proprioceptors are presynaptic to both projection and local EL interneurons; the former may send body posture information to the brain, while the latter may act locally to maintain left-right symmetric muscle length in each segment. Second, the Jaam interneurons are well positioned to process sensory information (e.g. from dorsal or ventral regions of the body wall) prior to transmitting information to the ELs. Although little is known about Jaam neurotransmitter expression or function, their position in the circuit raises the question of whether EL interneurons show state-dependent responses to proprioceptive inputs (Heckscher, 2015).

The data demonstrate that EL interneurons are presynaptic to motor neurons and can modify motor output. EL perturbation results in slow crawling and asymmetric left-right muscle contraction amplitude, while optogenetic stimulation of ELs induces motor neuron activity. The majority of ELs are cholinergic and likely excitatory, they provide direct input to contralateral motor neurons, and motor neurons are glutamatergic and excitatory. Thus, EL activity on one side of the body should result in increased contralateral motor neuron activity and contralateral muscle contraction. This may be reinforced by the disynaptic (EL-SA-MN) pathway, in which EL activity would prevent ipsilateral motor neuron activity if the SA neurons are inhibitory. This model awaits future characterization of SA neurotransmitter expression and function. It is proposed that ipsilateral muscle relaxation (via the EL-SA-MN pathway) together with contralateral muscle contraction (via the direct EL-MN pathway) is used for dynamic adjustment of body posture (Heckscher, 2015).

How do EL interneurons maintain left-right symmetric muscle contraction amplitude? Left-right differences in muscle contraction amplitude inevitably arise due to stochastic external (environmental) or internal (CNS/muscle) asymmetries. Without proper compensation, these perturbations would result in mismatched muscle contraction amplitude on left-right sides of the body. It is hypothesized that sensory input generates a representation of body wall curvature that is delivered to the EL interneurons. Left-right interactions among ELs would allow them to compare left versus the right sides of the body, followed by EL stimulation of motor output to restore left-right symmetric muscle length (Heckscher, 2015).

How does EL interneuron ablation and activation generate the same phenotype? A model is favored in which ELs are part of a 'perturbation-compensation' circuit. A larva that experiences an asymmetrical perturbation from an external or internal source would generate left-right mismatched muscle contraction amplitudes in the absence of any compensation. It is proposed that the EL circuit detects and compensates for these asymmetries. When the ELs are absent or constitutively active, they lose the ability to perform the left-right comparison and the asymmetries persist. In this way two 'opposite' manipulations yield the 'same' phenotype (Heckscher, 2015).

There is deep conservation of genetic programs that specify neuronal fate. This is particularly true for the Even-skipped+ (Eve or Evx+ in vertebrates) interneurons, which have been found in all bilateral animals examined to date except C. elegans. Annelids, chordates, insects, fish, birds, and mammals—as well as the presumed last common ancestor between invertebrates and vertebrates, Platynereis dumerilii —all contain Eve/Evx+ interneurons. Evx+ neurons in mice are commissural, excitatory, and directly contact motor neurons; this study shows that fly Eve+ interneurons are commissural, likely excitatory, and directly contact motor neurons. One hypothesis to explain the remarkable parallels between Eve/Evx+ interneurons is that the last common ancestor between vertebrates and invertebrates was segmented and motile; and thus the genetic programs used to create locomotor circuitry may be evolutionarily ancient (Heckscher, 2015).

This study has shown that the Drosophila Eve+ lateral interneurons are required to maintain left-right symmetrical motor output in the larva. Do Evx+ interneurons have a similar function in other organisms? Genetic removal of Evx1+ interneurons in mice did not reveal any specific function in either gross motor patterns or in the timing of left-right alternating motor neuronal activity as assayed by nerve root recordings. Subsequently, a broader genetic manipulation which reduced the number of Evx1+ interneurons to 25% of wild type levels, as well as ablating a large but unspecified number of Evx1− neurons, resulted in a hind limb hopping phenotype during fast locomotion. This study raised the possibility that Evx1+ interneurons regulate locomotion in mice. In this study it was shown that highly specific ablation or activation of Eve+ lateral interneurons disrupts larval crawling. It will be interesting to determine whether Evx1+ interneurons regulate bilaterally symmetric or alternating gait in other organisms, as well as whether Eve+ interneurons regulate alternating gait or symmetric flight in adult flies (Heckscher, 2015).

Much of the understanding of how neural networks develop is based on studies of sensory systems, revealing often highly stereotyped patterns of connections, particularly as these diverge from the presynaptic terminals of sensory neurons. Considerably less is known about the wiring strategies of motor networks, where connections converge onto the dendrites of motoneurons. This study investigated patterns of synaptic connections between identified motoneurons with sensory neurons and interneurons in the motor network of the Drosophila larva and how these change as it develops. As animals grow, motoneurons were found to increase the number of synapses with existing presynaptic partners. Different motoneurons form characteristic cell-type-specific patterns of connections. At the same time, there is considerable variability in the number of synapses formed on motoneuron dendrites, which contrasts with the stereotypy reported for presynaptic terminals of sensory neurons. Where two motoneurons of the same cell type contact a common interneuron partner, each postsynaptic cell can arrive at a different connectivity outcome. Experimentally changing the positioning of motoneuron dendrites shows that the geography of dendritic arbors in relation to presynaptic partner terminals is an important determinant in shaping patterns of connectivity. It is concluded that in the Drosophila larval motor network, the sets of connections that form between identified neurons manifest an unexpected level of variability. Synapse number and the likelihood of forming connections appear to be regulated on a cell-by-cell basis, determined primarily by the postsynaptic dendrites of motoneuron terminals (Couton, 2015).

Much of the current view of how sets of synaptic connections form and change during nervous system development is derived from studies of sensory systems. The connections that sensory neurons form are often tightly constrained, enabling the formation of accurate sensory maps, with numbers and distributions of synapses appropriate for network operation. Connectivity at lower-order synapses of the network can be almost invariant and cell autonomously specified. For example, Drosophila photoreceptor neurons reproducibly form ~50 synapses with specific postsynaptic lamina cells, irrespective of photoreceptor function or visual system defects. At higher-order synapses, in contrast, connectivity can be rather variable, reflecting both experience-dependent plasticity and distinct wiring strategies. For example, randomized connections in the mushroom body are thought to maximize coding space (Couton, 2015).

This study focused on the much less well-explored development of connectivity within a motor network. Motor systems manifest a great deal of flexibility, including their ability to adjust to changes in muscle size with growth and exercise, thus maintaining the capacity to trigger effective muscle contractions. This has been most extensively studied at the neuromuscular junction where the growth of the presynaptic terminal is matched with that of the postsynaptic muscle, regulated by muscle-derived retrograde signals. In addition, motoneurons also adjust centrally through changes in the size and connectivity of their dendritic arbors (Couton, 2015).

To investigate patterns of connectivity in a motor network and how these change as the animal develops and grows, this study used the Drosophila larva as a model. A paradigm was developed for studying identified partner neurons at the level of individual synaptic sites across different developmental stages. The following questions were asked: (1) How does connectivity change as the motor network develops? (2) How reproducible or variable are the sets of connections that form? (3) Is there evidence of synaptic patterning information residing with the presynaptic or postsynaptic partner? This study shows that from hatching to later larval stages, existing connections are progressively consolidated by addition of synapses. While patterns of connections are specific to each motoneuron type, considerable variability remains. Moreover, connectivity appears to be set on a cell-by-cell basis by the dendritic arbors of motoneurons, and dendritic positioning is a determinant of the connections that motoneurons make. Together, these findings argue in favor of a flexible regulation of connectivity in the assembly of the larval crawling circuit (Couton, 2015).

To study the emergence of synaptic connectivity in a motor network as it develops, genetic tools were developed for reliably visualizing and manipulating identified, connecting neurons in the Drosophila larval nerve cord. For pre-motor partner neurons, an intersectional 'split-Gal4' enhancer trap screen was fractionated through the set of cholinergic interneurons and sensory neurons, which provide the synaptic drive to motoneurons in this system. From >3,000 lines, those with sparse expression and terminations in the motor neuropile were identified. Single motoneurons ('aCC' and 'RP2') were visualized via a LexA/LexAOp and FLP recombinase-based quaternary system (Singh, 2013). To resolve synaptic sites, the presynaptic active zone marker UAS-brp::mRFP was combined with the GFP reconstitution across synaptic partners (GRASP)-based reporter for cell-cell contacts. Brp::mRFP-positive presynaptic specializations that coincide with physical appositions of presynaptic and postsynaptic membranes, as reported by GRASP, were scored as putative synapses. Thus patterns of connectivity during larval development, from 0 hr after larval hatching (ALH) to the third instar stage (48 hr ALH), were charted between the aCC and RP2 motoneurons and some of their presynaptic partners, made accessible to analysis by the Split-Gal4 line BF29^VP16.AD: two intersegmental descending interneurons and the ddaD and ddaE proprioceptive sensory neurons (Couton, 2015).

Focus was placed on the lateral interneuron (IN_lateral) within the BF29^VP16.AD expression pattern; its axon descends contralaterally from the sub-esophageal ganglion to segment A8 and forms putative en passant synapses with intersegmental nerve motoneurons. In mid-abdominal segments (A2-A6), the number of putative synaptic connections between this IN_lateral and the RP2 motoneuron increases steadily with developmental time from an average of 0.86 ± 0.26 at 0 hr ALH to 6.73 ± 0.78 at 24 hr ALH to 11.09 ± 0.97 at 48 hr ALH. This developmental increase in synapse number is compatible with electrophysiological recordings from these motoneurons. IN_lateral axons also form putative synapses with the two dendritic sub-arbors of the aCC motoneuron. The larger ipsilateral arbor, located on the same side as the aCC soma, receives more putative synapses from the IN_lateral than the smaller sub-arbor on the contralateral side. Both RP2 and aCC project to dorsal body wall muscles. To extend these observations to motoneurons that innervate ventral muscles, RP3 motoneurons were manually labeled with the lipophilic tracer dye DiD, and co-localization with IN_lateral Brp::mRFP sites were charted as putative connections. Here, too, it was found that the number of putative connections between this pair of neurons increases with developmental time, from 1 synapse (±0, n = 3) at 0 hr ALH to an average of 3.6 synapses (±0.4, n = 5) at 24 hr ALH (Couton, 2015).

Cell-type-specific differences in connectivity were documented. These are most evident in the likelihood with which the RP2 and aCC motoneurons receive putative synapses from the ddaD and ddaE sensory terminals (the high density of Brp::mRFP puncta in these sensory terminals prevents resolution of individual puncta). As larvae develop, this sensory-motor connection becomes increasingly frequent, although throughout aCC, motoneurons have a significantly lower probability than RP2 of forming putative synapses with these dda sensory terminals. In addition, it was found that motoneurons such as RP3, which are similar in operation to RP2 and aCC, i.e., in innervating longitudinal body wall muscles, also form putative synapses with the presynaptic IN_lateral, while motoneurons innervating antagonistic transverse muscles do not, even though their dendrites arborize within reach of the IN_lateral axon. For another pre-motor interneuron, INBF59, labeled with the BF59VP16.AD expression line, single cells were resolved by injecting INBF59 interneurons expressing UAS-brp::mRFP with the lipophilic tracer dye, Neuro-DiO, and different motoneurons with the spectrally distinct DiD. Co-localization of these three markers (Neuro-DiO, Brp::mRFP, and DiD) was taken as indicative of a putative synapse. The data suggest that different motoneurons, projecting to dorsal (aCC, RP2), lateral (MN-LL1), and ventral (RP3) muscles, may have different likelihoods of contacting the INBF59 (Couton, 2015).

In summary, in this motor network, the number of putative synapses between partner neurons generally increases as the network matures and the animal grows. Different motoneurons have different likelihoods of forming synapses with the same sets of presynaptic sensory neurons. Such qualitative differences are suggestive of motoneuron-type-specific regulation of connectivity (Couton, 2015).

It was striking by how variable connectivity between identified neurons seemed to be. For example, the number of putative synapses between IN_lateral and RP2 motoneurons ranged from 0 to 3 at 0 hr ALH and 6 to 16 at 48 hr ALH. Similarly, for the sensory-motor connection, only a fraction of RP2 and aCC motoneurons receive putative synaptic contacts from dda sensory terminals. Here, differences in connectivity are mirrored by the diverse routes by which individual neurons attain their connections. For instance, aCC motoneurons form putative synaptic connections with dda sensory axon terminals in every possible way: with contralateral, ipsilateral, or both groups of sensory projections, established by different routes, with dendrites from the main arbor or the soma. This shows that postsynaptic dendritic arbors of motoneurons are quite flexible in how they attain connections with presynaptic terminals (Couton, 2015).

Next, causes for the variable connectivity were explored. There is no clear indication that the connectivity that was measured becomes progressively more reproducible as the network matures. It was then asked whether differences in segmental identity contributed to the variability that was seen. Regression analyses show no statistically significant link between the segmental identity of RP2 and aCC motoneurons and the number of putative synapses that these receive from the IN_lateral at 0 hr ALH, 24 hr ALH, or 48 hr ALH (Couton, 2015).

Next, the effects that local and global network adjustments might have on connectivity were considered. To this end, focus was placed on pairs of RP2 and aCC neurons located in the same nerve cord and connected to the same IN_lateral, and it was asked whether having a common presynaptic partner leads to more similar numbers of synapses formed with the same axon. it was found that RP2 and aCC motoneurons can vary substantially in the number of putative connections they receive from the same presynaptic partner. These data imply that local interactions between individual pairs of neurons, rather than global network effects, might determine the outcome of connectivity (Couton, 2015).

In summary, these observations suggest that variability in connectivity might be an inherent feature of this motor network, at least for the cells analyzed in this study (Couton, 2015).

Since synapses are the product of interactions between presynaptic and postsynaptic terminals, it was asked whether the variability that was observe arises from one or the other synaptic partner. Testing the potential for an instructive role by the presynaptic interneuron, it was asked whether there was any pattern to the distribution of presynaptic sites along the IN_lateral axon. Along the IN_lateral axon (segments A2 to A8), the number of presynaptic sites per neuron was found to be highly variable, ranging from 48 to 107 (85 ± 16.8, SD, n = 17). At the same time, the distribution of presynaptic sites and the spacing between these are indistinguishable from random. Thus, no evidence was seen of positional patterning of en passant presynaptic sites along IN_lateral axons, which has been observed in other systems (Couton, 2015).

It was then asked whether differences in presynapse number could explain the variability in connectivity between different IN_lateral-motoneuron pairs. To this end, each IN_lateral-motoneuron pair the number of putative synapses formed was correlated with the local density of 'available' presynaptic Brp::mRFP puncta located within the IN_lateral axon along the span of the motoneuron dendritic tree. No significant correlation was found. This suggests that, at least in this system, the density of available presynaptic sites is not predictive of how many synaptic connections are formed with the postsynaptic motoneuron. Instead, these data are compatible with a model where the postsynaptic dendritic arbor regulates the number of connections that it forms (Couton, 2015).

Next,the role of postsynaptic motoneuron dendrites in determining connectivity was investigated. Previously, it was shown that postsynaptic dendritic arbors regulate the number of inputs they receive by adjusting dendritic growth. In motor networks, dendritic positioning has been suggested to be important in determining partner choice. To investigate the role of dendritic arbor positioning in shaping connectivity, the medio-lateral territories of motoneuron dendrites was changed. Increasing dendritic sensitivity to the midline attractant Netrin, by targeted overexpression of the cognate receptor Frazzled/DCC, shifts RP2 dendrites from principally lateral to more medial neuropil regions. This shift leads to a reduction of laterally positioned dendrites, so that fewer are in proximity to the IN_lateral axon, and a concomitant increase of dendrites in the medial neuropil, which is innervated by another interneuron with a medial descending projection (IN_medial). As a result, the proportion of synapses between motoneurons and the IN_lateral is drastically reduced, whereas the proportion of synapses with the IN_medial is greatly increased, as compared to controls (Figure 5C; t test, p = 0.0005 and p = 0.0194 for RP2 and aCCi, respectively). Although these observations do not assay for changes in partner choice (RP2 and aCC receive connections from both IN_lateral and IN_medial), these findings are compatible with a model where connections in motor systems emerge, to an extent, as a consequence of geographical overlap between presynaptic and postsynaptic terminals (Couton, 2015).

In summary, the data point to the existence of mechanisms that allow postsynaptic neurons to determine in a cell-type-specific fashion the number of presynaptic synapses they accept. Clearly, geographical overlap between presynaptic and postsynaptic terminals is necessary for synaptic connections to form, and the experiments suggest that dendritic positioning mechanisms contribute to the emergence of connectivity (Couton, 2015).

There is currently no consensus among views on how patterns of connections develop in a motor network. On the one hand, a great deal of genetically encoded specificity is evident in parts of the mouse spinal cord. For example, group 1a afferents target motoneuron pools with accuracy, and their connectivity is buffered, so that normal information flow is largely maintained in the face of considerable disturbances. Precision of wiring is perhaps most explicit in the selective positioning of inhibitory synapses by the so-called GABA pre-interneurons onto terminals of proprioceptive 1a sensory afferents. This precise and apparently invariant wiring is mediated by the expression of at least two sets of complementary heterophilic transsynaptic cell adhesion molecules. Contrasting with this view are studies from Xenopus tadpoles, where two-electrode recordings unequivocally demonstrated a surprising lack of specificity in synaptic connections during early stages of motor network development. Modeling based on these observations further suggests that such rather non-specific wiring patterns are able to generate swimming like motor outputs and that those patterns of connectivity could be formed simply through geographical overlap of coarsely defined presynaptic and postsynaptic termination zones. A limitation in those studies is that they look at groups of similar cells; this has precluded detailed insights at the level of individual synapses over developmental time. This study worked with identified partner neurons and studied how synaptic patterns in a motor network change, as the animal develops and grows (Couton, 2015).

A striking observation from this study is that at the output face of the network, motoneurons increase synaptic contacts with existing presynaptic partners over time. This correlates with previous observations that synaptic drive also increases during this period of larval development, although there is as yet no physiological readout for the specific anatomical changes detailed in this study. For motoneurons, the observed strengthening of existing connections is likely an adaptive mechanism that maintains the ability to effectively depolarize muscles as they enlarge during development. Although it has not been possible to assay for addition of new presynaptic partners during development, this wiring strategy contrasts with those proposed for cortical neurons, where pyramidal cells are thought to maximize the diversity of presynaptic inputs while keeping synapse number with each partner at a minimum (Couton, 2015).

Remarkably, reproducible cell-cell interactions during nervous system development can be genetically encoded, and this has been most clearly demonstrated with identified nerve cells of invertebrates—from highly specific substrate choices during axon path finding to the selection of synaptic partners and the number of synapses formed. In the Drosophila larval motor system, it was found that different motoneuron types have characteristic patterns of connections. For example, the likelihood of forming connections with the proprioceptive dda sensory neurons differs between the RP2 and aCC motoneurons. Qualitative differences in the specificity of partner choice are also present in that the IN_lateral forms connections with motoneurons that innervate longitudinal body wall muscles (e.g., aCC, RP2, and RP3), but not with motoneurons thought to be antagonistic in operation, despite close proximity of their dendrites (Couton, 2015).

At the same time, this motor system also manifests a considerable degree of variability, both in the likelihood and the number of connections that form between motor and pre-motor interneurons. Although some connection patterns seem to become more reproducible during early phases of network maturation, such as those between the RP2 motoneuron and dda sensory terminals, by and large, the observations suggest that connectivity is inherently flexible and that it is the outcome of local cell-cell interactions, at least between most cells that were studied. For example, two identical motoneurons (in different neuromeres) contacting the same IN_lateral axon can form quite different numbers of putative connections with the same presynaptic cell. It is conceivable that these connections are variable because they are not critical to motor system operation, and it remains to be seen to what extent the observations of this study are representative of connectivity elsewhere in this network (Couton, 2015).

Where does the information that determines these connectivity outcomes reside? No correlation was found with segmental identity or evidence for presynaptic patterning information: the number of presynaptic release sites that any one IN_lateral makes varies considerably, both between and within animals (left versus right homolog), and their distribution along the axon appears to be random, yet fairly even, with similar numbers of presynaptic sites per neuromere. Most compatible with the current data is the notion that patterns of connectivity are predominantly determined by the postsynaptic dendrites of motoneurons (Couton, 2015).

It has been previously shown that motoneurons achieve a specific range of synaptic input by adjusting the growth of their dendritic arbors. These structural adjustments mirror and complement homeostatic changes of neuronal excitable properties. This study shows that different dendritic growth patterns lead to different connectivity outcomes. For example, aCC motoneurons are capable of initiating growth of dendritic branches from different parts of the cell, which can form connections with the ipsilateral and/or contralateral dda terminals, or neither. In an analogous situation, in the mouse retina, differences in dendritic growth lead to distinct connection patterns between different bipolar cells and presynaptic photoreceptor terminals. This study experimentally tested how dendritic positioning impacts connectivity. Changing the bias so that motoneurons preferentially elaborate their dendrites toward the ventral midline results in changes in connectivity, namely reductions in the proportion of synapses with the lateral IN_lateral and concomitant increases in connections with the medially located IN_medial axon. Although this experiment does not inform about partner choice, since both the IN_lateral and IN_medial are normally contacted by these motoneurons, it suggests that the number of connections is determined by the extent to which presynaptic and postsynaptic terminal arbors are targeted to common regions. These experiments in the Drosophila larva support observations and models on connectivity in the motor network of Xenopus tadpoles, which suggest that the connectivity matrix might be determined in considerable part by geographical overlap of coarsely defined presynaptic and postsynaptic territories. There is evidence that the conserved Slit-Robo and Netrin-Frazzled/DCC guidance cue systems define such territories for positioning axon tracts and regions of dendritic arborization in the CNS and that these can contribute to shaping synaptic connectivity. That said, it remains to be established how the promiscuity of connections apparent in early Xenopus tadpoles changes over developmental time and to what extent hardwired specificity is genetically encoded elsewhere in the Drosophila or indeed in other motor networks (Couton, 2015).

Rhythmic motor patterns underlying many types of locomotion are thought to be produced by central pattern generators (CPGs). This study used the motor circuitry underlying crawling in larval Drosophila as a model to try to understand how segmentally coordinated rhythmic motor patterns are generated. Whereas muscles, motoneurons and sensory neurons have been well investigated in this system, far less is known about the identities and function of interneurons. A recent study identified a class of glutamatergic premotor interneurons, PMSIs (period-positive median segmental interneurons), that regulate the speed of locomotion. This study reports on the identification of a distinct class of glutamatergic premotor interneurons called Glutamatergic Ventro-Lateral Interneurons (GVLIs). Calcium imaging was used to search for interneurons that show rhythmic activity, and GVLIs were identified as interneurons showing wave-like activity during peristalsis. Paired GVLIs were present in each abdominal segment A1-A7 and locally extended an axon towards a dorsal neuropile region, where they formed GRASP-positive putative synaptic contacts with motoneurons. The interneurons expressed vesicular glutamate transporter (vGluT) and thus likely secrete glutamate, a neurotransmitter known to inhibit motoneurons. These anatomical results suggest that GVLIs are premotor interneurons that locally inhibit motoneurons in the same segment. Consistent with this, optogenetic activation of GVLIs with the red-shifted channelrhodopsin, CsChrimson ceased ongoing peristalsis in crawling larvae. Simultaneous calcium imaging of the activity of GVLIs and motoneurons showed that GVLIs' wave-like activity lagged behind that of motoneurons by several segments. Thus, GVLIs are activated when the front of a forward motor wave reaches the second or third anterior segment. It is proposed that GVLIs are part of the feedback inhibition system that terminates motor activity once the front of the motor wave proceeds to anterior segments (Itakura, 2015).

The motoneurons involved in Drosophila larval peristaltic locomotion are known to be responsive to at least three neurotransmitters, excitatory acetylcholine and inhibitory GABA and glutamate. Therefore, motoneurons likely generate rhythmic motor outputs by integrating multiple inputs. In order to clarify how interneurons contribute to the generation of motoneuronal rhythmic activity, it is essential to identify premotor interneurons and determine how they control the activity of motoneurons. This study identified GVLIs as putative premotor interneurons in this system (Itakura, 2015).

Four lines of evidence suggest that GVLIs are inhibitory premotor interneurons. First, GVLIs express vGluT, a vesicular transporter of glutamate, and thus likely secrete glutamate, a neurotransmitter known to elicit inhibitory responses in motoneurons. Second, vGluT-positive GVLI axon terminals are present in the dorsal region of the neuropile in the vicinity of motoneurons' dendrites in the same segment. Third, GVLIs form GRASP-positive putative synaptic contacts with motoneurons, although uncertainty remains as to the identity of the target motoneurons. The contact sites express the presynaptic markers Synaptotagmin and vGluT and show robust increases in calcium concentration during peristaltic waves, strongly suggesting that they are presynaptic terminals. Fourth, optogenetic activation of GVLIs inhibited motor function. Activation of GVLIs in crawling larvae disrupted ongoing peristaltic waves. Local activation of GVLIs in dissected larvae halted peristaltic waves in the corresponding region in the body wall. These results are consistent with the idea that GVLIs send inhibitory inputs locally to motoneurons. Taken together, anatomical and functional analyses strongly suggest that GVLIs are premotor local interneurons that inhibit motoneurons in the same segment. It should be noted, however, that this study has not examined whether GVLIs form synaptic connections with interneurons. Thus, it remains possible that GVLIs innervate some interneurons in addition to motoneurons. It is also important to note that axon terminals of GVLIs cover only a small portion of the dendritic region of motoneurons and thus likely innervate only a small subset of motoneurons. Considering the strong effect of GVLIs activation, GVLIs may well inhibit a large number of motoneurons via other interneurons (Itakura, 2015).

In Drosophila, several glutamate receptors (GluR) have been identified, such as metabotropic GluRs (DmGluR), AMPA/kinate receptor homologues, N-methyl-D-aspartate (NMDA) receptor homologues [56], and glutamate-gated chloride channels (GluCl). Thus Glu can have various effects on postsynaptic cells depending on the receptors expressed. For instance, Glu causes excitatory junction currents (EJCs) when released at neuromuscular junction (NMJ) and induces hyperpolarizing responses in antennal lobe neurons. Glutamate application elicits inhibitory responses in larval motoneurons. The effect is blocked by the chloride channel blocker picrotoxin, suggesting the existence of GluCl on motoneurons. Thus it is most likely that GVLIs inhibit motoneurons via GluCl. It should be noted, however, that the inhibitory effects of glutamate via GluCl has only been examined in subsets of motoneurons. It should also be noted that GVLIs may secrete other neurotransmitters in addition to Glu and/or transmit information through gap junctions. Future identification of the postsynaptic partners of GVLIs and the receptors expressed on the cells will provide more information on how GVLIs regulate the activity of downstream motoneurons (Itakura, 2015).

This study used calcium imaging to characterize the activity of GVLIs and aCCs in T3-A7 segments and the activity timing relationships among them. During forward locomotor waves, GVLIs are activated at a similar timing as are aCC neurons in the second or third more anterior neuromeres and later than aCC neurons in the same segment. The phase delay between GVLI and aCC activity remained relatively constant over wide range of wave durations. The identity of the postsynaptic motoneuron(s) of GVLIs remains to be determined. However, the axon terminals of GVLIs are located in a neuropile region occupied by dendrites of motoneurons that innervate dorsal/ventral muscles and are activated at the same timing as aCCs. GVLIs therefore are likely to be activated with a delay of 2-3 segments to their target motoneurons. It should be noted, however, the delay would be shorter if the target motoneurons are those innervating lateral muscles since they are known to be activated later than those innervating ventral/dorsal muscles (Itakura, 2015).

By studying the activity of aCCs and GVLIs during peristalsis at varying speeds, this study showed that phase delays between the two neurons remain relatively constant over a range of wave durations as in many undulatory movements spanning multiple body segments. The current results conform to a previous study that showed phase constancy based on the observation of muscle movements. The phase representation of the activation of aCCs and GVLIs, consisting of composite data derived from multiple larvae undergoing peristalsis at different speeds, well recapitulated the sequential activation from posterior to anterior segments observed in a single larva. Thus, use of the phase representation is adequate in the analyses of neural activity in this system. The phase delay data indicates that GVLIs, like motoneurons, are regulated by intersegmental networks that maintain phase constancy over different speeds of peristalsis. Although GVLIs were activated at a similar time as aCCs in the second or third anterior neuromere, they were not active at exactly the same time as aCC neurons. This suggests that upstream partners of GVLIs are different from those of motoneurons (Itakura, 2015).

The onset and termination of muscle contraction must be finely regulated to generate efficient forward movement during larval locomotion. Excitatory and inhibitory premotor neurons active at distinct phases of larval locomotion are likely to be involved in this regulation. During forward locomotion, muscles in three or more segments are simultaneously contracted at a given time. This indicates that muscle activity is shut down when the front of a muscle contraction wave reaches the third or more anterior segment. The activity pattern of GVLIs revealed by calcium imaging (phasic activation with a two-to-three segment delay compared to aCC motoneurons) is consistent with a role for GVLIs in this process. The anatomy of GVLIs is also consistent with a role in feedback inhibition: each GVLIs extend their putative dendritic processes to anterior neuromeres and their axonal processes to motoneurons in the same segment. GVLIs may thus inhibit motoneurons and help to terminate muscle contraction when the motor wave reaches the anterior segments, by integrating information from anterior segments and transmitting the signal to motoneurons in the same segment. Whether GVLIs indeed play essential roles in this process remains to be determined since functional analyses with currently available neural silencers failed to show any obvious phenotypes. It should also be noted that if GVLIs do play such a role, they should only be part of the system since their axonal terminals do not cover the entire dendritic field of motoneurons and thus likely innervate only a subset of motoneurons (Itakura, 2015).

In an independent study, another class of premotor inhibitory neurons PMSIs (period-positive median segmental interneurons) were identified. Like GVLIs, PMSIs are glutamatergic and inhibit motor function when activated, and show wave-like activity during peristalsis. However, they are activated at a different phase from that of GVLIs. They are activated much earlier than GVLIs, shortly after the activation of the postsynaptic motoneurons with a time delay of ~0.5 neuromere, and control the duration of motor bursting and the speed of locomotion. Thus, PMSIs appear to provide early-cycle inhibition that is critical for determining the duration of motor bursting. In contrast, GVLIs may contribute to late-cycle inhibition that terminates motor bursting. Future studies will elucidate how GVLI, PMSI and other premotor interneurons, active at distinct phases of a motor cycle, shape the motor pattern. For example, optogenetic activation of the interneurons can be combined with patch-clamp recordings in motoneurons to study how the activity manipulation changes the pattern of motor activity. Such analyses will pave the way for understanding how rhythm is generated during larval locomotion (Itakura, 2015).

Natural events present multiple types of sensory cues, each detected by a specialized sensory modality. Combining information from several modalities is essential for the selection of appropriate actions. Key to understanding multimodal computations is determining the structural patterns of multimodal convergence and how these patterns contribute to behaviour. Modalities could converge early, late or at multiple levels in the sensory processing hierarchy. This study shows that combining mechanosensory and nociceptive cues synergistically enhances the selection of the fastest mode of escape locomotion in Drosophila larvae. In an electron microscopy volume that spans the entire insect nervous system, the multisensory circuit was reconstructed supporting the synergy and spanning multiple levels of the sensory processing hierarchy. The wiring diagram revealed a complex multilevel multimodal convergence architecture. Using behavioural and physiological studies, functionally connected circuit nodes were identified that trigger the fastest locomotor mode, and others were identified that facilitate it. Evidence is provided evidence that multiple levels of multimodal integration contribute to escape mode selection. It is proposed that the multilevel multimodal convergence architecture may be a general feature of multisensory circuits enabling complex input-output functions and selective tuning to ecologically relevant combinations of cues (Ohyama, 2015).

Different combinations of nociceptive and mechanosensory stimulation induced different likelihoods of the key escape sequences: rolling followed by fast crawling versus fast crawling alone. Nociceptor activation alone evoked a relatively low likelihood of rolling and a high likelihood of fast crawling. Vibration alone evoked only fast crawling and essentially no rolling. Combined with nociceptor activation, vibration increased the likelihood of rolling; the effect is dose-dependent and super-additive (synergistic). This vibration-induced facilitation of rolling is mediated through the mechanosensory chordotonal neurons (Ohyama, 2015).

It is suspected that the information from the two modalities converges onto central neurons involved in the selection of rolling. To identify such neurons and thus determine where in the sensory processing hierarchy multisensory convergence occurs, a was performed behavioural screen for neurons whose thermogenetic activation triggers rolling. A 'hit' was identified in the R72F11 Drosophila line, that drove GAL4 expression in neurons potentially early in the sensory processing hierarchy. Activating the neurons in R72F11 triggered rolling in a significant fraction of animals, and inhibiting them significantly decreased rolling in response to bimodal stimulation (Ohyama, 2015).

R72F11 drives expression selectively in four lineage-related, segmentally repeated projection neurons with basin-shaped arbors in the ventral, sensory domain of the nerve cord; therefore they were named Basins-1-4. The dendrites of Basin-1 and Basin-3 span a ventrolateral domain of the nerve cord, where the mechanosensory chordotonal terminals are located. The dendrites of Basin-2 and Basin-4 span both the ventrolateral chordotonal domain and a ventromedial domain where the nociceptive MD IV terminals are located. It was therefore asked whether the mechanosensory chordotonal and the nociceptive MD IV neurons directly converge on Basin-2 and Basin-4 (Ohyama, 2015).

In an electron microscopy volume that spans 1.5 nerve cord segments, the chordotonal and MD IV arbors were scanned. The left and right Basin-1, -2, -3 and -4 among the reconstructed neurons (Ohyama, 2015).

Basin-1 and Basin-3 received many inputs (each >25 synapses and >15% of total input, on average) from chordotonal neurons, but very few (no more than 1% of total input synapses) from MD IV neurons. Basin-2 and Basin-4 received many inputs from both chordotonal neurons (on average >20 synapses and >10% total input) and MD IV neurons (on average >20 synapses and >10% total input), each on distinct dendritic branchesl Of all the 301 partners downstream of MD IV and chordotonal neurons, only Basin-2 and Basin-4 reproducibly received >5 synapses from both chordotonal and MD IV neurons, suggesting that they are probably key integrators of chordotonal and MD IV inputs (Ohyama, 2015).

To investigate whether the observed anatomical inputs from the sensory neurons onto Basins were functional and excitatory, calcium transients were imaged in response to MD IV or chordotonal activation collectively in all Basins or in individual Basin types, using lines that drive expression selectively in Basin-1 or Basin-4 (Ohyama, 2015).

In Basin-1, calcium transients were observed in response to vibration, but not in response to MD IV activation, consistent with the large number of synapses it receives from chordotonal neurons and the relatively few from MD IV neurons. In Basin-4, calcium transients were observed in response to both vibration and MD IV activation, consistent with the large number of synapses it receives from both sensory types. Basin-4 integrated the inputs from the two modalities, responding significantly more to bimodal than to unimodal (Ohyama, 2015).

Next, it was asked whether the multisensory Basin-4 interneurons contribute to rolling selection. Silencing Basin-4 neurons significantly decreased rolling in response to bimodal stimulation, indicating these neurons are involved in triggering rolling. Selective activation of the multisensory Basin-4 interneurons triggered rolling in a dose-dependent way, with strongest activation triggering rolling in 45% of animals (Ohyama, 2015).

It was also asked whether a second level of multimodal integration (that is, integration of information from distinct Basin types, that receive distinct combinations of chordotonal and MD IV inputs), enhances the selection of rolling. Indeed, co-activation of Basin-1 with the bimodal Basin-4 facilitated rolling, resulting in a significantly higher likelihood of rolling compared to activation of Basin-4 alone (70% versus 45%). Thus, information from distinct Basin types may converge again onto downstream neurons involved in triggering rolling (Ohyama, 2015).

To identify potential sites of convergence of information from the different types of first-order Basin interneurons a 'hit' was examined from the thermogenetic activation screen, R69F06, a Drosophila line that drove GAL4 expression in neurons that project far from the early sensory processing centres. Thermogenetic activation of neurons in R69F06 triggered rolling in a high fraction of larvae, and inhibiting them significantly decreased rolling in response to bimodal stimulation (Ohyama, 2015).

R69F06 drives expression in a few neurons in the brain, in the sub-oesophageal zone (SEZ) and in a pair of thoracic neurons whose axons descend through the dorsal, motor domain of the nerve cord. Selectively activating the single pair of thoracic neurons triggered rolling in 76% of larvae. These command-like neurons were named Goro (a romanization of the Japanese for rolling) (Ohyama, 2015).

Activation of Basins evoked strong calcium transients in the Goro neurons, indicating that these cell types involved in the same behaviour are functionally connected. To identify the shortest anatomical pathways from Basins to Goro that might support the observed functional connectivity and to determine whether the information from distinct Basin types converges onto Goro, electron microscopy reconstruction was again used (Ohyama, 2015).

A second electron microscopy volume (from a second larva) was used that spans the entire larval nervous system and therefore also includes Goro neurons. In the new volume, chordotonal, MD IV and Basin neurons were reconstructed from segment A1, as well as the Goro neurons (Ohyama, 2015).

To find putative pathways from distinct Basin types to Goro neurons, all neurons downstream of all axonal outputs were reconstructed from the four left and right Basin homologues from segment A1. Thirty-one pairs of reproducible downstream partners were identified. Among these second-order nerve cord interneurons were identified that constitute the shortest pathways from Basins to Goro neurons (called A05q and A23g where 'A' stands for abdominal neuron). They receive inputs from distinct Basin types and synapse onto Goro neurons. Thus, information from distinct Basin types, that receive distinct combinations of MD IV and chordotonal inputs, converges onto Goro neurons -providing a second level of multimodal convergence (Ohyama, 2015).

Ten distinct second-order projection neuron types downstream of Basins ascend to the brain. Some of these integrate Basin information across multiple distal segments of the body, either exclusively from a single Basin type, or from distinct Basin types (that receive distinct combinations of sensory inputs; for example, A00c-a4 and A00c-a5). Then, distinct second-order PNs, that receive distinct combinations of Basin inputs (and therefore distinct combinations of mechanosensory and nociceptive inputs), re-converge again on third-order interneurons in the brain. Thus, following convergence of local mechanosensory and nociceptive information from a single segment onto multisensory Basins, global mechanosensory and multisensory information from multiple segments is integrated within the brain pathway (Ohyama, 2015).

By tracing upstream of Goro dendritic inputs, brain neurons were identified that send descending axons that synapse onto Goro neurons. Tracing downstream of a multisensory second-order ascending projection neuron (A00c-a4), third-order projection neurons were identified connecting the ascending pathways from Basins to a descending path onto Goro neurons. Thus, the activity of the command-like Goro neurons may be modulated by the more local multisensory and unisensory information via the nerve cord Basin-Goro pathway and by the global body-wide nociceptive and mechanosensory multisensory information via the brain Basin-Goro (Ohyama, 2015).

A third-order SEZ feedback neuron was identified that receives convergent body-wide mechanosensory and multisensory information and descends through the nerve cord sensory domain. The SEZ feedback neuron synapses onto the first-order (Basins) and second-order neurons from both the nerve cord (A05q) and brain (A00c) Basin-Goro pathways. Both the nerve cord and brain Basin-Goro pathways may therefore be jointly regulated based on integrated global multisensory information (Ohyama, 2015).

Next, the functional role of the nerve cord and brain Basin-Goro pathways was explored. Basin activation could activate Goro neurons in the absence of the brain, suggesting the nerve cord Basin-Goro pathway is excitatory and sufficient for activating Goro neurons and triggering rolling. Consistent with this idea, Basin activation evoked calcium transients in their nerve-cord targets, the A05q neurons, and A05q activation evoked calcium transients in Goro neurons. Furthermore, thermogenetic activation of the neurons in a line that drives expression, among others, in the A05q neurons triggered (Ohyama, 2015).

Calcium imaging in the terminals of three Basin-target neurons that ascend to the brain (A00c-a6, A00c-a5 and A00c-a4)) revealed that, collectively, they respond to vibration, to MD IV activation, and to Basin activation, suggesting that this connection is also excitatory (Ohyama, 2015).

Silencing the A00c neurons decreased rolling in response to bimodal stimulation, and their co-activation with Basin-4 facilitated rolling. Therefore, downstream from early local multisensory integration by Basin-4, additional levels of integration of global mechanosensory and multisensory information appear to further facilitate the transition to rolling behaviour (Ohyama, 2015).

The rolling response triggered by multisensory cues (or by strong nociceptive cues alone) is followed by fast crawling. Similarly, optogenetic activation of the first-order multisensory Basin-4 neurons triggered both locomotor modes; rolling followed by fast crawling. However, optogenetic activation of the Goro neurons triggered only rolling, but not fast crawling, suggesting that they act as dedicated command-like neurons for rolling. This also suggests that the act of rolling itself is insufficient to trigger fast crawling. In the future it will be interesting to determine how all of the 31 novel neuron types directly downstream of the Basins identified in the electron microscopy reconstruction contribute to the selection of the two locomotor modes, rolling and crawling, in a defined sequence (Ohyama, 2015).

By combining behavioural and physiological studies with large-scale electron microscopy reconstruction this study has mapped a multisensory circuit that mediates the selection of the fastest mode of escape locomotion (rolling) in Drosophila larva. Mechanosensory and nociceptive sensory neurons were found to converge on specific types of first-order multisensory interneurons that integrate their inputs. Then, interneurons that receive distinct combinations of mechanosensory and nociceptive inputs converge again at multiple levels downstream, all the way to command-like neurons in the nerve cord. Activating just a single type of first-order multisensory interneuron triggers rolling probabilistically. Co-activation of first-order interneurons that receive distinct combinations of mechanosensory and nociceptive inputs increases rolling probability. Thus, action selection starts at the first-order multisensory interneurons and multiple stages of multimodal integration in the distributed network enhance this selection (Ohyama, 2015).

Given that spurious firing from distinct sensors is uncorrelated, whereas event-derived signals will be temporally correlated across the sensory channels, multimodal integration even at a single level improves the signal-to-noise ratio. Multilevel multimodal integration can offer additional advantages. Theoretical studies show that a multilevel convergence architecture enables more complex input-output relationships. Similarly, the multilevel multimodal convergence architecture described in this study could offer better discrimination between different kinds of multisensory events. The weights in such networks could be tuned either through experience or through evolution to respond selectively to highly specific combinations of two cues. Using a simple model, it can be demonstrated that compared to early-convergence a multilevel architecture could specifically enhance the selection of the fastest escape mode in the most threatening situations, either in response to weak multimodal or strong unimodal nociceptive cues (Ohyama, 2015).

The multilevel multimodal convergence architecture may be a general feature in multisensory integration circuits, enabling complex response profiles tunable to specific ecological needs. For example, physiological studies in mammals have identified multisensory neurons that integrate the same cues at several stages in the sensory processing hierarchy, although it is unclear whether the multisensory neurons at distinct levels are causally related to the same behaviour. Due to the size of networks involved, synaptic-level resolution studies of the underlying convergence architecture across multiple levels were unattainable (Ohyama, 2015).

In addition to the multilevel multimodal feed-forward convergence motif, electron microscopy reconstruction revealed higher-order and local feedback neurons. Recent theoretical models of multisensory integration suggest that the output of individual multisensory neurons is normalized by the activity of other multisensory neurons in that population, but the anatomical implementation of such feedback has not been identified. Some of the feedback neurons in the multisensory circuit described in this study may have roles in such normalization computations (Ohyama, 2015).

Another circuit motif revealed by the study is the divergence of sensory information into nerve cord and ascending brain pathways and subsequent re-convergence of the shorter and the longer pathway onto the same command-like neurons in motor nerve cord (Goro). The nerve cord pathway integrates nociceptive and mechanosensory information from a local region of the body (few segments), whereas the ascending brain pathway integrates the information across all body segments and provides a means of modulating command-like neuron activity based on global body-wide nociceptive and mechanosensory information. The multisensory circuit described in this study in a genetically tractable model system provides a resource for investigating in detail how multiple brain and nerve cord pathways interact with each other and contribute to the selection of different modes of locomotion (rolling and crawling) in a defined sequence (Ohyama, 2015).

The electron microscopy volume spanning the entire insect nervous system acquired for this study can be used to map circuits that mediate many different behaviours. Combining information from a complete wiring diagram with functional studies has been very fruitful in the 302-neuron nervous system of C. elegans. Recently, similar approaches have been applied to microcircuits in smaller regions of larger nervous systems. This study has demonstrated that relating local and global structure to function in a complete nervous system is now possible for the larger and more elaborate nervous system of an insect (Ohyama, 2015).

Animals control the speed of motion to meet behavioral demands. Yet, the underlying neuronal mechanisms remain poorly understood. This study shows that a class of segmentally arrayed local interneurons (period-positive median segmental interneurons, or PMSIs) regulates the speed of peristaltic locomotion in Drosophila larvae. PMSIs formed glutamatergic synapses on motor neurons and, when optogenetically activated, inhibited motor activity, indicating that they are inhibitory premotor interneurons. Calcium imaging showed that PMSIs are rhythmically active during peristalsis with a short time delay in relation to motor neurons. Optogenetic silencing of these neurons elongated the duration of motor bursting and greatly reduced the speed of larval locomotion. These results suggest that PMSIs control the speed of axial locomotion by limiting, via inhibition, the duration of motor outputs in each segment. Similar mechanisms are found in the regulation of mammalian limb locomotion, suggesting that common strategies may be used to control the speed of animal movements in a diversity of species (Kohsaka, 2014).

PMSIs are the first interneuronal population shown to be involved in Drosophila larval locomotion. Anatomical and functional analyses strongly suggest that PMSIs are premotor local interneurons that inhibit motor neurons in the same or a neighboring segment. Previous electrophysiological analyses showed that GABA or glutamate application elicits inhibitory responses in motor neurons that reverse at near resting potential and are blocked by the chloride channel blocker picrotoxin. Based on these observations, it has been suggested that motor neurons express Cl−-permeable GABA and glutamate receptors. Glutamate-gated inhibitory channels have been identified and well characterized in arthropods and other invertebrates including C. elegans. Although no such receptors are known in vertebrates, previous structural and pharmacological analyses suggest that invertebrate glutamate-gated chloride channels are orthologous to vertebrate glycine channels. Drosophila homologs of the receptors have been cloned and shown to produce a glutamate-gated chloride current when expressed in Xenopus oocytes and exhibit inhibitory action in Drosophila adult brain. Thus, it is likely that PMSIs inhibit motor neurons through glutamate-gated chloride channels. The motor neurons are also glutamatergic but send excitatory input to the muscles. Previous studies report that there are 40 putative vGluT-positive glutamatergic neurons in each hemisegment, of which 34 are motor neurons and six are interneurons. Since the number of PMSIs is comparable to that of the estimated glutamatagic interneurons, PMSIs most likely represent a majority of the glutamatergic interneurons in the ventral nerve cord (Kohsaka, 2014).

This study demonstrated that the duration of motor bursting and segmental muscle contraction is elongated when PMSIs are inhibited. The results indicate that PMSIs regulate the duration of motor output in each segment by terminating motor bursting. Consistent with this idea, dual-color Ca2+ imaging showed that activation of PMSIs is delayed with respect to that of the postsynaptic motor neurons. This temporal pattern allows PMSIs to regulate the time window of motor firing via inhibition. Thus, a main function of PMSIs seems to be to limit the duration of motor output (Kohsaka, 2014).

Similar roles in shaping motor outputs have been proposed for V1 neurons in mice and aIN neurons in Xenopus, both of which are inhibitory interneurons expressing Engrailed and have been proposed to share evolutionarily conserved roles. Loss or acute inactivation of V1 neurons elongates the duration of motor bursting during fictive locomotion in isolated mouse spinal cord. Xenopus aIN neurons provide early-cycle inhibition to motor neurons and other CPG interneurons during swimming. Thus, regulation by on-cycle inhibition seems to be a common mechanism for shaping the duration of motor outputs in vertebrates and in Drosophila larvae. Interestingly, PMSIs share several cellular properties with vertebrate V1 and aIN neurons. The three classes of neurons are all inhibitory premotor interneurons that are rhythmically activated during motor cycles. They are unipolar and send their axons first toward motor neurons and then extend an ascending ipsilateral axon longitudinally. Whereas V1 and aIN use glycine as the inhibitory neurotransmitter, PMSIs use glutamate, which is considered to be the invertebrate counterpart of glycine. These shared features may underlie the common function in motor control (Kohsaka, 2014).

Several mechanisms have been proposed for speed control of animal locomotion, including the recruitment of different motor neurons and change in electrophysiological properties of motor and other CPG neurons. The current results on PMSIs and previous studies on V1 and aIN neurons suggest that limiting the duration of motor firing by inhibition might be a phylogenetically conserved mechanism for speed control. In mice lacking V1 neurons, not only the duration of motor firing but also that of motor cycles is elongated, and thus the speed of locomotion is reduced. Although the role of aIN neurons in speed control has not been directly examined, close correlations have been observed between the activity of these neurons and the frequency of the tadpole swimming. This study demonstrates that blocking activities of PMSIs elongates the duration of motor bursting and reduces the speed of axial locomotion in Drosophila larvae. Taken together, these results suggest that evolutionarily distant organisms with anatomically and functionally distinct motor systems may adopt similar strategies for speed control of locomotion. It is important to note that both activation and inhibition of PMSIs activity lead to a decrease in locomotor speed (paralysis upon activation with ChR2 and slowed locomotion upon inhibition with Shits or NpHR). Thus, these neurons need to be activated at an optimum level and timing to output locomotion with appropriate speed (Kohsaka, 2014).

It still remains to be determined how the change in the duration of motor bursting affects the speed of locomotion. A simple model would be that since motor bursting in each segment is elongated in the absence of PMSI activity, it takes longer for the motor wave to propagate along the segments. In many undulatory movements, such as lamprey and leech swimming and Drosophila larval crawling, intersegmental phase lag (not intersegmental time lag) remains constant at different speeds. This is because the phase of muscle contraction in different segments must remain constant in order to maintain the same motor output pattern (e.g., forming approximately one full wave at a given time during lamprey swimming). Because of this intersegmental coordination, segmental lag of motor activity may have to be prolonged in the absence of PMSI activity to match up with the elongation of segmental motor bursting; otherwise, too many muscle segments would contract at the same time during peristalsis. Indeed, electrophysiological recordings showed that intersegmental time lag of motor firing was prolonged to a similar extent as the motor bursting (~2 fold) when PMSI activity was silenced. Likewise, in mice lacking V1 neurons, while the left-right and flexor-extensor coordination is maintained, both motor bursting and step cycles are elongated to a similar extent (2- to 3-fold). Thus, a common strategy, limiting the duration of motor bursting, may be used to regulate the speed of diverse animal locomotion such as larval locomotion and mammalian limb movements because it leads to changes in the most critical parameters of the speed, intersegmental time delay in axial locomotion, and left-right/flexor-extensor step cycle in limb locomotion. Understanding how intersegmental coordination is regulated in Drosophila larvae is an important future goal (Kohsaka, 2014).

It is also important to explore what might be the upstream neural circuits that activate PMSIs. Good candidates are multidendritic neurons, which are known to be required for fast larval locomotion and believed to feedback muscle contraction status. Another interesting possibility is that PMSIs control the speed of locomotion in response to environmental changes such as temperature or to meet internal demands such as hunger. Preliminary data using the GRASP technique suggest that PMSIs indeed receive afferent projections from sensory neurons. Once the upstream neurons are identified, the input-output relationship between these neurons and PMSIs can be systematically studied using optogenetics and other methods. It is anticipated that such analyses will not only clarify the roles of PMSIs in local neural circuits, but also shed light on conserved mechanisms by which inhibitory interneurons regulate animal locomotion (Kohsaka, 2014).

Most neurons of the adult Drosophila ventral nerve cord arise from a burst of neurogenesis during the third larval instar stage. Most of this growth occurs in thoracic neuromeres, which contain 25 individually identifiable postembryonic neuronal lineages. Initially, each lineage consists of two hemilineages--'A' (Notch(On)) and 'B' (Notch(Off))--that exhibit distinct axonal trajectories or fates. No reliable method presently exists to identify these lineages or hemilineages unambiguously other than labor-intensive lineage-tracing methods. By combining mosaic analysis with a repressible cell marker (MARCM) analysis with gene expression studies, a gene expression map was constructed that enables the rapid, unambiguous identification of 23 of the 25 postembryonic lineages based on the expression of 15 transcription factors. Pilot genetic studies reveal that these transcription factors regulate the specification and differentiation of postembryonic neurons: for example, Nkx6 is necessary and sufficient to direct axonal pathway selection in lineage 3. The gene expression map thus provides a descriptive foundation for the genetic and molecular dissection of adult-specific neurogenesis and identifies many transcription factors that are likely to regulate the development and differentiation of discrete subsets of postembryonic neurons (Lacin, 2014).

Understanding how cell-type diversity in nervous systems arises remains a key goal in developmental biology. Even simple nervous systems, such as those in insects, involve hundreds of different subtypes of cells. Over the last several decades, research using the Drosophila embryonic CNS as a model system has unveiled basic principles that underlie nervous system development in invertebrates and vertebrates. Drosophila and other holometabolous insects, however, undergo two distinct waves of neurogenesis: embryonic neurogenesis creates the larval nervous system; postembryonic neurogenesis creates the adult nervous system. Relative to embryonic neurogenesis, little is known about the genetic and molecular control of postembryonic neurogenesis (Lacin, 2014).

Within each hemisegment of the segmented embryonic nerve cord, 30 neuroblasts (NBs) divide in a stem cell manner to produce ~400 neurons and glia that interconnect to form a functional CNS. Towards the end of embryogenesis, NBs become quiescent or undergo apoptosis: in abdominal segments, most NBs die; in thoracic segments, 25 of the 30 NBs become quiescent and persist into larval stages. This study focused on the postembryonic neuronal lineages produced by these 25 NBs. During the second larval-instar stage, in response to glia-derived insulin signaling, thoracic NBs regain their proliferative activity. Initially, NBs divide slowly to produce a small number of large, Chinmo-positive (Chinmo+) neurons (termed early-born neurons). Shortly after larvae enter the third (last) instar stage, NBs divide more quickly and produce many, small Broad+ neurons (termed late-born neurons), ceasing their proliferation in the early pupal stage (Lacin, 2014).

Elegant mosaic analysis with a repressible cell marker (MARCM)-based lineage-tracing studies revealed that each neuronal lineage in the thoracic CNS is uniquely identifiable based on its relative position size and neuronal projection patterns. Each postembryonic NB, which resides in the ventral-most region of a lineage, divides in a stem cell manner to self-renew and produce a chain of secondary precursor cells, called ganglion mother cells (GMCs). Typically, each GMC divides to produce sibling post-mitotic neurons that adopt distinct fates based on the state of Notch signaling -- 'A' (NotchON), 'B' (NotchOFF). In contrast to the embryo, in which sequentially born 'A' (or 'B') daughter cells often adopt distinct identities, most A (or B) cells within a given postembryonic lineage manifest the same cellular phenotype, extending projections along a common path to a similar target region. Thus, initially, each postembryonic lineage consists of a NB and some GMCs in the ventral region of the clone and two major subtypes of neurons (A and B) more dorsally. In some lineages, most or all cells of the A (or B) hemilineage undergo apoptosis, resulting in a monotypic lineage that consists largely, if not exclusively, of cells from the A or B hemilineage (Lacin, 2014).

At present, the only reliable way to identify which lineage a group of postembryonic neurons belongs to is through labor-intensive MARCM-based lineage tracing methods. In the work reported in this study, by combining gene expression studies of 14 transcription factors with MARCM-based lineage tracing methods, a gene expression map was created that unambiguously identifies 23 of the 25 postembryonic neuronal lineages and 29 of the 34 major neuronal hemilineages (See: Schematic models of transcription factor expression in postembryonic neuronal lineages). Pilot functional studies reveal that the identified transcription factors direct the development and differentiation of the postembryonic neurons expressing them (Lacin, 2014).

By a ten-to-one ratio, postembryonic neurons outnumber embryonic neurons, and the adult fly CNS is composed almost entirely of postembryonic neurons. Yet much less is known about postembryonic neurogenesis than embryonic neurogenesis. The molecular marker map of postembryonic thoracic neuronal lineages presented in this study helps bridge this gap by extending the work of Truman who characterized these lineages on the basis of morphology. The map enables the identification of 23 of 25 postembryonic neuronal lineages based on gene expression alone, buttressing the descriptive foundation of postembryonic neurogenesis and illustrating that the combinatorial code of neuronal specification extends to the postembryonic thoracic CNS. The apparent lack of cell-type diversity in postembryonic neuronal lineages, the utility of the gene expression map in matching postembryonic lineages to their cognate embryonic lineages, and the similarity of hemilineages in flies to pools of neurons in vertebrates are discussed (Lacin, 2014).

Despite their larger size, postembryonic lineages appear less complex than their embryonic counterparts. For example, embryonic NB 3-1 produces four motor neurons and a variable number of intersegmental and local interneurons. Postembryonic lineage 4, which appears to derive from NB 3-1, generates almost 50 cells, but based on morphology and gene expression, neurons in this lineage can be grouped into at most two subtypes of neurons. Are thoracic postembryonic neuronal lineages less complex than their embryonic counterparts? The jury remains out. Studies of postembryonic neurogenesis have not reached the resolution of those in the embryo, and most have assessed postembryonic neuronal lineages at the end of larval life, when the vast majority of neurons have arisen, but still days away from their final differentiation. Thus, even though neurons in a given postembryonic lineage display simple gene expression profiles and extend axons along only one or two paths before metamorphosis, they may manifest complex patterns of target innervation and gene expression after metamorphosis. In this context, the molecular marker map is a key antecedent for studies that dissect the cellular and molecular complexity of postembryonic lineages at single-cell resolution at later stages of development (Lacin, 2014).

Within the thoracic nerve cord, postembryonic neurons derive from the same NBs that generate embryonic neurons. With few exceptions, it has proved difficult to pair postembryonic neuronal lineages with their cognate embryonic lineages and NBs. To do so would provide a continuum of knowledge from the genetic mechanisms that drive the NB formation and specification in the embryo to those that govern the development and differentiation of postembryonic neurons (Lacin, 2014).

In those cases in which the common ancestry of embryonic and postembryonic neuronal lineages is known, the two lineages share similar gene expression profiles. For example, postembryonic lineages 3 and 4 derive from embryonic NBs 7-1 and 3-1, respectively. In embryos, NB 7-1 expresses Nkx6 and generates Eve+ A-type motor neurons and Dbx+ B-type interneurons. In larvae, NB 7-1 continues to produce Dbx+ B-type interneurons and generates Nkx6+, rather than Eve+, A-type neurons. In embryos, NB 3-1 produces B-type Hb9+, Nkx6+, Lim3+ and Islet+ RP1, three to five motor neurons. In larvae, NB 3-1 produces B-type Hb9+ and Nkx6+, but Lim3- and Islet-, interneurons. Thus, lineally related embryonic and postembryonic neurons share similar gene expression profiles and are likely to share functional attributes (Lacin, 2014).

The shared gene expression profiles of lineally related postembryonic and embryonic neurons suggest the gene expression map will help match postembryonic lineages to their cognate embryonic lineages. For example, in embryos NB 2-2 generates six B-type neurons that express Hb9, Nkx6 and Lim3 (Lacin, 2009). In larvae, lineage 10, a monotypic B-type lineage, is the only postembryonic lineage that expresses this combination of transcription factors, and similar to the embryonic neurons produced by NB 2-2, lineage 10 neurons extend axons across the midline as part of the anterior commissure. A systematic pairing of embryonic and postembryonic lineages will still require sophisticated lineage tracing methods that induce clones in the early embryo and analyze the embryonic and postembryonic lineages of single NBs in the CNS of late third instar larvae. Here, the simultaneous use of molecular markers that identify defined embryonic and/or postembryonic neuronal lineages will enable the matching of individual embryonic and postembryonic lineages. Only through such studies will it be possible to follow CNS development uninterrupted from the embryo to the adult (Lacin, 2014).

The hemilineage has been identified as the developmental unit of the postembryonic CNS: most neurons within an individual hemilineage project axons within the same bundle to similar targets (Truman, 2010). This study extends these findings by showing that within a given lineage, most transcription factors are expressed in A- or B-type neurons, but not both. Thus, hemilineages are composed of tightly clustered groups of neurons that share common transcription factor expression profiles and extend axons in the same bundle to innervate similar targets (Lacin, 2014).

At the morphological and molecular level, neuronal hemilineages in flies resemble pools of neurons in vertebrates. Individual motor or inter-neuron pools are composed of clustered groups of neurons that share common transcription factor expression profiles and extend axons in the same bundle to innervate similar targets. For example, motor neurons with cell bodies located medially within the lateral motor column (LMC) express Islet and project axons to ventrally derived limb muscle; motor neurons with cell bodies located laterally in the LMC express Lim1 and project axons to dorsally derived limb muscles. These parallels between neuronal hemilineages in flies and pools of neurons in vertebrates suggest that individual pools of vertebrate neurons share a common lineage and state of Notch activation (Lacin, 2014).

Recent studies on Drosophila melanogaster and other insects have revealed important insights into the functions and evolution of neuropeptide signaling. In contrast, in- and output connections of insect peptidergic circuits are largely unexplored. Existing morphological descriptions typically do not determine the exact spatial location of peptidergic axonal pathways and arborizations within the neuropil, and do not identify peptidergic input and output compartments. Such information is however fundamental to screen for possible peptidergic network connections, a prerequisite to understand how the CNS controls the activity of peptidergic neurons at the synaptic level. This study provide a precise 3D morphological description of peptidergic neurons in the thoracic and abdominal neuromeres of the Drosophila larva based on fasciclin-2 (Fas2) immunopositive tracts as landmarks. Comparing the Fas2 "coordinates" of projections of sensory or other neurons with those of peptidergic neurons, it is possible to identify candidate input and output connections of specific peptidergic systems. These connections can subsequently be more rigorously tested. By immunolabeling and GAL4-directed expression of marker proteins, this study analyzed the projections and compartmentalization of neurons expressing 12 different peptide genes, encoding approximately 75% of the neuropeptides chemically identified within the Drosophila CNS. Results are assembled into standardized plates which provide a guide to identify candidate afferent or target neurons with overlapping projections. In general, this study found that putative dendritic compartments of peptidergic neurons are concentrated around the median Fas2 tracts and the terminal plexus. Putative peptide release sites in the ventral nerve cord were also more laterally situated. The results suggest that peptidergic neurons in the Drosophila ventral nerve cord have separated in- and output compartment. The lack of a strict segmentally reiterated pattern throughout the thoracic and abdominal neuromeres suggests that the restricted and differential distribution of peptidergic neurons reflects neuromere-specific functional connections. Other larval neuron types or circuits that match the observed peptidergic distribution patterns have not been characterized (Santos, 2007).

The last two abdominal neuromeres a8/9 have a unique pattern of peptidergic somata and projections (e.g. FMRFa, MIP or PDF neurons, and show the least serial homology to the more anterior neuromeres of the ventral ganglion. This finding also extends to descending processes. Descending axons may stop before or when reaching the border to a8 (HUG and DTK neurons), form extensive varicose ramifications within the neuropil of a8 (AST, corazonin or branch extensively in the terminal plexus of a9 (FMRFa-, leucokinin-, MIP and PDF-neurons. Belonging to the tail region, the segments a8/9 differ from the homomeric segments a1-7 with respect to the organization of muscles and sensory neurons. Furthermore, several unique structures such as the spiracles or the anal pads belong to these terminal segments. Unlike other segmental nerves, the segmental nerve of a9 innervates the hindgut musculature. The unique pattern of peptidergic neurons in a8/9 might thus, at least partially, reflect a segment-specific function related to e.g. control of spiracles or intestinal functions. For example, the PDF neurons innervate the hindgut, but their exact function is so far unknown. Similar segmental differences between a8 and the rest of the abdominal neuromeres have been found for neurons expressing biogenic amines (Santos, 2007).

The fusion construct syb.egfp has been developed as a presynaptic marker. Since synaptobrevin (SYB) is an integral membrane protein of small synaptic vesicles and large peptide-containing vesicles alike, SYB.EGFP also labels peptide vesicles and hence peptide accumulation and release sites (varicosities), which typically do not spatially coincide with synapses. Concomitantly, it is assumed that purely dendritic compartments of peptidergic neurons do not contain vesicles and show no or only weak SYB.EGFP labeling. These assumptions are supported by results obtained for PDF neurons in the brain, and the Tv and Va neurons that innervate neurohemal organs. SYB.EGFP was only found in the cell bodies (where the protein is made) and in the terminals in the neurohemal organs. The axonal projections as well as the arborizations within the VNC were unlabeled. Nevertheless, when interpreting the SYB.EGFP distribution, it has to be kept in mind that SYB.EGFP might also label presynaptic sites if the peptidergic neurons contain colocalized classical neurotransmitters (Santos, 2007).

The haemagglutinin-tagged GABAA receptor subunit RDL.HA has been shown to be a useful specific postsynaptic marker in motor neurons. Since The GABAA receptor subunit RDL is involved in mediating GABAergic postsynaptic currents, attempts were made to see whether ectopic RDL.HA expression indicates postsynaptic sites (dendrites) of peptidergic neurons also. The general expression level of RDL.HA was very weak, and only discernible labeling intensities were obtained with two different GAL4-drivers: Ccap- and c929-GAL4. Nevertheless, the labeling was spatially very confined to neuron compartments that showed no varicosities or only weak SYB.EGFP fluorescence. This suggests that RDL.HA labeled postsynaptic sites in peptidergic neurons (Santos, 2007).

Arborizations around the median DM and VM tracts turned out to be a prominent feature of most characterized peptidergic neurons with somata in the ventral ganglion, including the AST, CCAP, corazonin, FMRFa, MIP and PDF neurons. In contrast to motor neurons, the prominent midline arborizations of peptidergic neurons were rather short, and did not occupy large areas in the more lateral neuropils between the median and lateral tracts. For the CCAP neurons, ectopically expressed RDL.HA localized exclusively to these median arborizations. In contrast, SYB.EGFP as well as peptide-immunoreactivity was absent or relatively low in these arborizations. Also in the general peptidergic c929-GAL4-line, SYB.EGFP expression was low in the median compared to lateral fascicles. This might suggest that the median arborizations represent peptidergic dendrites. Descending processes of CCAP, EH, HUG and leucokinin neurons (originating from somata in the suboesophageal ganglion or in the brain) all have putative release sites around the DM and VM tracts. Of the peptidergic neurons with cell bodies in the VNC, only those expressing corazonin were found to have varicosities indicative of release sites around the DM and VM tracts (Santos, 2007).

Taken together, these findings suggest that the arborizations around the dorsomedial (DM) and ventromedial (VM) tracts are mainly input compartments for peptidergic VNC neurons, and point to this midline region as a main site for synaptic inputs onto peptidergic neurons including the CCAP neurons. The different putative sites of in- and outputs to peptidergic neurons in the VNC are summarized (see Assignment of putative main compartment identities as suggested by morphology, immunolabeling intensities and distribution of synaptic markers). Peptides released from varicosities of leucokinin, CCAP, HUG-, EH and corazonin neurites along the DM tract may modulate synaptic transmission around the DM tracts, or might represent direct input signals to peptidergic neurons. Also, the dorsal ap-let neurons with somata in the ventral ganglion expressing the peptide precursor Nplp1 appear to have their output sites along the DM tracts as indicated by strong peptide immunoreactivity. Unlike any of the peptidergic neurons characterized here, the dorsal ap-let neurons seem to have extensive arborizations within the neuropil of each hemineuromere, which appear to contain no or only little peptide immunoreactive material and hence might represent dendritic regions. Also the leucokinin neurons with somata in the ventral ganglion do not send projections towards the midline. Since leucokinin release is likely to occur at peripheral release sites on body wall muscles, it is possible that a synaptic input region is located along the VL tract, the only projection site of abdominal leucokinin neurons within the CNS neuropil (Santos, 2007).

For most model organisms in neuroscience, research into visual processing in the brain is difficult because of a lack of high-resolution maps that capture complex neuronal circuitry. The microinsect Megaphragma viggianii, because of its small size and non-trivial behavior, provides a unique opportunity for tractable whole-organism connectomics. This study imaged the Drosophila whole head using serial electron microscopy. Its compound eye was reconstructed, and the optical properties of the ommatidia as well as the connectome of the first visual neuropil-the lamina was analyze. Compared with the fruit fly and the honeybee, Megaphragma visual system is highly simplified: it has 29 ommatidia per eye and 6 lamina neuron types. Features are reported that are both stereotypical among most ommatidia and specialized to some. By identifying the "barebones" circuits critical for flying insects, these results will facilitate constructing computational models of visual processing in insects (Chua, 2023).

Advances in brain connectomics have demonstrated the extraordinary complexity of neural circuits. Developing neurons encounter the axons and dendrites of many different neuron types and form synapses with only a subset of them. During circuit assembly, neurons express cell-type-specific repertoires comprising many cell adhesion molecules (CAMs) that can mediate interactions between developing neurites. Many CAM families have been shown to contribute to brain wiring in different ways. It has been challenging, however, to identify receptor-ligand pairs directly matching neurons with their synaptic targets. This study integrated the synapse-level connectome of the neural circuit with the developmental expression patterns and binding specificities of CAMs on pre- and postsynaptic neurons in the Drosophila visual system. To overcome the complexity of neural circuits, focus was placed on pairs of genetically related neurons that make differential wiring choices. In the motion detection circuit, closely related subtypes of T4/T5 neurons choose between alternative synaptic targets in adjacent layers of neuropil. This choice correlates with the matching expression in synaptic partners of different receptor-ligand pairs of the Beat and Side families of CAMs. Genetic analysis demonstrated that presynaptic Side-II and postsynaptic Beat-VI restrict synaptic partners to the same layer. Removal of this receptor-ligand pair disrupts layers and leads to inappropriate targeting of presynaptic sites and postsynaptic dendrites. It is proposde that different Side/Beat receptor-ligand pairs collaborate with other recognition molecules to determine wiring specificities in the fly brain. Combining transcriptomes, connectomes, and protein interactome maps allow unbiased identification of determinants of brain wiring (Yoo, 2023).

Color and polarization provide complementary information about the world and are detected by specialized photoreceptors. However, the downstream neural circuits that process these distinct modalities are incompletely understood in any animal. Using electron microscopy, this study had systematically reconstructed the synaptic targets of the photoreceptors specialized to detect color and skylight polarization in Drosophila, and light microscopy was used to confirm many of the findings. Known and novel downstream targets were identified that are selective for different wavelengths or polarized light, and their projections were followed to other areas in the optic lobes and the central brain. The results revealed many synapses along the photoreceptor axons between brain regions, new pathways in the optic lobes, and spatially segregated projections to central brain regions. Strikingly, photoreceptors in the polarization-sensitive dorsal rim area target fewer cell types, and lack strong connections to the lobula, a neuropil involved in color processing. This reconstruction identifies shared wiring and modality-specific specializations for color and polarization vision, and provides a comprehensive view of the first steps of the pathways processing color and polarized light inputs (Kind, 2021).

Systematic reconstruction of all synaptic inputs and outputs of identified, functionally specialized Drosophila photoreceptors (pale and yellow R7-8 and two dorsal rim area photoreceptors, R7-DRA and R8-DRA) provides a comprehensive inventory of the first steps of the color and polarization pathways, from which all the computations of the dependent behaviors stem. These data revealed core connectomic motifs shared across column types, multiple new photoreceptor targets, and uncovered additional cell types as being connected to specific photoreceptor subtypes conveying specific color and polarization information to the central brain (Kind, 2021).

This study confirmed previously reported synaptic partners of the inner photoreceptors in the non-DRA medulla and identified new photoreceptor targets. As prior reconstructions were incomplete, it was unclear whether the unidentified connections were mainly onto new target neurons or represented more connections onto known cell types; reconstructions revealed both types of omissions. One functionally important set of missed connections are the synapses between inner photoreceptors from the same central and DRA ommatidia, which was found to be stronger than previously reported, due to significant numbers of synapses outside the medulla. These synapses likely contribute to color-opponent responses seen in central R7 and R8 terminals and the polarization-opponent signals measured from DRA photoreceptors. The reconstructions also support a larger-scale opponent process mediated by multicolumnar Dm9 cells, which also formed some synapses outside the medulla neuropil (Kind, 2021).

Other cell types also received inner photoreceptor input outside the medulla, notably the lamina monopolar cells L1 and L3. These lamina connections indicate that chromatic comparisons arising from R7 and R8 may feed into the motion vision pathway, and identifies a new site for interplay between the 'color' and 'motion' pathways. Together, these observations suggest that synapses in an unexpected location, outside the main synaptic layers of the medulla, could play a significant role in early visual processing (Kind, 2021).

In the non-DRA medulla, this study found, for the first time in an EM study, strong synaptic connections between R7s and MeTu cells that project to the AOTU, confirming previous claims based on light microscopy. This finding may reconcile disparate observations, such as a role in wavelength-specific phototaxis for cells matching MeTu morphology, as well as measurements of color-sensitive signals in the AOTU of bees. Previous anatomical studies partitioned MeTu cells into distinct subclasses that terminate in discrete subdomains of the AOTU. This study identified modality-specific MeTu-DRA cells that only integrate from the polarization-sensitive R7-DRA photoreceptors, while avoiding synaptic contacts with color-sensitive pale or yellow R7s. MeTu and MeTu-DRA cells target adjacent subdomains within the small unit of the AOTU, in agreement with proposals that parallel channels convey different forms of visual information from the eye to the central complex via the AOTU (Kind, 2021).

This study also identified connections from inner photoreceptors to several cell types either not previously described or not known to be photoreceptor targets, thus setting up a clear expectation that these cells should contribute to color or polarization vision. ML1 is a new, major target of R8, a cell type that connects the medulla to the lobula via a previously unknown, non-columnar pathway and the central brain. Previous studies have identified an important role for Tm5a/b/c and Tm20 cell types for chromatic processing in the lobula, and this study has confirmed that these cell types are targets of R7-8. Whether these Tm neurons and ML1 cells have common targets in the lobula, feed into shared central pathways or contribute to separate channels remain open questions; the lobula arbors of the Tm and ML1 cells are mostly in different layers arguing against direct synaptic interactions between the cells (Kind, 2021).

The reconstruction of DRA photoreceptor targets confirmed the modality-specific connectivity of Dm-DRA cell types within layer M6: Dm-DRA1 to R7-DRA, and Dm-DRA2 to R8-DRA. The R7-DRA and R8-DRA cells respond to orthogonal orientations of polarized light at each location of the DRA, so Dm-DRA1 and Dm-DRA2 likely process orthogonal e-vector orientations spatially averaged by pooling over ~10 ommatidia. The data also revealed additional DRA pathways into the PLP in the central brain via VPN-DRA cells, as well as to the contralateral DRA, via MeMe-DRA cells. Such interhemispheric connections have been demonstrated in larger insects, but not in Drosophila, and their synaptic input was not known. Interocular transfer contributes to navigation by desert ants that can see the celestial polarization but not visual landmarks, but the interactions between the DRA regions remain poorly understood. The identification of MeMe-DRA neurons may enable the mechanisms of such phenomena to be explored (Kind, 2021).

This study has additionally identified as pale-specific the aMe12 and ML-VPN1 cell types, and the yellow bias of Tm5c and the pale bias of Mi9. Detailed analysis of Dm8 inputs confirmed that these neurons receive most of their photoreceptor input in a central home column, consistent with pale and yellow subtypes of Dm8 cells having distinct chromatic properties. The selective photoreceptor input to these cell types, combined with input to columnar cells from a single photoreceptor subtype, indicates that wavelength-specific information is maintained in the medulla and lobula. The projections of aMe12 and ML-VPN1 to the central brain indicate the possibility that wavelength-specific photoreceptor responses are directly conveyed into the central brain by these cells, although they likely integrate input from other chromatically sensitive cell types (Kind, 2021).

By focusing on a small number of columns, this study has delivered a near-complete picture of local connectivity, but the possibility cannot be ruled out that certain cell types may have been overlooked. For instance, an arbitrary threshold was chosen of two to three synapses below which no attempts were made to reconstruct synaptic targets to the extent required to uniquely identify them. As a result this study may, in principle, have missed large cells that receive small, but significant inputs across many columns. Furthermore, regional specializations in the optic lobes, such that specific cell types might be found outside of the seed columns, cannot be ruled out. For example, some MeTu cells are only found in the dorsal third of the medulla, where incoming R7 cells are known to co-express Rh3 and Rh4 rhodopsins (Kind, 2021).

Taken together, the data presented in this study provides access to the full complement of R7 and R8 photoreceptor targets from functionally specialized optical units. By reconstructing these local circuits within a full-brain EM volume, it was possible to establish the complete morphology of large, multicolumnar cell types, that are strongly connected to photoreceptors, but had eluded previous connectomic reconstruction efforts. The sparse tracing approach that was implemented and described in this study enabled efficient identification of the complete set of upstream and downstream partners of the inner photoreceptors, in a manner that is complementary to the dense connectomes generated from smaller-scale medulla volumes. As an example of this synergistic usage of complementary data sets, this study has returned to the 7-column data and used whole-cell morphologies to match the bodies of previously unidentified photoreceptor targets. In that process strong candidates were established for MeTu, ML1, and perhaps for aMe12 and several aspects of their connectivity in the Zheng (Cell 174(3): 730-743. PubMed ID: 30033368) data set were confirmed. In so doing, it is now possible to have access to the additional connectivity data provided by the 7-column dense reconstruction. While full exploration and follow-up analyses of these combined data is beyond the scope of this work, the combined analysis of the Zheng (2018), and the medulla-7-column connectome revealed several intriguing connectivity patterns, including new candidate paths for the integration of output from different photoreceptor types. For example, the MeTu cells that are postsynaptic to R7 also receive indirect R8 input via the R8 target Mi15, and ML1 combines direct R8 input with indirect input from outer photoreceptors via lamina neurons. As further connectome data sets are completed, this comparative interplay between data sets with unique advantages and limitations will be an important step in both cross-validating and extending the applicability of all related data sets (Kind, 2021).

Reconstruction of the DRA photoreceptor targets has provided the first EM-based connectomic data set for modality-specific cell types likely to process skylight information in any insect and will be important for developing refined models of skylight navigation. Core motifs shared between DRA and central columns are prime candidates for circuit elements that perform computations, such as establishing opponency, that are key for both polarization and color processing, whereas cell types with preferential connections to either pale or yellow columns are promising candidates for the study of specific aspects of color processing in the insect brain. This comprehensive catalog of the neurons carrying signals from R7 and R8 photoreceptors deeper into the brain establishes a broad foundation for further studies into the mechanistic basis of color vision and its contributions to perception and behavior (Kind, 2021).

Topographic maps, the systematic spatial ordering of neurons by response tuning, are common across species. In Drosophila, the lobula columnar (LC) neuron types project from the optic lobe to the central brain, where each forms a glomerulus in a distinct position. However, the advantages of this glomerular arrangement are unclear. This study examine the functional and spatial relationships of 10 glomeruli using single-neuron calcium imaging. Novel detectors were discovered for objects smaller than the lens resolution (LC18) and for complex line motion (LC25). Glomeruli are spatially clustered by selectivity for looming versus drifting object motion and ordered by size tuning to form a topographic visual feature map. Furthermore, connectome analysis shows that downstream neurons integrate from sparse subsets of possible glomeruli combinations, which are biased for glomeruli encoding similar features. LC neurons are thus an explicit example of distinct feature detectors topographically organized to facilitate downstream circuit integration (Klapoetke, 2022).

Topographic maps, where neurons form ordered representation of the external world, have been long observed in many diverse brain regions. The prevalence of these maps has been suggested to support efficient organization of local circuit computation. In the early visual system, the vertebrate retina and insect optic lobe share many anatomic and functional similarities in encoding visual information retinotopically, and both output information to higher-order brain regions via visual projection neurons. The LC visual projection cells of flies, similar to retinal ganglion cells in vertebrates, have been hypothesized to encode behaviorally relevant local visual features. In Drosophila melanogaster, there are ∼20 anatomically distinct LC types (including LC-like visual projection cell types called lobula plate lobula columnar, or LPLC, neurons, each comprising a population of between 30 and 220 neurons. The dendrites of individual member neurons within an LC type tile retinotopic space in the optic lobe, but the axons of a given type collectively project to a target region in the fly's central brain called an optic glomerulus. The nonoverlapping optic glomeruli form a highly ordered, anatomically identifiable set in the central brain. However, the visual feature sensitivities of the different glomeruli, and whether the stereotyped anatomical glomerular arrangement relates to functional tuning, is not well understood. To date, feature encoding properties and behavioral links have been identified for only a handful of LC types. This study systematically explored the diversity of visual feature encoding by a broad set of LC neurons and further explored their connectivity onto downstream target neurons in order to establish how the anatomical arrangement of the glomeruli relates to their function and readout (Klapoetke, 2022).

Systematic survey of LC visual projection neuron response properties reveals a topographic visual feature map in the fly central brain. LC axon terminals bundle into discrete central brain glomeruli by cell type, and each glomerulus is distinctly tuned to a visual moving object class, with neighboring glomeruli more similarly tuned than distant ones. Indeed, the physical arrangement of the glomeruli is largely recapitulated by sorting LC neuron responses by their first two principal components, which roughly correspond to selectivity for objects versus looming motion and object size tuning (Klapoetke, 2022).

The LCs that were measured all have distinct visual feature tunings across object size and speed, with each cell type's preferred object size generally eliciting large responses across one order of magnitude of movement speeds. Whether size selectivity is preserved across different speeds in LC11 and LC18 responses was further examined, and it was observed that both cell types retain selectivity to moving small objects, but the peak size sensitivity shifts to smaller sizes at slower speeds and to bigger sizes at higher speed. It is noteworthy that this trend for peak sensitivity shifting to slightly larger sizes at faster motion speed may have a simple mechanistic basis (a larger, faster moving object can have a similar stimulus dwell time as a smaller, slower moving object). This peak sensitivity shift is also consistent with the typical perceived size-speed relationship at different depths (such as motion parallax cues) because at a distance, an object will appear to be smaller and to move across the retina more slowly but will become both faster and larger as it approaches the observer (Klapoetke, 2022).

A minimal model was generated that recapitulates LC18's size tuning to objects smaller than an individual ommatidium's lens resolution using common circuit motifs such as differing contrast responses to ON and OFF signals and crossover inhibition between the ON and OFF pathways. Because the hemibrain connectome is truncated in the optic lobe, it was not possible to systematically examine the inputs to LC18 to identify whether the propoaws circuit motifs exist in the connectome. Previous studies have indicated that LC11 neurons, which also detect small objects, receive inputs from T2 cells. The T2 neurons have distinctive axonal terminal morphology, which allowed identification of individual putative T2 neurons as inputs to both LC11 and LC18 in the hemibrain. Interestingly, these same putative T2 neurons are strong inputs to LC4 and LPLC1, which have size tunings distinct from both LC11 and LC18. These findings suggest that the diverse feature tuning observed across LC types is unlikely to be inherited directly from the properties of the input neurons. Understanding the cell-type specific integration of distinct combinations of input cell types, including inhibitory inputs, will be required to account for the size and speed tuning differences between LC11, LC18, and other LCs (Klapoetke, 2022).

What can the visual feature map that this study uncovered tell about what Drosophila perceive? Eemarkably precise encoding of particular features was found including looming detection (LPLC2), very small objects (LC18), and thin lines (LC25). It is unlikely that these features, especially in isolation, can be meaningfully interpreted to drive a behavioral response. Indeed, activation of LC18 and LC25 neurons did not produce any detectable behavior change in a previous study (although LPLC2 activation drives a takeoff jump). Rather, it may be necessary to integrate multiple visual features to generate a more reliable estimate of external moving objects. For example, LPLC2's signals are integrated together with LC4's by a downstream descending neuron that serves to detect approaching objects, such as predators, and mediates a visual-looming-evoked takeoff jump (Klapoetke, 2022).

Consistent with this expectation, the readout across LC neurons by downstream cells was examined in the central brain, and half of these strongly connected targets integrated their inputs from more than one LC type. These downstream neurons typically integrate from only a small fraction of the possible combinations of glomeruli, and these combinations are highly biased to neighboring glomeruli, with similar feature encoding. The downstream neurons are a highly diverse set, with cells that output to other central brain regions, provide feedback to the optic lobe, project to the contralateral hemisphere, or even directly project to the ventral nerve cord. Furthermore, many downstream targets that only integrate from a single LC type project contralaterally, suggesting that binocular interactions are a substantial feature of the glomeruli, as they are for circuits downstream of LC6, which have previously been explored (Klapoetke, 2022).

Although this analysis focused exclusively on the projection neurons from the lobula to the ventrolateral central brain, there is another prominent lobula visual projection, the LC10 neurons, that target a more dorsal central brain area called the AOTU. LC10 neurons have distinct, but overlapping, sensitivity to visual features, especially for moving objects, from the ventrolateral-projecting LCs. The LC10 axon terminals also feature prominent retinotopic organization within the AOTU, which is quite different from the other optic glomeruli. Furthermore, LC10 neurons are critical for visual tracking during courtship behavior, whereas other LCs are involved in locomotor behaviors such as walking forward, turning, freezing, or escape takeoff. Future work should evaluate whether LC10 pathways ultimately converge with neurons downstream of the other LCs to coordinate diverse behaviors guided by these different visual pathways (Klapoetke, 2022).

Circuit level analysis of feature integration has been limited by the lack of detailed connectivity information between neurons with known feature encoding. In the fly brain, most central projections of visual pathways are not yet characterized. By functionally characterizing a large set of spatially adjacent visual projection neuron types and then taking advantage of the recent central brain connectome, this study found that the integration of LC visual features is highly selective, with 33% of possible feature pairs observed. Furthermore, these feature combinations are highly biased for nearby glomeruli, suggesting local circuit wiring economy as a likely driving force for the glomeruli spatial positions. Such structured integration serves as an exemplar for how brains can organize sensory information into ordered feature maps to compose diverse pathways that represent critical conduits between vision and behavioral controls (Klapoetke, 2022).

Animals have evolved sophisticated visual circuits to solve a vital inference problem: detecting whether or not a visual signal corresponds to an object on a collision course. Such events are detected by specific circuits sensitive to visual looming, or objects increasing in size. Various computational models have been developed for these circuits, but how the collision-detection inference problem itself shapes the computational structures of these circuits remains unknown. Inspired by the distinctive structures of LPLC2 neurons (ultra-selective looming detecting neuron, lobula plate/lobula columnar, type II (LPLC2) in Drosophila) in the visual system of Drosophila, this study built anatomically-constrained shallow neural network models and trained them to identify visual signals that correspond to impending collisions. Surprisingly, the optimization arrives at two distinct, opposing solutions, only one of which matches the actual dendritic weighting of LPLC2 neurons. Both solutions can solve the inference problem with high accuracy when the population size is large enough. The LPLC2-like solutions reproduces experimentally observed LPLC2 neuron responses for many stimuli, and reproduces canonical tuning of loom sensitive neurons, even though the models are never trained on neural data. Thus, LPLC2 neuron properties and tuning are predicted by optimizing an anatomically-constrained neural network to detect impending collisions. More generally, these results illustrate how optimizing inference tasks that are important for an animal's perceptual goals can reveal and explain computational properties of specific sensory neurons (Zhou, 2022).

Many sighted animals solve this inference problem with high precision, thanks to robust loom-selective neural circuits evolved over hundreds of millions of years. The neuronal mechanisms for response to looming stimuli have been studied in a wide range of vertebrates, from cats and mice to zebrafish, as well as in humans. In invertebrates, detailed anatomical, neurophysiological, behavioral, and modeling studies have investigated loom detection, especially in locusts and flies. An influential mathematical model of loom detection was derived by studying the responses of the giant descending neurons of locusts. This model established a relationship between the timing of the neurons' peak responses and an angular size threshold for the looming object. Similar models have been applied to analyze neuronal responses to looming signals in flies, where genetic tools make it possible to precisely dissect neural circuits, revealing various neuron types that are sensitive to looming signals (Zhou, 2022).

However, these computational studies did not directly investigate the relationship between the structure of the loom-sensitive neural circuits and the inference problem they appear to solve. On the one hand, the properties of many sensory circuits appear specifically tuned to the tasks that they are executing. In particular, by taking into account relevant behaviors mediated by specific sensory neurons, experiments can provide insight into their tuning properties. On the other hand, computational studies that have trained artificial neural networks to solve specific visual and cognitive tasks, such as object recognition or motion estimation, have revealed response patterns similar to the corresponding biological circuits or even individual neurons. Thus, in this study it is asked whether the properties associated with neural loom detection can be reproduced simply by optimizing shallow neural networks for collision detection (Zhou, 2022).

The starting point for the computational model of loom detection is the known neuroanatomy of the visual system of the fly. In particular, the loom-sensitive neuron LPLC2 (lobula plate/lobula columnar, type 2) has been studied in detail. These neurons tile visual space, sending their axons to descending neurons called the giant fibers (GFs), which trigger the fly's jumping and take-off behaviors. Each LPLC2 neuron has four dendritic branches that receive inputs at the four layers of the lobula plate (LP). The retinotopic LP layers host the axon terminals of motion detection neurons, and each layer uniquely receives motion information in one of the four cardinal directions. Moreover, the physical extensions of the LPLC2 dendrites align with the preferred motion directions in the corresponding LP layers. These dendrites form an outward radial structure, which matches the moving edges of a looming object that expands from the receptive field center. Common stimuli such as the wide-field motion generated by movement of the insect only match part of the radial structure, and strong inhibition for inward-directed motion suppresses responses to such stimuli. Thus, the structure of the LPLC2 dendrites favors responses to visual stimuli with edges moving radially outwards, corresponding to objects approaching the receptive field center (Zhou, 2022).

The focus of this paper is to investigate how loom detection in LPLC2 can be seen as the solution to a computational inference problem. Can the structure of the LPLC2 neurons be explained in terms of optimization-carried out during the course of evolution-for the task of predicting which trajectories will result in collisions? How does coordination among the population of more than 200 LPLC2 neurons tiling a fly's visual system affect this optimization? To answer these questions, simple anatomically-constrained neural network models were built that receive motion signals in the four cardinal directions. The model using artificial stimuli to detect visual objects on a collision course with the observer. Surprisingly, optimization finds two distinct types of solutions, with one resembling the LPLC2 neurons and the other having a very different configuration. How each of these solutions detects looming events and where they show distinct individual and population behaviors were analyzed. When tested on visual stimuli not in the training data, the optimized solutions with filters that resemble LPLC2 neurons exhibit response curves that are similar to those of LPLC2 neurons measured experimentally. Importantly, although it only receives motion signals, the optimized model shows characteristics of an angular size encoder, which is consistent with many biological loom detectors, including LPLC2. These results show that optimizing a neural network to detect looming events can give rise to the properties and tuning of LPLC2 neurons (Zhou, 2022).

The radially structured dendrites of the LPLC2 neuron in the LP can account for its response to motion radiating outward from the receptive field center. The results show that the logic of this computation can be understood in terms of inferential loom detection by the population of units. In particular, for an individual detector unit, an inward structure can make a better loom detector than an outward structure, since it is sensitive to colliding objects originating from a wider array of incoming angles. As the number of units across visual space increases, the performance of the outward-sensitive receptive field structure comes to match the performance of the inward solutions. As the number of units increases, the inhibitory component of the outward solutions also becomes broader as the population size becomes larger, which is crucial for reproducing key experimental observations, such as peripheral inhibition. The optimized solutions depend on the number of detectors, and this is likely related to the increasing overlap in receptive fields as the population grows. This result is consistent with prior work showing that populations of neurons often exhibit different and improved coding strategies compared to individual neurons. Thus, understanding anatomical, physiological, and algorithmic properties of individual neurons can require considering the population response. The solutions that were found to the loom inference problem suggest that individual LPLC2 responses should be interpreted in light of the population of LPLC2 responses (Zhou, 2022).

Our results shed light on discussions of η-like (encoding angular size) and ρ-like (encoding angular velocity) looming sensitive neurons in the literature. In particular, these optimized models clarify an interesting but puzzling fact: LPLC2 neurons transform their inputs of direction-selective motion signals to computations of angular size. Consistent with this tuning, the model also shows a linear relationship between the peak time relative to collision and the R/η ratio, which should be followed by loom sensitive neurons that encode angular size. In both cases, these properties appear to be the simple result of training the constrained model to reliably detect looming stimuli (Zhou, 2022).

The units of the outward solution exhibit sparsity in their responses to looming stimuli, in contrast to the denser representations in the inward solution. During a looming event, in an outward solution, most of the units are quiet and only a few adjacent units have very large activities, reminiscent of sparse codes that seem to be favored, for instance, in cortical encoding of visual scenes. Since the readout of the model is a summation of the activities of the units, sparsity does not directly affect the performance of the model, but is an attribute of the favored solution. For a model with a different loss function or with noise, the degree of sparsity might be crucial. For instance, the sparse code of the outward model might make it easier to localize a hit stimulus, or might make the population response more robust to noise (Zhou, 2022).

Experiments have shown that inhibitory circuits play an important role for the selectivity of LPLC2 neurons. For example, motion signals at the periphery of the receptive field of an LPLC2 neuron inhibit its activity. This peripheral inhibition causes various interesting response patterns of the LPLC2 neurons to different types of stimuli. However, the structure of this inhibitory field is not fully understood, and the model provides a tool to investigate how the inhibitory inputs to LPLC2 neurons affect circuit performance on loom detection tasks. The strong inhibition on the periphery of the receptive field arises naturally in the outward solutions after optimization. The extent of the inhibitory components increases as more units are added to models. The broad inhibition appears in the model to suppress responses to the non-hit stimuli, and as in the data, the inhibition is broader than one might expect if the neuron were simply being inhibited by inward motion. These larger inhibitory fields are also consistent with the larger spatial pooling likely to be supplied by inhibitory LPi inputs (Zhou, 2022).

The synthetic stimuli used to train models in this study were unnatural in two ways. The first way was in the proportion of hits and non-hits. Training was carried out with 25% of the training data representing hits. The true fraction of hits among all stimuli encountered by a fly is undoubtedly much less, and this affects how the loss function weights different types of errors. It is also clear that a false-positive hit (in which a fly might jump to escape an object not on collision course) is much less penalized during evolution than a false-negative (in which a fly doesn't jump and an object collides, presumably to the detriment of the fly). It remains unclear how to choose these weights in the training data or in the loss function, but they affect the receptive field weights optimized by the model (Zhou, 2022).

The second issue with the stimuli is that they were caricatures of stimulus types, but did not incorporate the richness of natural stimuli. This richness could include natural textures and spatial statistics, which seem to impact motion detection algorithms. This richness could also include more natural trajectories for approaching objects. Another way to enrich the stimuli would be to add noise, either in inputs to the model or in the model's units themselves. This was explored briefly by adding self-rotation-generated background motion; under those conditions, both solutions were present but optimized outward solutions performed better than the inward solutions. This indicates that the statistics of the stimuli may play an important role in selecting solutions for loom detection. However, it remains less clear what the true performance limits of loom detection are, since most experiments use substantially impoverished looming stimuli. Moreover, it is challenging to characterize the properties of natural looming events. An interesting future direction will be to investigate the effects of more complex and naturalistic stimuli on the model's filters and performance, as well as on LPLC2 neuron responses themselves (Zhou, 2022).

For simplicity, the models did not impose the hexagonal geometry of the compound eye ommatidia. Instead, it was assumed that the visual field is separated into a Cartesian lattice with 5° spacing, each representing a local motion detector with two spatially separated inputs. This simplification alters slightly the geometry of the motion signals compared to the real motion detector receptive fields. This could potentially affect the learned spatial weightings and reproduction of the LPLC2 responses to various stimuli, since the specific shapes of the filters matter. Thus, the hexagonal ommatidial structure and the full extent of inputs to T4 and T5 might be crucial if one wants to make comparisons with the dynamics and detailed responses of LPLC2 neurons. However, this geometric distinction seems unlikely to affect the main results of how to infer the presence of hit stimuli (Zhou, 2022).

The model requires a field of estimates of the local motion. Here, the simplest model was used - the Hassenstein-Reichardt correlator model Equation 3 - but the model could be extended by replacing it with a more sophisticated model for motion estimation. Some biophysically realistic ones might take into account synaptic conductances and could respond to static features of visual scenes. Alternatively, in natural environments, contrasts fluctuate in time and space. Thus, if one includes more naturalistic spatial and temporal patterns, one might consider a motion detection model that could adapt to changing contrasts in time and space (Zhou, 2022).

Although the outward filter of the unit emerges naturally from the gradient descent training protocol, that does not mean that the structure is learned by LPLC2 neurons in the fly. There may be some experience dependent plasticity in the fly eye , but these visual computations are likely to be primarily genetically determined. Thus, one may think of the computation of the LPLC2 neuron as being shaped through millions of years of evolutionary optimization. Optimization algorithms at play in evolution may be able to avoid getting stuck in local optima, and thus work well with the sort of shallow neural network found in the fly eye (Zhou, 2022).

This study focused on the motion signal inputs to LPLC2 neurons, and other inputs to LPLC2 neurons were neglected, such as those coming from the lobula that likely report non-motion visual features. It would be interesting to investigate how this additional non-motion information affects the performance and optimized solutions of the inference units. For instance, another lobula columnar neurons, LC4, is loom sensitive and receives inputs in the lobula. The LPLC2 and LC4 neurons are the primary excitatory inputs to the GF, which mediates escape behaviors. The inference framework set out here would allow one to incorporate parallel non-motion intensity channels, either by adding them into the inputs to the LPLC2-like units, or by adding in a parallel population of LC4-like units. This would require a reformulation of a probabilistic model. Notably, one of the most studied loom detecting neurons, the lobula giant movement detector (LGMD) in locusts, does not appear to receive direction-selective inputs, as LPLC2 does. Thus, the inference framework set out here could be flexibly modified to investigate loom detection under a wide variety of constraints and inputs, which allow it to be applied to other neurons, beyond LPLC2 (Zhou, 2022).

Self-motion generates visual patterns on the eye that are important for navigation. These optic flow patterns are encoded by the population of local direction-selective cells in the mouse retina, whereas in flies, local direction-selective T4/T5 cells are thought to be uniformly tuned. How complex global motion patterns can be computed downstream is unclear. This study shows that the population of T4/T5 cells in Drosophila encodes global motion patterns. Whereas the mouse retina encodes four types of optic flow, the fly visual system encodes six. This matches the larger number of degrees of freedom and the increased complexity of translational and rotational motion patterns during flight. The four uniformly tuned T4/T5 subtypes described previously represent a local subset of the population. Thus, a population code for global motion patterns appears to be a general coding principle of visual systems that matches local motion responses to modes of the animal's movement (Henning, 2022).

Color and polarization provide complementary information about the world and are detected by specialized photoreceptors. However, the downstream neural circuits that process these distinct modalities are incompletely understood in any animal. Using electron microscopy, this study has systematically reconstructed the synaptic targets of the photoreceptors specialized to detect color and skylight polarization in Drosophila, and light microscopy was used to confirm many of the findings. Known and novel downstream targets were identified that are selective for different wavelengths or polarized light, and their projections were followed to other areas in the optic lobes and the central brain. The results revealed many synapses along the photoreceptor axons between brain regions, new pathways in the optic lobes, and spatially segregated projections to central brain regions. Strikingly, photoreceptors in the polarization-sensitive dorsal rim area target fewer cell types, and lack strong connections to the lobula, a neuropil involved in color processing. This reconstruction identifies shared wiring and modality-specific specializations for color and polarization vision, and provides a comprehensive view of the first steps of the pathways processing color and polarized light inputs (Kind, 2021).

Both the wavelength and the polarization angle of light contain valuable information that can be exploited by many visual animals. For instance, color gradients across the sky can serve as navigational cues, and skylight's characteristic pattern of linear polarization can also inform navigation by indicating the orientation relative to the sun. The spectral content of light is detected by groups of photoreceptor cells containing rhodopsin molecules with different sensitivities, often organized in stochastic retinal mosaics, and specialized, polarization-sensitive photoreceptors have been characterized in many species, both vertebrates and invertebrates. These two visual modalities, color and polarization vision, require the processing of signals over a wide range of spatial and temporal scales, and many questions remain about how the signals from functionally specialized photoreceptors are integrated in downstream neurons. Are color and polarization signals mixed at an early stage, or are they processed by different, modality-specific cell types? Do separate pathways exist that selectively process and convey information from photoreceptor types in the retinal mosaic to targets in the central brain? The full scope of the early synaptic stages of color and polarization circuitry is unknown in any animal, and the analysis of electron microscopy (EM) connectomes is ideally suited to exhaustively answer these questions, especially when corroborated with genetic labeling of cell types and circuit elements imaged with light microscopy. This study makes significant progress on these questions by mapping the neuronal connections of specialized, identified photoreceptors within the Drosophila full adult fly brain data set and validates many of these results by using the powerful genetic tools available in Drosophila (Kind, 2021).

While studies in many insects have contributed to the understanding of polarized light and color vision, the visual system of Drosophila offers many advantages for the exploration of neural circuits. Anatomical studies are facilitated by the stereotyped, repetitive structure of the optic lobes, with many cell types, the so-called columnar neurons, found in repeated circuit units, called visual columns, that are retinotopically arranged and each correspond to one of the ~800 unit eyes (ommatidia) of the compound eye. Over 100 optic lobe cell types have been described in classical Golgi work, by more recent studies combining genetic labeling with light microscopy and, for some cell types, through EM reconstructions that have revealed not only cell morphologies but also detailed synaptic connectivity. Furthermore, genetic tools, and, most recently, gene expression data are available for many optic lobe cell types. An enabling feature of this emerging body of work is that in nearly all cases, distinct cell types can be reliably identified across data sets, such that new studies often directly enrich prior ones (Kind, 2021).

Each Drosophila ommatidium contains eight photoreceptors whose output is processed in a series of neuropils called the lamina, medulla, lobula, and lobula plate that together form the optic lobes of the fly. Outer photoreceptors R1-6 project to the lamina neuropil, and serve as the main input to the motion vision circuitry; inner photoreceptors R7 and R8 pass through the lamina without connections and project directly to the deeper medulla neuropil, which also receives lamina projections. The organization of the inner photoreceptors along the dorsal rim area (DRA) of the eye characteristically differs from that of the rest of the retina. In the non-DRA part of the retina, R7 and R8 differ in their axonal target layers, with R7 projecting to layer M6, and R8 to layer M3. R7 and R8 also differ in their rhodopsin expression, being sensitive to short wavelength ultraviolet (UV, R7) and blue (R8), respectively, in so-called 'pale' ommatidia, and to long wavelength UV (R7) and green (R8) in 'yellow' ommatidia. Pale and yellow ommatidia are distributed randomly, at an uneven ratio that is conserved across insects. Meanwhile, DRA ommatidia are morphologically and molecularly specialized for detecting skylight polarization, that Drosophila can use to set a heading. In the DRA, the inner photoreceptors express the same UV rhodopsin (Rh3) and detect perpendicular angles of polarized UV light. In contrast to the rest of the medulla, R7-DRA and R8-DRA project to the same medulla layer (M6), where their targets include polarization-specific cell types. Across insects, a 'compass pathway' connects the DRA to the central brain via the anterior optic tubercle (AOTU). Anatomical and functional data from Drosophila suggests that the non-DRA medulla is also connected to the compass pathway, potentially forming parallel pathways for processing different celestial cues (Kind, 2021).

EM studies have already revealed some of the circuitry downstream of R7 and R8. For example, axons of R7 and R8 from the same ommatidium are reciprocally connected with inhibitory synapses, leading to color-opponent signals in their presynaptic terminals. Interestingly, R7-DRA and R8-DRA also inhibit each other. Other known R7 and R8 targets in the main medulla include local interneurons (e.g. Dm8) and projection neurons that provide connections to deeper optic lobe regions (e.g. Tm5 and Tm20 neurons). A previous light microscopy study identified a single cell type, Tm5a, that is specific for yellow medulla columns; this neuron has been used to identify pale and yellow columns in an EM volume. Using genetic labeling techniques, four classes of TmY cells have also been reported as specific targets of pale versus yellow photoreceptors, yet previous connectomic studies did not reveal similar cells. The currently most comprehensive EM study of the medulla reconstructed the connections between neurons in seven neighboring medulla columns, revealing a detailed, yet incomplete, inventory of cell types connected to R7-8. This data set, now publicly available, is remarkable for its dense reconstruction of columnar circuits, but could not be used to identify many multicolumnar neurons, that were cut-off at the edge of the data volume, leaving ~40% of R7-8 synapses onto unidentified cell types. In addition, no EM-based reconstructions of DRA neurons and their connections are currently available (Kind, 2021).

This study presents a comprehensive reconstruction of all inner photoreceptor synaptic outputs and inputs, from pairs of pale and yellow columns and from three DRA columns in the data set. These reconstructions were carried out within a full-brain volume, such that, for the first time, nearly all neurons connected to these photoreceptors were identified. Light microscopy revealed a large visual projection neuron (VPN) with distinctive morphology, named accessory medulla cell type 12 (aMe12), that selectively innervates pale ommatidia across the medulla. This cell was reconstructed and subsequently used to identify pale and yellow columns, from which the connectivity of R7 and R8 was enumerated with known and novel cell types within the optic lobes and projecting to the central brain, including more cells with pale-yellow specificity, and synapses on axons between neuropils. In the DRA, it was shown that cellular diversity is reduced, with local interneurons and projection neurons to the AOTU dominating, and connections to the lobula virtually missing. Circuit motifs shared between DRA and central columns were identified, and modality-specific cell types were identified, including cells with interhemispheric connections and projections to the central brain. Together, this study identified the connected neurons that account for 96% of these inner photoreceptor synapses, a comprehensive set of the neurons that comprise the first step of the pathways through which color and polarization signals are transduced to the rest of the brain (Kind, 2021).

This systematic reconstruction of all synaptic inputs and outputs of identified, functionally specialized Drosophila photoreceptors (pale and yellow R7-8, R7-DRA, and R8-DRA) provides a comprehensive inventory of the first steps of the color and polarization pathways, from which all the computations of the dependent behaviors stem. These data revealed core connectomic motifs shared across column types, multiple new photoreceptor targets, and uncovered additional cell types as being connected to specific photoreceptor subtypes conveying specific color and polarization information to the central brain (Kind, 2021).

Previously reported synaptic partners of the inner photoreceptors in the non-DRA medulla were confirmed, and new photoreceptor targets were identified. As prior reconstructions were incomplete, it was unclear whether the unidentified connections were mainly onto new target neurons or represented more connections onto known cell types; the current reconstructions revealed both types of omissions. One functionally important set of missed connections are the synapses between inner photoreceptors from the same central and DRA ommatidia, which this study found to be stronger than previously reported, due to significant numbers of synapses outside the medulla. These synapses likely contribute to color-opponent responses seen in central R7 and R8 terminals and the polarization-opponent signals measured from DRA photoreceptors. The reconstructions also support a larger-scale opponent process mediated by multicolumnar Dm9 cells, which also formed some synapses outside the medulla neuropil (Kind, 2021).

Other cell types also received inner photoreceptor input outside the medulla, notably the lamina monopolar cells L1 and L3. These lamina connections indicate that chromatic comparisons arising from R7 and R8 may feed into the motion vision pathway, and identifies a new site for interplay between the 'color' and 'motion' pathways. Together, these observations suggest that synapses in an unexpected location, outside the main synaptic layers of the medulla, could play a significant role in early visual processing (Kind, 2021).

In the non-DRA medulla, this study found, for the first time in an EM study, strong synaptic connections between R7s and MeTu cells that project to the AOTU, confirming previous claims based on light microscopy. This finding may reconcile disparate observations, such as a role in wavelength-specific phototaxis for cells matching MeTu morphology, as well as measurements of color-sensitive signals in the AOTU of bees. Previous anatomical studies partitioned MeTu cells into distinct subclasses that terminate in discrete subdomains of the AOTU. T identified modality-specific MeTu-DRA cells that only integrate from the polarization-sensitive R7-DRA photoreceptors, while avoiding synaptic contacts with color-sensitive pale or yellow R7s. Our MeTu and MeTu-DRA cells target adjacent subdomains within the small unit of the AOTU (Figure 10E), in agreement with proposals that parallel channels convey different forms of visual information from the eye to the central complex via the AOTU (Kind, 2021).

This study also identified connections from inner photoreceptors to several cell types either not previously described or not known to be photoreceptor targets, thus setting up a clear expectation that these cells should contribute to color or polarization vision. ML1 is a new, major target of R8, a cell type that connects the medulla to the lobula via a previously unknown, non-columnar pathway and the central brain. Previous studies have identified an important role for Tm5a/b/c and Tm20 cell types for chromatic processing in the lobula, and this study has confirmed that these cell types are targets of R7-8. Whether these Tm neurons and ML1 cells have common targets in the lobula, feed into shared central pathways or contribute to separate channels remain open questions; the lobula arbors of the Tm and ML1 cells are mostly in different layers arguing against direct synaptic interactions between the cells (Kind, 2021).

The reconstruction of DRA photoreceptor targets confirmed the modality-specific connectivity of Dm-DRA cell types within layer M6: Dm-DRA1 to R7-DRA, and Dm-DRA2 to R8-DRA (see Comparison of central versus dorsal rim area (DRA), and pale versus yellow pathways). The R7-DRA and R8-DRA cells respond to orthogonal orientations of polarized light at each location of the DRA, so Dm-DRA1 and Dm-DRA2 likely process orthogonal e-vector orientations spatially averaged by pooling over ~10 ommatidia. The data also revealed additional DRA pathways into the PLP in the central brain via VPN-DRA cells, as well as to the contralateral DRA, via MeMe-DRA cells. Such interhemispheric connections have been demonstrated in larger insects, but not in Drosophila, and their synaptic input was not known. Interocular transfer contributes to navigation by desert ants that can see the celestial polarization but not visual landmarks, but the interactions between the DRA regions remain poorly understood. The identification of MeMe-DRA neurons may enable the mechanisms of such phenomena to be explored (Kind, 2021).

For color vision, this study has confirmed the pale-yellow specificity of Tm5a and Tm5b cells. The aMe12 and ML-VPN1 cell types were identified as pale-specific, and the yellow bias of Tm5c and the pale bias of Mi9 were identified as well. Detailed analysis of Dm8 inputs confirmed that these neurons receive most of their photoreceptor input in a central home column, consistent with pale and yellow subtypes of Dm8 cells having distinct chromatic properties. The selective photoreceptor input to these cell types, combined with input to columnar cells from a single photoreceptor subtype, indicates that wavelength-specific information is maintained in the medulla and lobula. The projections of aMe12 and ML-VPN1 to the central brain indicate the possibility that wavelength-specific photoreceptor responses are directly conveyed into the central brain by these cells, although they likely integrate input from other chromatically sensitive cell types (Kind, 2021).

By focusing on a small number of columns, this study has delivered a near-complete picture of local connectivity, but the possibility cannot be ruled out that certain cell types may have been overlooked. For instance, an arbitrary threshold was chosen of two to three synapses below which attempts were not made to reconstruct synaptic targets to the extent required to uniquely identify them. As a result, in principle, this study may have missed large cells that receive small, but significant inputs across many columns. Furthermore, regional specializations in the optic lobes, such that specific cell types might be found outside of the seed columns, cannot be ruled out. For example, some MeTu cells are only found in the dorsal third of the medulla, where incoming R7 cells are known to co-express Rh3 and Rh4 rhodopsins (Kind, 2021).

Taken together, the data presented in this study provides access to the full complement of R7 and R8 photoreceptor targets from functionally specialized optical units. By reconstructing these local circuits within a full-brain EM volume, it was possible to establish the complete morphology of large, multicolumnar cell types, that are strongly connected to photoreceptors, but had eluded previous connectomic reconstruction efforts. The sparse tracing approach implemented and described in this study enabled efficient identification of the complete set of upstream and downstream partners of the inner photoreceptors, in a manner that is complementary to the dense connectomes generated from smaller-scale medulla volumes. As an example of this synergistic usage of complementary data sets, this study has returned to the 7-column data and used whole-cell morphologies to match the bodies of previously unidentified photoreceptor targets. In that process, strong candidates have been established for MeTu, ML1, and perhaps for aMe12 and several aspects of their connectivity were confirmed in a previous data set. In so doing, this study now has access to the additional connectivity data provided by the 7-column dense reconstruction. While full exploration and follow-up analyses of these combined data is beyond the scope of this work, the combined analysis a previous study and the medulla-7-column connectome revealed several intriguing connectivity patterns, including new candidate paths for the integration of output from different photoreceptor types. For example, the MeTu cells that are postsynaptic to R7 also receive indirect R8 input via the R8 target Mi15, and ML1 combines direct R8 input with indirect input from outer photoreceptors via lamina neurons. As further connectome data sets are completed, this comparative interplay between data sets with unique advantages and limitations will be an important step in both cross-validating and extending the applicability of all related data sets (Kind, 2021).

Reconstruction of the DRA photoreceptor targets provides the first EM-based connectomic data set for modality-specific cell types likely to process skylight information in any insect and will be important for developing refined models of skylight navigation. Core motifs shared between DRA and central columns are prime candidates for circuit elements that perform computations, such as establishing opponency, that are key for both polarization and color processing, whereas cell types with preferential connections to either pale or yellow columns are promising candidates for the study of specific aspects of color processing in the insect brain. This comprehensive catalog of the neurons carrying signals from R7 and R8 photoreceptors deeper into the brain establishes a broad foundation for further studies into the mechanistic basis of color vision and its contributions to perception and behavior (Kind, 2021).

Color vision is an important sensory capability of humans and many animals. It relies on color opponent processing in visual circuits that gradually compare the signals of photoreceptors with different spectral sensitivities. In Drosophila, this comparison begins already in the presynaptic terminals of UV-sensitive R7 and longer wavelength-sensitive R8 inner photoreceptors that inhibit each other in the medulla. How downstream neurons process their signals is unknown. This study reports that the second order medulla interneuron Dm8 is inhibited when flies are stimulated with UV light and strongly excited in response to a broad range of longer wavelength (VIS) stimuli. Inhibition to UV light is mediated by histaminergic input from R7 and expression of the histamine receptor ort in Dm8, as previously suggested. However, two additional excitatory inputs antagonize the R7 input. First, activation of R8 leads to excitation of Dm8 by non-canonical photoreceptor signaling and cholinergic neurotransmission in the visual circuitry. Second, activation of outer photoreceptors R1-R6 with broad spectral sensitivity causes excitation in Dm8 through the cholinergic medulla interneuron Mi1, which is known for its major contribution to the detection of spatial luminance contrast and visual motion. In summary, Dm8 mediates a second step in UV/VIS color opponent processing in Drosophila by integrating input from all types of photoreceptors. These results demonstrate novel insights into the circuit integration of R1-R6 into color opponent processing and reveal that chromatic and achromatic circuitries of the fly visual system interact more extensively than previously thought (Pagni, 2021).

Appropriate termination of the photoresponse in image-forming photoreceptors and downstream neurons is critical for an animal to achieve high temporal resolution. Although the cellular and molecular mechanisms of termination in image-forming photoreceptors have been extensively studied in Drosophila, the underlying mechanism of termination in their downstream large monopolar cells remains less explored. This study shows that synaptic ACh signaling, from both amacrine cells (ACs) and L4 neurons, facilitates the rapid repolarization of L1 and L2 neurons. Intracellular recordings in female flies show that blocking synaptic ACh output from either ACs or L4 neurons leads to slow repolarization of L1 and L2 neurons. Genetic and electrophysiological studies in both male and female flies determine that L2 neurons express ACh receptors and directly receive ACh signaling. Moreover, the results demonstrate that synaptic ACh signaling from both ACs and L4 neurons simultaneously facilitates ERG termination. Finally, visual behavior studies in both male and female flies show that synaptic ACh signaling, from either ACs or L4 neurons to L2 neurons, is essential for the optomotor response of the flies in high-frequency light stimulation. This study identifies parallel synaptic ACh signaling for repolarization of L1 and L2 neurons and demonstrates that synaptic ACh signaling facilitates L1 and L2 neuron repolarization to maintain the optomotor response of the fly on high-frequency light stimulation (Wu, 2021).

Color vision is an important sensory capability of humans and many animals. It relies on color opponent processing in visual circuits that gradually compare the signals of photoreceptors with different spectral sensitivities. In Drosophila, this comparison begins already in the presynaptic terminals of UV-sensitive R7 and longer wavelength-sensitive R8 inner photoreceptors that inhibit each other in the medulla. How downstream neurons process their signals is unknown. This paper reports that the second order medulla interneuron Dm8 is inhibited when flies are stimulated with UV light and strongly excited in response to a broad range of longer wavelength (VIS) stimuli. Inhibition to UV light is mediated by histaminergic input from R7 and expression of the histamine receptor ort in Dm8, as previously suggested. However, two additional excitatory inputs antagonize the R7 input. First, activation of R8 leads to excitation of Dm8 by non-canonical photoreceptor signaling and cholinergic neurotransmission in the visual circuitry. Second, activation of outer photoreceptors R1-R6 with broad spectral sensitivity causes excitation in Dm8 through the cholinergic medulla interneuron Mi1, which is known for its major contribution to the detection of spatial luminance contrast and visual motion. In summary, Dm8 mediates a second step in UV/VIS color opponent processing in Drosophila by integrating input from all types of photoreceptors. These results demonstrate novel insights into the circuit integration of R1-R6 into color opponent processing and reveal that chromatic and achromatic circuitries of the fly visual system interact more extensively than previously thought (Pagni, 2021).

The accurate processing of contrast is the basis for all visually guided behaviors. Visual scenes with rapidly changing illumination challenge contrast computation because photoreceptor adaptation is not fast enough to compensate for such changes. Yet, human perception of contrast is stable even when the visual environment is quickly changing, suggesting rapid post receptor luminance gain control. Similarly, in the fruit fly Drosophila, such gain control leads to luminance invariant behavior for moving OFF stimuli. This study shows that behavioral responses to moving ON stimuli also utilize a luminance gain, and that ON-motion guided behavior depends on inputs from three first-order interneurons L1, L2, and L3. Each of these neurons encodes contrast and luminance differently and distributes information asymmetrically across both ON and OFF contrast-selective pathways. Behavioral responses to both ON and OFF stimuli rely on a luminance-based correction provided by L1 and L3, wherein L1 supports contrast computation linearly, and L3 non-linearly amplifies dim stimuli. Therefore, L1, L2, and L3 are not specific inputs to ON and OFF pathways but the lamina serves as a separate processing layer that distributes distinct luminance and contrast information across ON and OFF pathways to support behavior in varying conditions (Ketkar, 2022).

The present study establishes that contrast and luminance are basic visual features that interact with both ON and OFF pathways. In both pathways, the interaction between these features enables stable visual behaviors across changing conditions. The lamina neurons L1, L2, and L3 act as the circuit elements segregating both contrast and luminance information. Behavioral experiments show that luminance-sensitive input neurons scale behavioral responses to contrast in both ON and OFF pathways. While L1 and L2 provide distinct contrast inputs, L1 also encodes luminance, together with L3. Whereas L3 activity non-linearly increases with decreasing luminance, L1 shows a linear relationship with luminance. Input from both luminance-sensitive neurons is differently used in ON and OFF pathways. Thus, L1, L2, and L3 are not ON or OFF pathways specific inputs, but they instead distribute the two most basic visual features, contrast and luminance, across pathways to enable behaviorally relevant computations (Ketkar, 2022).

Changing visual environments impose a common challenge onto the encoding of both ON and OFF contrasts, namely the contrasts are underestimated in sudden dim light. This work shows that visual behaviors guided by both ON and OFF pathways approach luminance invariance and are not susceptible to underestimation of contrast in sudden dim conditions. Similarly, luminance invariance has been shown in human perception of both ON and OFF contrasts, and in neural responses in cat LGN at fast time scales. This argues that the implementation of a rapid luminance gain is a common feature of all visual systems, which is relevant for any species that relies on visual information for its survival in changing visual environments. In Drosophila, luminance information from both L1 and L3 are required for rapid luminance gain control, but the impact of the two neurons on behavior is pathway dependent. In the OFF pathway, losing either L1 or L3 function leads to a strong deviation from luminance invariance, such that the dim light stimuli are underestimated. On the contrary, ON motion-driven behavior only underestimates dim stimuli if both L1 and L3 neuron types are not functional. Furthermore, L2 neurons, which were formerly thought to be OFF-pathway inputs, contribute contrast-sensitive information to ON behavior. Notably, ON and OFF contrast constancy is not achieved symmetrically at every processing stage. For example, in the vertebrate retina, ON RGCs encode a mixture of luminance-invariant and absolute (i.e. luminance-dependent) contrast, whereas OFF RGCs encode predominantly absolute contrast. Thus, asymmetrical implementation of contrast-corrective mechanisms can be common across visual systems, too (Ketkar, 2022).

Input from the three lamina neurons is differentially utilized across ON and OFF pathways. How does this fit with the established notion that L1 is an input to the ON and L2 and L3 are inputs to OFF pathways? The luminance-varying stimuli sets used in this study were able to pull out lamina neuron contributions that were not obvious with simpler stimuli. For example, the data show that L1 and L2 provide redundant contrast input to the ON pathway at 100% contrast and varying luminance. However, L1 is still strictly required for ON responses if different contrasts are mixed. This is consistent with a more complex ON-pathway input architecture and hints at a role for the L1 pathway in contrast adaptation. Interestingly, Mi1, an important post-synaptic partner of L1, shows an almost instantaneous and strong contrast adaptation (Ketkar, 2022).

While all three lamina neuron types hyperpolarize to light onset and depolarize to light offset, contrast selectivity emerges downstream of these neurons: post-synaptic partners of L1 acquire ON contrast selectivity due to inhibitory glutamatergic synapses, whereas cholinergic L2 and L3 synapses retain OFF contrast selectivity. L3 had furthermore mostly been considered an OFF-pathway neuron because the OFF-pathway neuron Tm9 receives its strongest input from L3. However, L3 itself actually makes most synaptic connections with the Mi9 neuron that plays a role in guiding behavioral responses to ON stimuli. Further synapses of L3 with the ON-selective Mi1 neuron are similar in number to those with Tm9. Finally, L3 can potentially also convey information to the chromatic pathway, as Tm20 is its second strongest postsynaptic connection. There, L3 luminance sensitivity might play a relevant role in achieving color constancy, that is color recognition irrespective of illumination conditions. Altogether, anatomical and functional data indicate that it is time to redefine L3 as part of a luminance-encoding system rather than a mere OFF-pathway input. Other synaptic connections that link L2 to downstream ON-selective neurons still have to be investigated in detail (Ketkar, 2022).

A role of L1 beyond the ON pathway is supported by functional connectivity studies showing that Tm9 properties rely in part on L1 input, and that Tm9 together with other OFF-pathway interneurons displays contrast-opponent receptive fields, showing the presence of ON information in the OFF pathway. Connectomics data did not identify any known OFF-pathway neurons postsynaptic to L1, but among the strongest postsynaptic partners of L1 are the GABAergic interneurons C2 and C3 that connect to the OFF pathway. Intercolumnar neurons downstream of L1, such as Dm neurons, could further carry information to OFF-selective neurons, likely through disinhibition from ON-selective inputs. In the vertebrate retina, intercolumnar amacrine cells mediate interaction between ON and OFF bipolar cells, which has been shown to extend the operating range of the OFF pathway (Ketkar, 2022).

Altogether, it now becomes evident that a split in ON and OFF circuitry only truly exists in downstream medulla neurons and direction-selective cells. The luminance and contrast features encoded differently in L1, L2 and L3 lamina neurons are shared by both pathways. Importantly, the distinct features that are passed on by the specific inputs downstream of photoreceptors guide distinct behavioral roles (Ketkar, 2022).

Despite being postsynaptic to the same photoreceptor input, L1, L2, and L3 all show different contrast and luminance sensitivities. L1 was previously considered the ON-pathway sibling of the contrast-sensitive L2, both with regard to its temporal filtering properties and at the transcriptome level. However, L1 calcium signals show a transient and a sustained response component, which are contrast- and luminance-sensitive, respectively. Compared to photoreceptors, which also carry both contrast and luminance components, L1 still amplifies the contrast signals received from the photoreceptors, since its transient component is more pronounced than the one seen in the photoreceptor calcium traces. In other insect species, different types of lamina neurons have also been distinguished based on their physiological properties, although their specific luminance and contrast sensitivities are yet unknown (Ketkar, 2022).

The two luminance-sensitive neurons L1 and L3 differ in their luminance-encoding properties. L1's initial transient contrast response might reduce the operating range of the subsequent luminance-sensitive baseline. L3's calcium responses show little adaptation and can utilize most of its operating range to encode luminance. L3 seems to invest this wider operating range into amplifying the darkest luminance values selectively and non-linearly. Thus, a predominantly luminance-sensitive channel among LMCs may have evolved to selectively process stimuli in the low luminance range. The different linear and non-linear properties of L1 and L3 might further increase the dynamic range of luminance signaling. Together with the pure contrast sensitivity of L2, the first-order interneurons in flies exhibit a wide range of sensitivities with respect to contrast and luminance, and different functional relevance. Diversifying feature encoding through distinct temporal properties of first-order interneurons is a strategy employed to reliably handle wide luminance ranges (Ketkar, 2022).

In flies, three first-order interneurons feed contrast and luminance information into downstream circuitry. In the mouse retina, more than 30 functionally distinct bipolar types show a spectrum of temporal filter properties rather than a strict transient-sustained dichotomy, thus capturing a larger diversity of temporal information in parallel channels. Many bipolar cell types resemble L1, in that they have both luminance and contrast signals in distinct response components. However, the degree of transiency varies from cell type to cell type, and some predominantly sustained bipolar cell types are also found, closely resembling the luminance-sensitive L3. Such diversification of feature extraction at the periphery has been shown to be computationally advantageous, especially when processing complex natural scenes. For example, during daylight, visual scenes can differ in intensity by 4-5 log units, whereas electrical signals in cone photoreceptors reach a dynamic range of only two orders of magnitude (Ketkar, 2022).

Although the vertebrate retina apparently has a much larger diversity of cell types to handle the wide and complex statistics of the visual environments, there is only a single layer of processing between photoreceptors and the first direction-selective cells, whereas in insects, there are two: the lamina and the medulla. It seems as if the combined properties of bipolar cells are spread across these two processing stages in the fly visual system: whereas some properties, such as diversity of temporal filtering starts in LMCs, contrast selectivity only emerges in medulla neurons and not directly in the first-order interneurons as it happens in bipolar cells. In both vertebrates and invertebrates, the emergence of ON selectivity occurs through inhibitory glutamatergic synapses, but whereas this happens at the photoreceptor-to-bipolar cell synapse in vertebrates, it happens one synapse further down between lamina and medulla neurons in flies. Taken together, LMCs and downstream medulla neurons combined appear to be the functional equivalents of vertebrate bipolar cell layers. Given the size limitations of the fly visual system to encode the same complex environment effectively, one benefit of this configuration with an extra layer could be that it allows more combinations. Furthermore, the photoreceptor-to-lamina synapse in the fly superposition eye already serves to spatially pool information from different photoreceptors. In both visual systems, diversifying distinct information across several neurons could serve as a strategy to reliably respond to contrast when luminance conditions vary (Ketkar, 2022).

The direction-selective T4/T5 cells innervate optic-flow processing projection neurons in the lobula plate of the fly that mediate the visual control of locomotion. In the lobula, visual projection neurons coordinate complex behavioral responses to visual features, however, the input circuitry and computations that bestow their feature-detecting properties are less clear. A highly specialized small object motion detector, LC11, was studied, and its responses were shown to be suppressed by local background motion. LC11 expresses GABA-A receptors that serve to sculpt responses to small objects but are not responsible for the rejection of background motion. Instead, LC11 is innervated by columnar T2 and T3 neurons that are themselves highly sensitive to small static or moving objects, insensitive to wide-field motion and, unlike T4/T5, respond to both ON and OFF luminance steps (Keles, 2020).

The cellular mechanisms of motion vision have become rapidly advanced owing to genetic, optogenetic, and in vivo imaging tools developed in Drosophila melanogaster. As in the mammalian retina, the fly optic lobe segregates ON and OFF polarity luminance changes into parallel cellular pathways. The ON- and OFF-selective pathways supply directionally selective columnar T4 and T5 neurons, respectively. The terminals of these small-field retinotopic motion detectors innervate the third optic ganglion, the lobula plate, where their synaptic output is integrated within the large planar dendrites of projection neurons that map specific wide-field patterns of optic flow onto descending pre-motor neurons to coordinate visual behavior (Keles, 2020).

In parallel to the motion vision pathway of the lobula plate, projection neurons identified in the lobula have been shown to encode moving features such as edges or objects to influence complex visual behaviors. Roughly 20 classes of lobula columnar neurons (LCs) project to the protocerebrum where axon terminals of each class form tight glomerular neuropils. Individual LC11 neurons as well as the glomerular ensemble are highly responsive to small contrasting objects moving in any direction across the ipsilateral field of view. Unlike the output cell types of the lobula plate, little is known about how the receptive field properties of LC11 arise. This study investigated the interactions between background motion and object responses in LC11, has identified a role for GABA-mediated inhibition in shaping object detection by LC11, and identifies presynaptic inputs to LC11. Columnar neurons T2 and T3 projecting from the medulla and terminating in the second and third layers of the lobula overlap with LC11 dendrites. T2 and T3 synapse with dendrites of LC11, and T3 supplies excitatory input to LC11. Finally, it was demonstrated that T2 and T3 neurons are highly selective for small objects, are suppressed by wide-field background motion, and unlike T4/T5, show full-wave rectified ON-OFF excitatory responses to rapid transitions in luminance (Keles, 2020).

In vertebrates, neurons in the retina partially encode object information, but fail to discriminate flicker from coherent motion. Yet, higher-order neurons in the mouse superior colliculus respond strongly only to moving stimuli. Similarly, this study found that T2/T3 neurons are selective for small objects, but respond to ON and OFF flicker as well, whereas downstream LC11 is responsive to object motion, not stationary flicker. It is proposed that LC11 computes continuous object motion from local ON-OFF transients conveyed by T2/T3. Future work on examining the spatiotemporal patterning of columnar inputs to LC11, as well as the cognate neurotransmitters and receptors should reveal how these computations are achieved (Keles, 2020).

In prior work (Keles, 2017), it was demonstrated that bath applied PTX, which selectively blocks chloride currents carried by GABA-A channels or glutamate channels, resulted in LC11 displaying uncharacteristic responses to elongated bars and gratings. This result was predicted under the presumption that inhibition actively filtered wide-field input from LC11. Curiously, in the same preparation, the small object responses for LC11 were essentially eradicated. How can global loss of inhibition by bath applied PTX explain both enhanced wide-field responses and diminished small object responses in LC11? Several lines of evidence suggest that postsynaptic inhibitory neuromodulation acts on LC11 in a center-surround fashion. LC11 expresses both acetylcholine receptors and a GABA-gated chloride channel subunit Rdl. Blocking GABA-A mediated synaptic currents by genetic disruption of Rdl specifically in LC11 neurons results in a decrease in response amplitude to the smallest object tested, and yet, importantly, had no effect on the normal attenuated responses to bars or normal absence of wide-field grating responses. The results support a working model in which Rdl knockout unmasks an ON-pathway input while decreasing the normal OFF object response of LC11. These properties could be explained by an ON-pathway GABAergic input to LC11 through Rdl that normally occludes ON excitation and disinhibits OFF responses. The corollary is that suppression of responses to large objects or wide-field motion occurs upstream of these object detectors. Indeed, LC11 appears to inherit its sensitivity to small object motion from excitatory T2/T3 inputs, perhaps themselves having surround inhibition mechanisms similar to T4/T5. There appears to be two mechanisms of action that are disrupted by PTX application on LC11 receptive field properties (Keles, 2017): crossover inhibition in T2/T3, which would explain their size-tuning, and local inhibition on LC11 that normally enhances small object responses. Thus, it is proposed that upon PTX delivery abnormal bar and wide-field motion responses are conveyed from T2/T3 to LC11 and small object responses are no longer boosted (Keles, 2020).

The importance of dynamic, stimulus-specific inhibition for spatial vision has been elucidated by other studies. In mice, cortical V1 center-surround receptive fields reveal stronger inhibitory currents than excitatory currents in both the surround and center, while inhibitory currents are spread more laterally than excitatory currents. In a visual collision detection circuit in the locust, feedforward inhibitory neurons actively encode dynamical variables such as object angular size. The inhibitory GABA-A receptor subunit Rdl is expressed by nearly all neurons of the fly visual system so far tested, highlighting the ubiquity and importance of inhibition for spatial vision (Keles, 2020).

T2 and T3 neurons share several key features with LC11. First, both show significantly larger responses to small solid objects than to single object edges or elongated bars, with virtually no response to moving wide-field gratings. In the large calliphorid fly Phaenicia sericata, T2 neurons have been examined with intracellular sharp electrodes, which showed that these columnar neurons depolarize to the OFF-phase of flicker, and hyperpolarize to the ON-phase. This contrasts to the GCaMP6f recordings in Drosophila, in which T2 is excited by both ON and OFF luminance transitions. Additionally, in Phaenicia T2 responded robustly to 80 x 62° moving gratings, whereas in Drosophila no response was observed in either T2 or T3 to gratings that filled the 108 x 63° display. The T2a cell type, with similar anatomy but different presynaptic inputs to T2, may show responses more closely matching those from larger flies (Keles, 2020).

An important feature of Drosophila T2/T3 neurons is that unlike T4 and T5 columnar motion detectors, which act as half-wave rectifiers that segregate ON and OFF edge stimuli, respectively, both T2 and T3 neurons show full-wave rectification in that they are excited by both ON and OFF phases of flicker. Notably, T5 shows similar amplitude responses to the OFF edges generated by either a solid two-edged dark object or a single moving OFF edge, whereas T3 responses are markedly stronger for the solid object presenting an OFF-ON sequence than to a single progressing OFF edge. T3 appears to receive input from a combination of neurons that reside in the ON and OFF pathways, including Mi1 and Tm3, providing a possible explanation for this result. Full-wave rectification of ON and OFF stimuli is consistent with single point correlation computations proposed to comprise elementary small target motion detectors (ESTMDs), which underlie the high performance object detection seen in lobula wide-field STMD neurons of hoverflies and dragonflies. Future work must explore the mechanisms that shape responses in T2 and T3, and how the spatiotemporal patterns of input from T2 and T3 confer discrimination of object motion from flicker in LC11 (Keles, 2020).

A neuron is conventionally regarded as a single processing unit. It receives input from one or several presynaptic cells, transforms these signals, and transmits one output signal to its postsynaptic partners. Exceptions exist: amacrine cells in the mammalian retina or interneurons in the locust mesothoracic ganglion are thought to represent many electrically isolated microcircuits within one neuron. An extreme case of such an amacrine cell has recently been described in the Drosophila visual system. This cell, called CT1, reaches into two neuropils of the optic lobe, where it visits each of 700 repetitive columns, thereby covering the whole visual field. Due to its unusual morphology, CT1 has been suspected to perform local computations, but this has never been proven. Using 2-photon calcium imaging and visual stimulation, this study found highly compartmentalized retinotopic response properties in neighboring terminals of CT1, with each terminal acting as an independent functional unit. Model simulations demonstrate that this extreme case of compartmentalization is at the biophysical limit of neural computation (Meier, 2019).

Understanding the circuit mechanisms behind motion detection is a long-standing question in visual neuroscience. In Drosophila melanogaster, recently discovered synapse-level connectomes in the optic lobe, particularly in ON-pathway (T4) receptive-field circuits, in concert with physiological studies, suggest a motion model that is increasingly intricate when compared with the ubiquitous Hassenstein-Reichardt model. By contrast, knowledge of OFF-pathway (T5) has been incomplete. This study presents a conclusive and comprehensive connectome that, for the first time, integrates detailed connectivity information for inputs to both the T4 and T5 pathways in a single EM dataset covering the entire optic lobe. With novel reconstruction methods using automated synapse prediction suited to such a large connectome, previous findings in the T4 pathway were successfully corroborated, and inputs and receptive fields for T5 were comprehensively identified. Although the two pathways are probably evolutionarily linked and exhibit many similarities, interesting differences and interactions were uncovered that may underlie their distinct functional properties (Shinomiya, 2019).

Over half a century ago, Hassenstein and Reichardt working on the beetle Chlorophanus, and later Reichardt working on flies and studying rabbit retinal ganglion cells, all independently presented evidence for motion detection circuits that incorporate a delay-and-compare strategy. In both insect and mammalian model groups, two or more independent, parallel inputs from upstream neurons provide input to elementary motion detector (EMD) circuits. Both models use a similar mechanism to compute the direction of motion, but they differ depending on how they produce a direction-selective response. The Barlow-Levick type circuit detects the preferred-direction signals by suppressing signals in the non-preferred direction; the Hassenstein-Reichardt detector generates an enhancement of signals in the preferred direction (Shinomiya, 2019 and references therein).

The fly's optic lobe consists of four consecutive neuropils: the lamina, medulla, lobula, and lobula plate. Each of these comprises columnar units that correspond to the array of ommatidia in the retina. The motion pathway in the optic lobe arises from the photoreceptor cells (PRs), which receive light signals in the compound eye and extend their axons to the lamina. R1-R6 cells expressing rhodopsin rh1 provide signals to lamina monopolar cells in the lamina cartridges, which project to the distal medulla. The lamina neurons are presynaptic to various types of medulla neurons in the distal medulla. Among them, medulla columnar neurons including Mi, Tm, and TmY cells further provide inputs to the dendritic arbors of T4 in the M10 layer of the medulla and T5 in the Lo1 layer of the lobula (Shinomiya, 2019).

The dendritic arbors of T4 cells receive parallel inputs from multiple columns, and a single arbor receives inputs from columns that signal different positions of the visual field, depending on the cell types of the input neurons. Recent developments in techniques for three-dimensional electron microscopy (3D-EM) have accelerated the identification of neurons and their synaptic circuits, or their connectome, in the brain of the fruit fly Drosophila melanogaster. In the visual system, motion detection pathways in the optic lobe have been a prominent goal for such connectomic approaches, which identify the component neurons using 3D-EM reconstructions of their arbors (Shinomiya, 2019).

The medulla dendritic arbors of T4 cells provide a substrate for the elementary motion detector (EMD) in the ON-edge motion pathway. Using serial-section transmission EM (ssTEM), Mi1 and Tm3 as major inputs to the T4 cell dendrites. A later approach using focused ion beam scanning EM (FIB-SEM) comprehensively revealed other medulla neurons providing inputs to T4. These medulla neurons relay input to T4 from L1, the first of two repeated neuron classes in the first neuropil, or lamina; L1 in turn receives input from the terminals of photoreceptors R1-R6 in the overlying compound eye (Shinomiya, 2019).

Complementary to the T4 cells, narrow-field T5 cells constitute the first output stage of the OFF-edge pathway, and some of T5's input neurons have also been identified from their terminals reconstructed using ssTEM. These inputs relay signals from L2 cells, which partner L1 in all columns, or cartridges, of the lamina and which also receive input from R1-R6. Therefore, the separation between the ON and OFF motion pathways is already established at the level of the lamina neurons (Shinomiya, 2019).

Finally, T4 and T5 cell axons transfer motion information to the fourth neuropil, or lobula plate, where it is integrated and further processed to extract specific motion modalities, before being conducted to the central brain by visual projection neurons (VPNs). VPNs include various types of lobula plate tangential neurons (LPTCs) and lobula plate/lobula columnar cells (Shinomiya, 2019).

The ON and OFF motion pathways are similar in their function, component neurons, and patterns of synaptic connections. Both T4 and T5 cells are direction-selective neurons, and each is further grouped into four subtypes: T4 as T4a, T4b, T4c and T4d; and T5 as T5a, T5b, T5c, and T5d. These T4 and T5 cells specifically signal motion in the four canonical directions. The subtypes a-d detect front-to-back, back-to-front, upward, and downward motion, respectively. Each subtype projects its axon to one of the lobula plate's four strata, depending on the direction of motion that it signals. Developmentally, both T4 and T5 are known to originate from the same subset of progenitor cells in the inner proliferation center and to express a proneural gene, Atonal, uniformly (Shinomiya, 2019).

Given the dimensional constraints of the respective ssTEM and FIB-SEM datasets, however, the T4 and T5 pathways, and their respective input neurons, have been reconstructed independently in separate reports using 3D-EM methods. Series of ultrathin sections have been used to identify medulla cell inputs to T4 cells; these included medulla intrinsic (Mi) and transmedulla (Tm) cells but not their terminals in the lobula, which were lacking from the EM dataset. Similarly, inputs to T5 terminals in the lobula arise from Tm cells, but the medulla arbors of these were also lacking from previous reconstructions. Subsequent reports that repeated the analysis of cells for seven medulla columns, using FIB-SEM, also failed to identify the lobula, but comprehensively identified additional inputs to, and connections between, T4 cells. Consequently, results from these studies cannot be compared directly to the same field size and at the same resolution in a single dataset. This makes it difficult to recognize and resolve deep similarities in the inputs to both pathways, which might support further evolutionary comparisons between those inputs, and which might also enable functional comparisons, especially for the inputs to T5 which to date are known only for four main Tm cells: Tm1, Tm2, Tm4, and Tm9 (Shinomiya, 2019).

This study exhaustively identified the synaptic inputs to T5 cells and described their spatial layouts. The anatomical properties of the dendritic terminals of T4 and T5 were also assessed, after identifying all neurons that have synaptic contacts with the motion-sensing output cells in the medulla and lobula. As a result, this report concludes the connectomic analysis of both the ON- and the OFF- motion-sensitive pathways in Drosophila (Shinomiya, 2019).

Classical correlation models of the motion detection circuit, including the Hassenstein and Reichardt (1956) and Barlow and Levick (1965) models, consider only two independent upstream inputs in the detection of motion. Several studies have provided physiological evidence that the elementary motion detector EMD circuit may be approximated by either of these models or their modified versions. These models cannot sufficiently address the asymmetrical responses of the T4 and T5 pathways, however, because the types and numbers of the neurons involved are limited, and are indeed not consistent with the findings that the input neurons to both T4 and T5 dendrites are clustered into three groups, not two. Previous studies, in particular, failed to include inhibitory inputs from CT1 to the T4 and T5 dendritic arbors. Although CT1 differs from the other medulla neurons providing inputs to T4 or T5 insofar as it lacks a direct synaptic partnership with lamina cells, it still receives indirect inputs from those cells via Mi1 and Mi9 (in M10) and Tm1 and Tm9 (in Lo1). CT1 is the only inhibitory columnar input to the T5 dendrites, and also the only element that is displaced from the other two excitatory legs. In the ON-edge side, CT1, together with the other two GABAergic neurons, C3 and Mi4, provides a measurable input to the base of the T4 dendrites (Shinomiya, 2019).

As a foretaste to human anatomy, a new EMD circuit model with three parallel inputs was recently proposed, based on the two classical motion detection models, on computer simulations, and on activity recordings from T4 and T5 cells. Inputs to the motion-detecting unit include a non-delayed direct input, delayed enhancer input, and a delayed suppressor input (null direction suppression) located on the side opposite to the enhancer signal input, the two inputs bracketing the direct input. The direction from the enhancer signal input to the suppressor signal input corresponds to the preferred direction (PD), and is opposite to the non-preferred direction (ND). This three-way input model incorporates the classical two-inputs models, the Hassenstein and Reichardt (H-R) and Barlow and Levick (B-L) models, as its subsets, so that the outputs, as well as the temporal tuning patterns of the circuits, are still consistent with the previous physiological studies. Applying this model to T4: a) Mi1 and Tm3, which provide inputs to the center of the dendritic arbor, would be direct inputs; b) Mi9 innervating the tip would be an enhancer input; and c) CT1, Mi4, and C3, which innervate the base, would be suppressor inputs. In the case of T5: a) Tm1, Tm2, and Tm4 would be direct inputs; b) Tm9 would be an enhancer, and; c) CT1 would be a suppressor input. Although Tm9 innervates the tip of T5 dendrites and therefore would fall into an enhancer location here, silencing experiments suggested that the contribution of Tm9 to the dark-edge detection was even larger than that of Tm1 and Tm2 combined, so it might not be reasonable to regard Tm9 as an input that simply enhances the main direct inputs (Shinomiya, 2019).

The motion circuit neurons activated by signals from ON-edge motion in either the preferred (A), or non-preferred (B) directions, and by OFF-edge motion in either the preferred (C), or non-preferred. Among these inputs to the T-cells, CT1, C3, and Mi4 are known to be GABA-positive, Mi9 is glutamate-positive, and all other cells, including T4 and T5, are positive for cholinergic reagents. As all neurons providing inputs to the base are putatively GABAergic, and all neurons to the center putatively cholinergic, the connections of the ON- and OFF-edge pathways partially match the three-way EMD circuit model (Shinomiya, 2019).

Both Mi9 and Tm9 cells relay signals from L3, and send inputs to the tip of the T4/T5 dendritic arbors. They both produce slow and sustained responses that serve as low-pass filters. Although Tm9 is putatively cholinergic and provides excitatory signals to T5, putatively glutamatergic signals from Mi9 to T4 are supposed to be inhibitory on the basis of behavioral response assays. This possibility is also suggested by the observation that T4 expresses a glutamate-gated chloride channel, GluClα, which mediates inhibitory signals. Tm9 shows an increased response during OFF stimulation, and Mi9 is also thought to be activated similarly but provides input to the T4 pathway. These neurons may therefore modulate T4 and T5 using independent but opposing respective mechanisms. Mi9 may function as a temporal low-pass filter of the OFF signal and may cancel noise during an OFF stimulus by suppressing T4, whereas Tm9 could enhance OFF signals to T5 along with the direct inputs from Tm1 and Tm2 (see Schematic diagram of inputs onto and outputs from T4 and T5 dendrites). Besides Mi9 and Tm9, the tips of the T4 and T5 dendrites also receive excitatory inputs from T4 and T5 of the same cell type, a-d, which could also enhance signals to the EMD circuits by themselves, although these inputs are fewer in number and presumably would be far weaker than those from Mi9 or Tm9 (Shinomiya, 2019).

CT1, which innervates the base of T4 and T5 dendrites, is an interesting wide-field cell that receives signals from the lamina cell pathways indirectly via other medulla neurons, including Mi1, Mi9, Tm1, and Tm9. The other inputs to the base of T4, Mi4 and C3, likewise lack direct inputs from lamina cells, suggesting that inhibitory signals from CT1, Mi4, and C3 are delayed by an additional synapse relative to those from the direct inputs (Shinomiya, 2019).

The connectivity diagrams are summarized of ON- and OFF-edge EMD circuits that have now been demonstrated anatomically (see Schematic diagram of inputs onto and outputs from T4 and T5 dendrites). A schematized and simplified EMD circuit diagram summarizes three downstream pathways from the photoreceptors (R1-R6). Among the lamina cells, L1 signals to ON pathways selectively, whereas L2 and L3 both signal to OFF pathways. None of the inputs to the base of T4 and T5, that is from CT1, Mi4, and C3, receives direct inputs from lamina interneurons. CT1 terminals in M10 receive only indirect input from L1 and L3, via Mi1 and Mi9, whereas those in Lo1 receive information from L2 and L3 only via Tm1 and Tm9. Mi4 also receives inputs from L1 and L3, through L5 and Mi9. The transfer of information to these neurons may be delayed at additional synaptic relays. Mi4, Mi9, and Tm9 themselves show delayed and sustained calcium responses against white-noise stimuli when compared with the responses of Mi1, Tm3, Tm1, Tm2, and Tm4, suggesting that the responses of Mi4, Mi9, and Tm9 serve as delayed arms in the EMD model. The response properties of CT1 are still unknown. It is suggested that the anatomical pathways of the two EMD circuits have now been reported in sufficient detail in this and previous accounts, but their physiological correlates, including the neurotransmitters and receptors of some constituent neurons, as well as temporal delays at each cell, all still need further analysis to complete the picture of how the T-cells signal motion information (Shinomiya, 2019).

On the basis of the described neuronal connectivity, speculative timing diagrams against ON- and OFF-edge signals to motion in the preferred and non-preferred directions are also shown. Among the lamina cells, only L1 activates the downstream cells during ON-edge signals. The direct inputs (Mi1 and Tm3) and suppressor inputs (CT1 and Mi4) to the T4 dendrites may therefore contribute to detecting ON-edge signals at the level of the EMD circuit. During OFF-edge signals, on the other hand, all three input legs to T5 as well as Mi9, which provides inhibitory inputs to T4, are activated. In both ON- and OFF-edge circuits, excitatory inputs provide signals to T4 or T5 cells first, before delayed inhibitory inputs suppress the responses of these cells to stimuli in the preferred direction. For signals in the non-preferred direction, T4 and T5 cells are inhibited by the suppressor inputs and will not be excited. Besides these, T4 is also likely to be suppressed by Mi9 not during ON-edge signals but during OFF-edge signals, presumably cutting off spontaneous noise from responses to non-preferred stimuli. Such a mechanism is lacking in the T5 pathway (Shinomiya, 2019).

Inputs to T4 and T5 from their upstream neurons reveal significant anatomical similarities between the ON- and OFF-edge EMD circuits. The Mi1 cell in the ON pathway, for example, is similar to the Tm1 and Tm2 cells in the OFF pathway, insofar as all these cells provide inputs to the center of the T4/T5 cell dendrites and use acetylcholine as their neurotransmitter, and differ only in the neuropil of their termination. Like T4, T5 expresses transcripts of two different nicotinic cholinoceptors, as well as those of an A-type muscarinic cholinoceptor, suggesting that T5 receives cholinergic Tm inputs by means of both ionotropic and metabotropic cholinoceptors. Additional neurons, CT1 and TmY15, innervate both M10 and Lo1 and form synapses there with T4 and T5, respectively. On the other hand, three types of GABA-positive neurons provide inputs to the base of T4 dendrites, whereas only one type of GABAergic neuron (CT1) sends inputs to the base of T5. Inputs from CT1 make up a much larger proportion of the total input to T5 cells than to T4 cells. This difference might compensate for the lack of other inhibitory inputs to the base of T5. There are also two tangential elements in Lo1 (LT33 and Tm23) that make synapses with T5 dendrites, and counterparts are not found in M10. These then constitute differences betweeen the inputs to T5 and T4 cells. As the cell body site and projection trajectory of LT33 are both similar to those of CT1, a GABAergic cell, and as the cell is both presynaptic and postsynaptic to the T5 cell dendrites, it is possible that LT33 is also inhibitory and inhibits the activity of T5 regardless of the direction specificity. Even so, the T4 and T5 motion stimulus responses are very similar (Shinomiya, 2019).

The T4 and T5 cells not only share functional characteristics, but also correlate closely in their development and are suggested evolutionary siblings. T4 and T5 are produced from the same lobula plate neuroblasts, and the expression of Notch specifies the generation of these two morphologically similar cell types. Assuming that, during the course of evolution, T4 and T5 arose as duplicates from an ancestral cell population, as has been proposed, the neurons that provide their synaptic partnerships, most notably Mi and Tm cells, could also have been duplicated, possibly in an event that was induced by the duplication of T4/T5 (Shinomiya, 2019).

T4 and T5, or morphologically similar optic lobe neurons, have been found in a broad range of arthropod species, including various Diptera, the honeybee, butterfly, and crab, suggesting that the origin of these cells can be traced back to the Cambrian, when pancrustacean ancestors are thought to have given rise to hexapod and crustacean species. Anatomical and functional differences in the T4 and T5 pathways of the fly's brain may therefore have accumulated during the course of evolution, adapting ancestral forms to their living environments, by changes in the synaptic connections of their partner neurons (Shinomiya, 2019).

The two novel cells described in this report, CT1 and TmY15, share important similarities: both are putatively inhibitory, and both provide input to both T4 and T5. They differ chiefly in their anatomical field size. TmY15 is narrow-field, spanning at least 10 columns, whereas CT1 spans the entire field. The architecture of the connecting networks also differs. Thus, both are anatomically qualified to provide an inhibitory surround to the field of T-cells that they innervate (Shinomiya, 2019).

In addition to its similarity to TmY15, CT1 may support a local computation of the inhibitory elements of a Barlow-Levick circuit. Calculations of the space constant for the CT1 arbor suggest that a delay between adjacent columns is sufficient to allow local inhibition, in addition to any global inhibition that this cell may mediate. Such local computation has both the right sign and location (on the leading edge of the anti-preferred direction) in both the ON-pathway of T4 in the medulla, where CT1 is excited by Mi1 and Tm3, and the OFF-pathway of T5 in the lobula, where the inputs are instead Tm1 and Tm9. The latter, in turn, derive their inputs from L1 (to T4) and L2 (to T5) (Shinomiya, 2019).

The functional significance of TmY15 in motion processing must remain speculative. Not only does this cell receive a wide range of inputs from various types of cells in the medulla, lobula, and lobula plate, but it also has much weaker synaptic contacts with T4 and T5 in M10 and Lo1 when compared with other Mi or Tm cell inputs. It specifically innervates the second and third strata (Lop2 and Lop3) in the lobula plate, and preliminary observations show that it receives inputs from cells T4b, T4c, T5b, and T5c, suggesting that TmY15 may also work as a feedback loop to suppress responses in T4 and T5 during regressive and upward motions (Shinomiya, 2019).

In summary, this report comprehensively identifies input and output neurons of the dendritic arbors of T4 and T5 cells, and uses a single dataset to reveal (at synapse level) the detailed similarities between the connections of these two motion-signaling output cells. Together with the functional contribution of individual neurons in the motion-detection circuits shown in several studies, the detailed connectivity diagram that this study provides should further facilitate functional analyses in these cells, through behavioral assays, calcium imaging, and electrophysiological recordings, and by providing comparisons with known neurons in the ON-pathway (Shinomiya, 2019).

Understanding how complex brain wiring is produced during development is a daunting challenge. In Drosophila, information from 800 retinal ommatidia is processed in distinct brain neuropiles, each subdivided into 800 matching retinotopic columns. The lobula plate comprises four T4 and four T5 neuronal subtypes. T4 neurons respond to bright edge motion, whereas T5 neurons respond to dark edge motion. Each is tuned to motion in one of the four cardinal directions, effectively establishing eight concurrent retinotopic maps to support wide-field motion. A mode of neurogenesis was discovered where two sequential Notch-dependent divisions of either a horizontal or a vertical progenitor produce matching sets of two T4 and two T5 neurons retinotopically coincident with pairwise opposite direction selectivity. Retinotopy is shown to be an emergent characteristic of this neurogenic program and derives directly from neuronal birth order. This work illustrates how simple developmental rules can implement complex neural organization (Pinto-Teixeira, 2018).

The retinotopic organization of the fly visual system is crucial for circuit function, as exemplified by motion detection circuits. Within the optic lobe, visual motion information is processed in two parallel pathways: the ON pathway detecting bright edge motion and the OFF pathway that processes dark edge motion. The two pathways bifurcate early since distinct lamina neurons, the first to make contact with photoreceptors, connect to different sets of medulla neurons, which themselves then synapse with dendrites of T4 neurons (ON) in the medulla or T5 neurons (OFF) in the lobula. T4 and T5 neurons are the first neurons in each pathway that are direction selective. They process the visual signal originating from one main column and integrate it with ~7 neighboring columns to compute local motion (Pinto-Teixeira, 2018).

Both T4 and T5 neurons exist in four subtypes (termed a, b, c, and d) directionally tuned to one of the four cardinal directions (front-to-back, back-to-front, upward, and downward). Thus, for each column, four T4 and four T5 neurons, one of each subtype, represent eight independent motion detectors. T4 (ON) and T5 (OFF) neurons with the same directional tuning project retinotopically to one of the four layers of the lobula plate that is organized into two layers for horizontal motion (layer a, front-to-back; layer b, back-to-front) and two layers for vertical motion (layer c, upward; layer d, downward). Within each layer, T4 and T5 neurons synapse with the dendrites of lobula plate tangential cells that integrate the retinotopic local motion signals from T4 and T5 neurons to produce direction-selective wide-field motion responses. Thus, the retinotopy of the T4/T5 circuit is crucial for detecting broad field motion: at the level of T4 (medulla) and T5 (lobula) dendrites, where the retinotopic organization of the inputs onto T4/T5 dendrites allows direction selectivity to first emerge, and at the level of their axons, where retinotopic organization allows for efficient, selective coding of specific global motion patterns. It is therefore critical that the correct number of each T4 and T5 neuronal subtype be produced so that each medulla column is innervated by the four T4 neuronal subtypes and each lobula column by the four T5 subtypes. Furthermore, all eight subtypes of T4 and T5 neurons must project retinotopically to individual layers of the lobula plate (Pinto-Teixeira, 2018).

The four neuropiles of the optic lobes develop during the larval and the early pupal stages from two crescent-shaped neuroepithelial domains: the outer proliferation center (OPC), which produces neurons of the lamina and medulla, and the inner proliferation center (IPC), which generates neurons of the lobula and lobula plate. The IPC crescent is localized between the OPC and the developing central brain. It is divided into three domains: the surface IPC (sIPC) marked by Wingless expression (which will not be discussed further in this work) that is attached to the proximal IPC (pIPC), and a distal domain (dIPC). T4/T5 neurons are produced by progenitors that originate from the pIPC. Unlike the OPC neuroepithelium that is sequentially converted by a proneural wave into neuroprogenitors (neuroblasts (NBs) in the fly), the pIPC neuroepithelium produces Dichaete+ migrating progenitors that move distally to generate the dIPC. Once migrating progenitors reach the dIPC, they acquire a NB identity and divide to produce neurons (Pinto-Teixeira, 2018).

dIPC NBs progress through two temporal windows. First, Dichaete+ NBs divide to self-renew and produce the distal C2, C3, T2, and T3 neurons (C/T neurons) to the outside of the dIPC crescent. In the second temporal window, these NBs express Atonal (Ato) and Dachshund (Dac) and produce T4 and T5 neurons to the inside of the crescent (Pinto-Teixeira, 2018).

This study investigated the developmental program that establishes the identity of the four T4 and four T5 neuronal subtypes and how this program leads to their eight coincident retinotopic maps. A causal link was identified between a mode of neurogenesis and retinotopy in which a single NB produces two ON and two OFF neurons with opposite motion direction selectivity (along the horizontal or the vertical axis) that innervate a single column in three neuropiles. It was also shown that vertical and horizontal T4/T5 motion detectors are produced by different NBs distinguished by Decapentaplegic (Dpp) activity. It is concluded that retinotopy results from the features of this neurogenic program, which depends on neuronal birth order and a unique mode of NB division to pattern a complex and highly organized neural network. Thus, simple developmental rules can generate a complex neural organization across three neuropils of the optic lobes (Pinto-Teixeira, 2018).

As neurons are produced and their identities are specified, they must be precisely incorporated into neuronal circuits. Understanding how neurons are specified, how the developing brain orchestrates the correct targeting of a myriad of individual neurons, and in which way these two developmental processes are related, are difficult problems to solve. These were addressed by studying how each of the eight T4/T5 neuronal subtypes is specified and how their eight retinotopic maps are precisely established. Typically, NBs change their transcription factor identity at each division. Neuronal progeny inherit this identity through an intermediate GMC to dictate their fate. This study identified a mode of neurogenesis that relies on two consecutive Notch binary cell-fate decisions to produce four distinct T4/T5 neurons from a single NB temporal window. Because T4/T5 neurons with opposite motion direction selectivity for one retinotopic position are produced by a single NB at the same time, these four neurons innervate their target neuropiles synchronously, connecting with the same, newly produced target column to establish retinotopy. If each of the four T4 and T5 neurons were produced independently, synchronization of their projection patterns between three neuropiles could be much more difficult to achieve. This would require the establishment of a deterministic spatiotemporal molecular code, such that each column would use a unique molecular code recognized by all the neurons that are supposed to target it. The stepwise, synchronous production of sibling retinotopic neurons described in this study reduces the target possibilities at each time point since the progeny of one NB always find the newest column produced in the medulla or lobula. The results illustrate how the developmental program that specifies T4/T5 fate meets the functional requirements of the motion circuit by establishing coherent retinotopic maps within horizontal and vertical systems (Pinto-Teixeira, 2018).

Such successive divisions that rely on the reutilization of the Notch pathway are reminiscent of the divisions of Drosophila sensory organ precursors (SOPs). Although these cells are not bona fide NBs, SOPs divide in a Notch-dependent manner multiple times to first produce two distinct cells (pIIa and pIIb) that divide once (pIIa) or twice (pIIb) more to give rise to the full complement of cells that form the sensory organ, only one of which is a neuron. In olfactory sensilla, a similar precursor also appears to divide several times in a Notch-dependent fashion to produce up to four olfactory neurons, as well as sensilla cells. In this case, some of the four progeny die, producing 1, 2, 3, or, in some rare cases, 4 neurons per sensillum (Pinto-Teixeira, 2018).

In the case of T4/T5 neurogenesis, this study has demonstrated that Notch signaling is used in two consecutive divisions: after the final NB division, the Notch target in one of the two GMCs is E(spl)mγ (but not Hey), while in the GMC division, Hey [but not E(spl)mγ] marks only one of the early born neurons (T5). How Notch differs in these distinct contexts and how such precise temporal control is established is not known. However, the observation that Notch signaling activates different reporters in different contexts and cell types supports the notion that differential transcriptional programs are activated in different cell types. Furthermore, Notch activity is rather transient, which helps explain how Notch signaling instructs different gene expression programs at each round of division (Pinto-Teixeira, 2018).

A recent preprint on a similar topic is in line with the findings on the role of Dpp and Notch in the specification of the eight T4/T5 subtypes and shows that both Dac and Ato are required for the transition between neuroblasts competence states in the dIPC and for the switch to T4/T5 neuron formation. Another upcoming report (Mora, 2018) addresses the role of the temporal transition from Ase+ to Ato+ in dIPC neuroblasts and shows how Ato expression is required for subsequent neuronal differentiation of T4/T5 neurons. It further suggests that Ato+ neuroblasts divide symmetrically to self-amplify before producing the T4/T5 progeny. However, the data reported above, including the precise lineage analysis, do not support such an amplification step that would disrupt the stoichiometry of production of the C/T neurons and eight T4/T5 subtypes (Pinto-Teixeira, 2018).

The lineage of the T4/T5 direction-selective neurons suggests how motion circuitry and the optic lobe neuropiles themselves might have evolved. Horizontal and vertical motion selective neurons originate from two distinct pIPC neuroepithelial domains whose identity is established by Dpp signaling. In the absence of Dpp signaling, Brk expression was expanded to the Dpp domains, suggesting that the default status of the neuroepithelium is to express Brk. Horizontal and vertical motion-selective neurons were produced by distinct progenitor pools and both rely on the special type of neurogenesis described above to produce their complement of T4/T5 neurons. The most parsimonious evolutionary history for this developmental program is that the Notch-mediated binary fate decisions that specify layers of the lobula plate with opposite tuning, as well as T4 (moving bright edges) versus T5 (moving dark edges) fate, was implemented before the specification of horizontal and vertical motion-selective subtype identity. The ancestor might have only responded to horizontal motion (Brk) before splitting of the neuroepithelium occurred, allowing the acquisition of vertical motion vision (Dpp), perhaps when the animals developed the capacity for flight (Pinto-Teixeira, 2018).

T4 and T5 neurons share morphological and functional similarities, but also important differences, such as the organization of their dendritic processes in the medulla (T4) versus lobula (T5), where each subtype (a,b,c, and d) must be oriented according to its local motion direction preference. Dpp signaling and the two Notch binary fate decisions establish the specification of the four T4 and four T5 subtypes. Future studies will be required to understand how the dendrites of each subtype are properly organized (Pinto-Teixeira, 2018).

Sensory maps and neural circuits are largely genetically 'hardwired' in Drosophila and are usually activity independent. Despite this developmental rigidity, there is a very limited understanding of how genetic programs drive developmental processes that are able to establish precise neural circuits. This study shows that the neurogenic program that specifies the identity of the eight T4/T5 neuron subtypes is also sufficient to establish the coherent retinotopy that supports global motion perception in the fly. It provides an example of how the establishment of connectivity within a neural circuit can only be fully understood in its developmental context. The existence of a causal link between the genetic program that specifies cell fate and the circuit these cells build provides an example of how a complex hardwired neuronal circuit can be built from simple developmental rules (Pinto-Teixeira, 2018).

The computational organization of sensory systems depends on the diversification of individual cell types with distinct signal-processing capabilities. The Drosophila visual system, for instance, splits information into channels with different temporal properties directly downstream of photoreceptors in the first-order interneurons of the OFF pathway, L2 and L3. However, the biophysical mechanisms that determine this specialization are largely unknown. This study shows that the voltage-gated Ka channels Shaker and Shal contribute to the response properties of the major OFF pathway input L2. L3 calcium response kinetics postsynaptic to photoreceptors resemble the sustained calcium signals of photoreceptors, whereas L2 neurons decay transiently. Based on a cell-type-specific RNA-seq data set and endogenous protein tagging, this study identified Shaker and Shal as the primary candidates to shape L2 responses. Using in vivo two-photon imaging of L2 calcium signals in combination with pharmacological and genetic perturbations of these Ka channels, it was shown that the wild-type Shaker and Shal function is to enhance L2 responses and cell-autonomously sharpen L2 kinetics. These results reveal a role for Ka channels in determining the signal-processing characteristics of a specific cell type in the visual system (Gur, 2019).

Visual objects can be discriminated by static spatial features such as luminance or dynamic features such as relative movement. Flies track a solid dark vertical bar moving on a bright background, a behavioral reaction so strong that for a rigidly tethered fly, the steering trajectory is phase advanced relative to the moving bar, apparently in anticipation of its future position. By contrast, flickering bars that generate no coherent motion, or whose surface texture moves in the direction opposite to the bar generate steering responses that lag behind the stimulus. It remains unclear how the spatial properties of a bar influence behavioral response dynamics. A dark bar defined by its luminance contrast to the uniform background drives a co-directional steering response that is phase-advanced relative to the response to a textured bar defined only by its motion relative to a stationary textured background. The textured bar drives an initial contra-directional turn and phase-locked tracking. The qualitatively distinct response dynamics could indicate parallel visual processing of a luminance versus motion-defined object. Calcium imaging shows that T4/T5 motion detecting neurons are more responsive to a solid dark bar than a motion defined bar. Genetically blocking T4/T5 neurons eliminates the phase-advanced co-directional response to the luminance-defined bar, leaving the orientation response largely intact. It is concluded that T4/T5 neurons mediate a co-directional optomotor response to a luminance defined bar, thereby driving phase-advanced wing kinematics, whereas separate unknown visual pathways elicit the contra-directional orientation response (Keles, 2018).

Many animals rely on vision to detect objects such as conspecifics, predators, and prey. Hypercomplex cells found in feline cortex and small target motion detectors found in dragonfly and hoverfly optic lobes demonstrate robust tuning for small objects, with weak or no response to larger objects or movement of the visual panorama. However, the relationship among anatomical, molecular, and functional properties of object detection circuitry is not understood. This study characterized a specialized object detector in Drosophila, the lobula columnar neuron LC11. By imaging calcium dynamics with two-photon excitation microscopy, it was shown that LC11 responds to the omni-directional movement of a small object darker than the background, with little or no responses to static flicker, vertically elongated bars, or panoramic gratings. LC11 dendrites innervate multiple layers of the lobula, and each dendrite spans enough columns to sample 75 degrees of visual space, yet the area that evokes calcium responses is only 20 degrees wide and shows robust responses to a 2.2 degrees object spanning less than half of one facet of the compound eye. The dendrites of neighboring LC11s encode object motion retinotopically, but the axon terminals fuse into a glomerular structure in the central brain where retinotopy is lost. Blocking inhibitory ionic currents abolishes small object sensitivity and facilitates responses to elongated bars and gratings. These results reveal high-acuity object motion detection in the Drosophila optic lobe (Keles, 2017).

When navigating in their environment, animals use visual motion cues as feedback signals that are elicited by their own motion. Such signals are provided by wide-field neurons sampling motion directions at multiple image points as the animal maneuvers. Each one of these neurons responds selectively to a specific optic flow-field representing the spatial distribution of motion vectors on the retina. This study describes the discovery of a group of local, inhibitory interneurons in the fruit fly Drosophila key for filtering these cues. Using anatomy, molecular characterization, activity manipulation, and physiological recordings, this study demonstrates that these interneurons convey direction-selective inhibition to wide-field neurons with opposite preferred direction and provides evidence for how their connectivity enables the computation required for integrating opposing motions. The results indicate that, rather than sharpening directional selectivity per se, these circuit elements reduce noise by eliminating non-specific responses to complex visual information (Mauss, 2015).

Diverse sensory experiences can result in largely overlapping patterns of activation within sensory circuits yet require fundamentally different behavioral responses. An underlying key operation is the extraction of features relevant for specific behaviors by hierarchical layers of neuronal networks with increasing selectivity. A well-studied example of such feature extraction is the computation of the optic flow associated with self-motion -- that is, the feedback motion cues created by an animal progressing through its environment. Across many animals studied, motion-sensitive neurons covering large receptive fields (those that receive input from cues spanning the visual field) tend to be motion opponent, i.e., are excited by motion along one and inhibited along the opposite direction. However, the functional significance of motion opponency is unclear and has to date not been experimentally challenged. This problem was addressed in Drosophila, which has emerged as a powerful model system to study the mechanisms underlying motion vision (Mauss, 2015).

The Drosophila optic lobe consists of four neuropiles called lamina, medulla, lobula, and lobula plate. Each of these neuropiles is built from about 750 repetitive columns arranged in a retinotopic way. Monopolar L1 and L2 cells, among others, receive photoreceptor input in the lamina and feed into two motion pathways. Within each pathway, the direction of motion is computed separately, with the L1-pathway selectively processing motion of brightness increments (ON) and the L2-pathway motion of brightness decrements (OFF). The outputs of the ON and OFF pathways are represented by arrays of small-field T4 and T5 cells, respectively. Each T4 and T5 cell is tuned to one of four cardinal directions and terminates in one of the four layers of the lobula plate such that opposite directions are represented in adjacent layers (layer 1: front to back; layer 2: back to front; layer 3: upward; layer 4: downward). These directions match the preferred directions of wide-field motion-sensitive tangential cells that extend their dendrites in the respective layers: horizontal system cells with dendrites in layer 1 depolarize during front-to-back motion and hyperpolarize during back-to-front motion, Hx cells in layer 2 exhibit the opposite tuning, and vertical system (VS) cells with dendrites mostly in layer 4 depolarize primarily during downward and hyperpolarize during upward motion. With T4/T5 cells blocked, tangential cells lose all of their motion sensitivity, and flies become completely motion blind. Combining optogenetic stimulation of T4/T5 cells with various pharmacological antagonists, the connections between T4/T5 and tangential cells have recently been characterized as monosynaptic, excitatory, and cholinergic. T4/T5 cells thus account for the depolarization of the tangential cells during preferred direction motion. What remains unclear is the mechanism and functional role of subtracting information about motion in the opposite or null direction (Mauss, 2015).

This study characterize a hitherto unknown class of vertical system lobula plate intrinsic (LPi) neurons and demonstrates how they contribute to motion opponency. First, anatomical and molecular characterization, as well as combined optogenetic stimulation and electrophysiological recordings, reveal that LPi neurons are bi-stratified and inhibit tangential cells in single lobula plate layers via glutamatergic synapses. Second, two-photon calcium imaging demonstrated that LPi neurons are activated in response to motion directions similar to their presumed T4/T5 inputs and opposite to their postsynaptic targets. Third, genetically silencing LPi cell output selectively abolishes null direction inhibitory potential changes in tangential cells. It is therefore concluded that LPi neurons hyperpolarize tangential cells during null direction motion through sign-inverting layer interactions, thus forming the cellular basis of motion opponency in the fly. As a final point, the identification of LPi neurons enabled the long-sought functional relevance of motion opponency to be experimentally addressed. As blocking the activity of LPi neurons renders their postsynaptic wide-field motion-sensitive neurons responsive to a variety of moving patterns, these experiments suggest that motion opponency is essential for flow-field selectivity, thereby improving the ability to reliably estimate self-motion trajectories based on complex visual information (Mauss, 2015).

Motion detection is a fundamental function of all higher visual systems. It is a paradigmatic model for sensory feature extraction since motion information is not explicitly encoded in the single receptor response but has to be computed by downstream neural circuits. Motion detection can be described as a two- stage process: In the first stage, direction-selective signals are generated by correlating the output from neighboring photoreceptors after asymmetric temporal filtering. Neural substrates corresponding to these correlators are, for instance, the T4/T5 cells of the fly optic lobes and the dendrites of starburst amacrine cells in the mammalian retina. In the second stage, signals from oppositely tuned correlators are subtracted from each other, giving rise to a fully opponent output. This processing step is implemented in the fly optic lobe on the dendrites of the lobula plate tangential cells, which receive two kinds of inputs: (1) a direct excitatory input from T4/T5 cells terminating within the same lobula plate layer, giving rise to depolarization during preferred direction motion; and (2) as shown in this study, an indirect inhibitory input via bi-stratified LPi neurons from T4/T5 cells terminating in the adjacent layer, causing hyperpolarization during null direction motion (Mauss, 2015).

GABAergic inhibition has been shown to shape response properties of interneurons in early visual processing by mediating lateral antagonistic effects in Drosophila. Work in the Calliphora visual system has ascribed a more specialized role for GABAergic transmission in mediating null direction inhibition, based on experiments using picrotoxinin as a GABA receptor antagonist. Unexpectedly, in the same context, glutamate has been identified as the underlying neurotransmitter in Drosophila. This discrepancy is perhaps due to neglecting the action of the pharmacologic compound as a rather unspecific chloride channel blocker in earlier work. It should also be noted that, in Calliphora, picrotoxinin application was shown to have two effects on tangential cell motion processing: preferred direction depolarization was enlarged, and null direction hyperpolarization was replaced by noticeable depolarization. This was interpreted as evidence for weak directional tuning of the inputs, i.e., the later identified T4/T5 cells. A similar result was observed in Drosophila. The LPi3-4 block in Drosophila, however, did not produce a prominent null direction depolarization, and preferred direction excitation was indistinguishable from the control condition. Since a recent study demonstrated narrow directional tuning of the T4/T5 cells, rendering postsynaptic directional response sharpening unnecessary, it is suggested that picrotoxinin off-target effects on glutamate or GABA receptors in the upstream circuit are responsible for this inconsistency, and genetic LPi block represents a more suitable approach to eliminate null direction inhibition (Mauss, 2015).

This analysis focused on the LPi3-4 neurons and their postsynaptic partners in layer 4, the VS cells, because of their experimental accessibility. However, the current findings can be most likely extended to the other layers. Tangential cells with dendrites in layer 3 have been identified in other fly species. Such so-called V2 cells are motion opponent with preference to upward flow, in agreement with their presumed inputs from excitatory layer 3 T4/T5 cells. Since the data indicate that the LPi4-3 neurons convey glutamatergic signals selective for downward motion to lobula plate layer 3, it seems plausible that a motion-opponent wiring complementary to the LPi3-4/VS cell connectivity exists as well. The preference of LPi4-3 cells to ON over OFF edges is unexpected because in contrast to tangential cells, LPi4-3 neurons appear to be able to differentiate between T4/ON and T5/OFF input. Whether this finding hints toward an ON-selective null direction inhibition in layer 3 postsynaptic cells, perhaps dictated by certain natural stimulus statistics, or whether it reflects merely a bias of the driver line for an ON-selective LPi4-3 subgroup remains to be investigated. Some presynaptic swellings of the complementary LPi3-4 cells also exhibited polarity preference, but at present, it is unclear whether this indicates a similar T4/ T5 selectivity on a cell-by-cell basis or stochastic sampling of inputs. The functional architecture of lobula plate layers 1 and 2 strongly resembles the one of layers 3 and 4 with a 90 directional tuning shift: motion-opponent HS cells with a preference for front-to-back motion ramify their dendrites exclusively in layer 1, while motion-opponent Hx cells that prefer back-to-front motion confine their dendrites to layer 2. Therefore the existence of at least two complementary horizontal LPi cell types is anticipated in those layers too. It thus seems that global motion information is processed initially in two segregated horizontal and vertical subsystems with little direct interaction. Rather than representing the cardinal directions in a clock- or counter-clockwise manner, the four lobula plate layers are arranged such that opposite directions are represented side by side. This functional organization might serve to facilitate efficient nearest-neighbor interactions of motion-opponent signals (Mauss, 2015).

Similar to the fly lobula plate, the dorsal lateral geniculate nucleus (dLGN) in mammals relays direction-selective signals from the retina to higher brain centers. Some fundamental parallels in the organization of the two brain regions seem to exist. Their input channels, T4/T5 neurons in flies and ON/OFF direction-selective ganglion cells in mammals, predominantly encode the four cardinal directions of motion up, down, left, and right. The anatomical separation of the vertical and horizontal subsystems in flies seems to be mirrored, at least to a degree, in the dLGN, where opposing horizontal direction information resides in the superficial region of mouse dLGN, segregated from vertical motion. Moreover, a feed-forward inhibitory principle to generate motion opponency that this study describes in the fly might also prevail in the dLGN, where directionally selective output neurons were suggested to integrate opposing signals from retinal ganglion cells, possibly directly and indirectly via local inhibitory neurons. However, many mammalian dLGN neurons are also orientation selective, potentially obtaining this property by integrating opponent excitatory direction-selective input (Mauss, 2015).

Associated with their proposed role as matched filters for sensing the optic flow generated by an animal's self-motion, in contrast to dLGN neurons, lobula plate tangential cells have large receptive fields, in some cases covering more than 100 degrees of visual space. Independent movement, e.g., originating from conspecifics or foliage, thus poses a challenge to the system by providing excitatory drive to tangential cells not associated with self-motion. The current experiments with intact and silenced LPi neurons support the idea that such inputs are attenuated by antagonistic signals from oppositely moving objects elsewhere in the visual scene. Perhaps more importantly, different flight maneuvers generate ambiguous optic flow patterns in sub-parts of the receptive field. For instance, both lift and forward translation cause downward optic flow in the ventral visual field, while only the latter produces upward flow dorsally. Taking into account excitation only, a reliable distinction between those patterns, especially under varying stimulus intensities, i.e., contrasts as experienced in natural scenes, seems inconceivable. This study has demonstrated that LPi cells strongly reduce such ambiguities, most likely by cancelling the excitation caused in one part of the dendrite by inhibition in another part. Motion opponency is thus reminiscent of other neural opponent mechanisms. In the classical example of color opponency, neural comparison discriminates sensory signals that are ambiguous at the level of photoreceptors in terms of wavelength and stimulus intensity. Notably, while color vision requires at least two separate measurements at any point in space, motion opponency disambiguates different optic flowfields derived from the same photoreceptor responses. Given that wide-field motion-sensitive neurons in various other systems are also motion opponent, it is suggested that such a mechanism might be universally required to increase sensitivity and selectivity for optic flow-fields associated with selfmotion. Similar neural comparators might be widely used for the extraction of equally complex sensory features (Mauss, 2015).

How neurons form appropriately sized dendritic fields to encounter their presynaptic partners is poorly understood. The Drosophila medulla is organized in layers and columns and innervated by medulla neuron dendrites and photoreceptor axons. This study shows that three types of medulla projection (Tm) neurons extend their dendrites in stereotyped directions and to distinct layers within a single column for processing retinotopic information. In contrast, the Dm8 amacrine neurons form a wide dendritic field to receive approximately 16 R7 photoreceptor inputs. R7- and R8-derived Activin selectively restricts the dendritic fields of their respective postsynaptic partners, Dm8 and Tm20, to the size appropriate for their functions. Canonical Activin signaling promotes dendritic termination without affecting dendritic routing direction or layer. Tm20 neurons lacking Activin signaling expand their dendritic fields and aberrantly synapse with neighboring photoreceptors. It is suggested that afferent-derived Activin regulates the dendritic field size of their postsynaptic partners to ensure appropriate synaptic partnership (Ting, 2014).

Studies of the past decade have revealed the molecular mechanisms by which photoreceptor and lamina axons target to specific medulla layers and restricting their axonal terminals to their retinotopic columns. However, obtaining comparable information on their synaptic partners, the dendrites of the medulla neurons, has been stymied by the morphological complexity of the dendrites and the lack of genetic and analytical tools. This study reports new techniques for visualizing and analyzing the dendritic structures of the medulla neurons. The dual-imaging technique reported in this study requires only regular confocal microscopes to generate 3D images of dendrites, and the registration method exploits the regular array structure of the medulla to permit the standardization of dendritic patterns for analyses. In addition, recent studies have begun to unravel the transcription programs that govern the development of diverse medulla cell types and serial EM reconstruction studies have illustrated their complex synaptic circuits. These advances are paving the way to studies on dendrite patterning in three-dimensional space and the establishment of synaptic partnerships between axons and dendrites in the Drosophila medulla (Ting, 2014).

Morphometric analyses of Tm neurons revealed dendritic properties distinct from those of the da sensory neurons in the PNS. Different classes of da neurons have distinguishable morphometric parameters, such as branching geometry, and their distinguishing dendritic morphological features are correlated with specific branching orders. In contrast, the three Tm neuron types have similar dendritic branching and terminating frequencies, and share similar dendritic geometrical properties. As such, the standard morphometrics are ineffective in differentiating Tm types. Instead, the type-specific dendritic attributes of Tm neurons, revealed by registration, are directly related to the layer and column structure of the medulla. These distinctions likely reflect the different functions of the dendrites of da and Tm neurons serve: da dendrites are free-ending sensory branches that tile the two-dimensional body wall, whereas Tm dendrites receive synaptic inputs from retinotopically directed afferents organized in layers and columns (Ting, 2014).

Examination of the dendritic attributes of Tm neurons provides insights into how dendritic morphology is achieved in these neurons. The axons of Tm9 and Tm2 were localized to the posterior side of their cognate columns, and their dendrites extended anteriorly to innervate their cognate columns, while Tm20 axons were at the anterior side of the column and their dendrites projected posteriorly. These type-specific axonal localizations are probably a consequence of their ontogeny: Tm9 and Tm2 axons are fasciculated, and these neurons are presumably derived from one or closely related cell lineages. Layer-specific termination of dendrites, the most prominent type-specific attribute of medulla neurons, matches the layer-specific termination of their presynaptic partners. Previous developmental studies have showed that columns and layers form as the different types of afferents sequentially innervate the medulla during development. The observation that type-specific dendritic attributes are directly related to medulla layers and columns thus led to the hypothesis that different types of medulla neurons respond differentially to various afferent-derived cues to pattern dendrites (Ting, 2014).

This study has revealed that Activin signaling regulates the dendritic field size of the columnar neuron Tm20 and the wide-field amacrine neuron Dm8. TGF-β family morphogens have been shown to regulate a broad range of neurodevelopmental processes, including proliferation and cell-fate determination during early stage of development as well as the promotion of axonal development through a non-canonical pathway involving Lim kinase. This study has found that Activin signals through the canonical pathway to regulate dendritic patterning in Tm20 and Dm8 neurons: the transcription factor dSmad2 was functionally required for Activin-dependent dendritic patterning, and the lack of Activin signaling caused a global dendritic phenotype (Ting, 2014).

The specificity of Activin signaling in regulating dendritic patterning is underscored by the distinct dendritic patterning defects, as well as the absence of cell-fate transformation and axonal targeting defects, observed in mutant neurons. Activin-signaling-deficient Tm20 neurons still expressed the appropriate transcription factors and markers, and projected their axons to the correct target layer in the lobula; mutant Dm8 elaborated their dendrites in the appropriate medulla layer M6. Instead, lack of Activin signaling reduced dendritic terminating frequency and expanded dendritic fields while the planar projection orientations and layer-specific terminations were largely unaffected. It is speculated that Activin signaling leads to the transcription regulation of as-yet-unidentified factor(s) that facilitate the termination of dendritic arbors (Ting, 2014).

Several lines of evidence suggest that Activin for Tm20 and Dm8 is, at least in part, provided by their respective presynaptic partners, R8 and R7. R7 and R8, but not R1-R6, expressed Activin during pupal stages, and RNAi-mediated knockdown of Activin in photoreceptors resulted in expanded dendritic fields in the Dm8 and Tm20 neurons. The cell-specific requirement for Activin is underscored by the observations that the genetic ablation of R7s phenocopied babo in Dm8 but not in Tm20 neurons, and that stalling R8 axons in the superficial M1 layer (thereby preventing the delivery of R8-derived Activin to the M3 layer) and Activin knock-down in R8s, disrupt the normal dendritic patterning of Tm20. It is noted that either Activin-RNAi or R8 stalling caused milder Tm20 dendritic phenotypes than did removing Baboon but that their combination did not significantly enhance Tm20 dendritic phenotypes. Thus, there is likely a secondary and partially redundant Activin source, for example, other medulla neurons, for Tm20. Nonetheless, the specific requirement of afferents indicates that R8-derived Activin cannot substitute for that from R7s, and vice-versa, even though these terminals reside only a few micrometers apart. It is therefore suggested that R7- and R8-derived Activin functions, in a short-range or layer-restricted fashion, to regulate the dendritic patterning of Dm8 and Tm20, respectively (Ting, 2014).

Photoreceptor-derived morphogens, such as Hedgehog, EGF, and Jelly Belly, have been shown to act, in an anterograde fashion early in development, to regulate the proliferation and differentiation of their target area. The current finding, that photoreceptor-derived Activin regulates the dendrite development of their synaptic targets, further suggests anterograde signaling as an effective mechanism for coordinating afferent-target development, even late in development. The wide-field neuron Dm8, which mediates innate UV preference, pools approximately 16 R7 inputs and relays the information to a few Tm5c neurons in the center of their dendritic field, while the Tm20 neurons form synapses in one-to-one correspondence with R8s, and process retinotopic information. Conceivably, the formation of appropriate dendritic field sizes and the correct synaptic partnership is critical for the functions of Tm20 and Dm8 neurons. It is reasoned that the dendritic tiling mediated by mutual repulsion, an important regulatory feature of the da neurons, alone is not suitable for restricting Tm20s' and Dm8s' dendritic fields because Tm20's dendrites arborized in three-dimensional space, and Dm8's dendrites overlap extensively with their neighbors'. The use of afferent-derived Activin to regulate dendritic patterning provides an adaptable and self-compensating mechanism for afferents to control the dendritic field sizes of their synaptic targets (Ting, 2014).

Many visual animals have innate preferences for particular wavelengths of light, which can be modified by learning. Drosophila's preference for UV over visible light requires UV-sensing R7 photoreceptors and specific wide-field amacrine neurons called Dm8. This study identified three types of medulla projection neurons downstream of R7 and Dm8 and has shown that selectively inactivating one of them (Tm5c) abolishes UV preference. Using a modified GRASP method to probe synaptic connections at the single-cell level, each Dm8 neuron was shown to form multiple synaptic contacts with Tm5c in the center of Dm8's dendritic field but sparse connections in the periphery. By single-cell transcript profiling and RNAi-mediated knockdown, it was determined that Tm5c uses the kainate receptor Clumsy to receive excitatory glutamate input from Dm8. It is concluded that R7s-->Dm8-->Tm5c form a hard-wired glutamatergic circuit that mediates UV preference by pooling approximately 16 R7 signals for transfer to the lobula, a higher visual center (Karuppudurai, 2014).

Understanding how visual systems translate light impulses into adaptively tuned percepts to guide behavior is a central goal in neurobiology. The Drosophila visual system, with its amenability to genetic manipulation, has enabled increasingly deep investigation of the molecular and cellular basis of visual-driven behaviors, including the spectral preference for UV light examined in this study. UV spectral preference has previously been shown to require first-order interneurons in the visual medulla (i.e., the wide-field amacrine Dm8 neurons) that receive inputs from multiple UV-sensing photoreceptors. This study shows that a subclass of Tm5 neurons, called Tm5c, receive excitatory glutamatergic input from Dm8 neurons through the kainate-type receptor Clumsy. Glutamatergic signaling, both to and by Tm5c, was shown to be necessary for normal UV preference. Together these results define not only critical elements of the molecular and cellular machinery underlying UV preference, but also patterns of connectivity and information flow at the first several processing stations of this important visual circuit (Karuppudurai, 2014).

By sparse reconstruction of serial-section transmission EM (ssTEM), Previous work identified Dm8 as the major postsynaptic partner for R7. This study shows that Tm5c interneurons are required for transducing Dm8's signal to the lobula, a higher visual center, in the UV preference pathway. The thin and complex dendrites of Tm5c make them challenging to resolve by ssTEM, due to the limited axial resolution (∼50 nm) of this method. To visualize Dm8->Tm5c and R8->Tm5c synapses, this study therefore resorted to the GRASP technique. To differentiate synaptic contacts from mere membrane contacts and to visualize the spatial distribution of Dm8->Tm5c synapses, the GRASP method was adapted to permit single-cell identification of presumptive presynaptic neurons in which active zones were fluorescently tagged. By applying this single-cell GRASP method, it was demonstrated that each Dm8 neuron has multiple synaptic contacts to one (or, at most a few) Tm5c neuron in the center of its dendritic field but sparse synaptic contacts in the periphery. The nature of synaptic signaling at the R8->Tm5c and Dm8->Tm5c contacts was established by single-cell transcript profiling and functional studies. Tm5c expresses the histamine-gated chloride channel Ort, and restoring Ort expression in Tm5c in an ort mutant background drove strong green preference. Dm8 expresses VGlut, and Tm5c correspondingly expresses glutamate-gated ionotropic receptors. RNAi-mediated knockdown of VGlut in Dm8, or the Clumsy iGluR in Tm5c, abolished UV preference. It is believed that the approach taken in this study, which combines single-cell GRASP, transcript profiling, RNAi-mediated knockdown, and behavioral assays, could be profitably applied to the dissection and characterization of other complex neural circuits (Karuppudurai, 2014).

Using single-cell transcript profiling, four kainate-type glutamate receptor subunits (Clumsy, CG11155, DKaiRIC, and DKaiRID) expressed in Tm5c were identified. These four iGluRs and CG9935 share sequence homology and domain structures with vertebrate kainate-type iGluRs (GluK1/2/3 and GluK4/5). RNAi-mediated knockdown further revealed that Clumsy is functionally required in Tm5c for UV preference. This demonstrates that kainate-type iGluRs function in the Drosophila CNS. In vertebrates, functional kainate receptors assemble tetramerically as dimers of dimers; GluK1-3 are capable of forming functional homotetramers while GluK4/5 are obligatory heteromers. Fly kainate receptor subunits share with vertebrates the key hydrophobic resides at the dimerization interfaces, suggesting that they assemble in a similar way to the vertebrate receptors. RNAi knockdown of CG11155, DKaiRIC, or DKaiRID did not enhance UV preference defects caused by RNAi knockdown of Clumsy. However, simultaneous RNAi knockdown of all three iGluR subunits significantly reduced UV preference, suggesting that they are functionally redundant. It is thus suggested that Clumsy forms functional heterotetramers with one of the other kainate receptors in Tm5c to mediate UV preference. Further in vitro assembly and electrophysiological studies will be needed to determine the exact subunit composition of the functional iGluRs. Ionotropic glutamate receptors in flies have been exclusively studied at the neuromuscular junction, in part as a surrogate model for CNS glutamate synapses. Functional identification of kainate-type iGluRs in the fly visual system, in combination with the robust UV preference behavior they mediate, opens the door to studying the assembly, function, and regulation of this important class of glutamate receptors in the Drosophila CNS (Karuppudurai, 2014).

As the wide-field Dm8 neurons have no presynaptic sites or axonal projections outside of the external medulla, they depend on medulla projection neurons to transduce signals to higher visual centers. While all three subtypes of Tm5 neurons appear to be postsynaptic to Dm8 and therefore are capable of transducing the Dm8 signal to the lobula, this study shows that only Tm5c is functionally required for UV preference. Inactivating Tm5a/b or Dm8 in addition to Tm5c did not enhance UV preference defects, suggesting that Dm8's function in UV preference is solely communicated through Tm5c. Tm5c differs from Tm5a/b in axonal morphology and neurotransmitter usage. It is thus suggested that Tm5c has unique synaptic functions and/or targets in that visual compartment that account for its role in UV preference (Karuppudurai, 2014).

The spatial organization of the R7s->Dm8->Tm5c circuit suggests a neural pooling mechanism for UV preference. Each Dm8 amacrine neuron has a large dendritic field that receives ∼16 R7 inputs, while a single Tm5c is present in most, if not all, medulla columns and receives direct retinotopic inputs from R8s. Single-cell GRASP experiments reveal that each Dm8 forms multiple synaptic contacts with one, or at most a few, Tm5c neurons in the center of Dm8's dendritic field but forms few synapses with Tm5c outside of the center. By pooling 16 R7 inputs to a single Tm5c, Dm8 could increase UV sensitivity by up to 16-fold at some cost in spatial resolution. It is interesting to note that the amplification magnitude of the R7s->Dm8->Tm5c circuit depends primarily on the size of the Dm8's dendritic field, which is negatively regulated by R7-derived Activin during development: excess Activin reduces Dm8's dendritic field size while lack of Activin enhances it. It is thus tempting to speculate that size of the Dm8 arbor, and thus the trade off between UV sensitivity and spatial resolution, has been adjusted in the course of insect evolution to meet each insect's ecological needs (Karuppudurai, 2014).

In addition to the excitatory glutamate input from Dm8, Tm5c neurons also receive inhibitory histaminergic inputs directly from R8 photoreceptors. Thus, the R7s->Dm8->Tm5c pooling circuit is superimposed on the retinotopic circuit R8->Tm5c. Tm5c expresses Ort, and restoring Ort expression in Tm5c in various ort mutant backgrounds rescued green phototaxis. Thus, both direct (R8->Tm5c) and indirect (R7s->Dm8->Tm5c) pathways transduce sign-inverting signals to Tm5c and both pathways are capable of driving phototaxis. However, RNAi-mediated knockdown of ort in Tm5c, which prevents the reception of R8 inputs, did not affect normal UV preference. This observation is consistent with previous conclusions that the indirect pathway through R7s->Dm8 is both required and sufficient for optimal UV preference, at least under the test condition, and that multiple redundant pathways mediate green phototaxis. While the direct pathway is not involved in UV preference, it might play a role in true color vision. Notably, chloride ions are a known modulator for kainate receptor and the direct pathway signals through the histamine-gated chloride channel Ort. Given that multiple pathways function redundantly in true color vision, differentiating these possibilities must await single-unit electrophysiological recording and/or new genetic techniques to segregate their functions (Karuppudurai, 2014).

The receptor mechanism for color vision has been extensively studied. In contrast, the circuit(s) that transform(s) photoreceptor signals into color percepts to guide behavior remain(s) poorly characterized. Using intersectional genetics to inactivate identified subsets of neurons in the optic lobe, this study has uncovered the first-order interneurons that are functionally required for hue discrimination in Drosophila. A novel aversive operant conditioning assay was developed for intensity-independent color discrimination (true color vision) in Drosophila. Single flying flies are magnetically tethered in an arena surrounded by blue and green LEDs (light-emitting diodes). The flies' optomotor response is used to determine the blue-green isoluminant intensity. Flies are then conditioned to discriminate between equiluminant blue or green stimuli. Wild-type flies are successfully trained in this paradigm when conditioned to avoid either blue or green. Functional color entrainment requires the function of the narrow-spectrum photoreceptors R8 and/or R7, and is within a limited range, intensity independent, suggesting that it is mediated by a color vision system. The medulla projection neurons, Tm5a/b/c and Tm20, receive direct inputs from R7 or R8 photoreceptors and indirect input from the broad-spectrum photoreceptors R1-R6 via the lamina neuron L3. Genetically inactivating these four classes of medulla projection neurons abolished color learning. However, inactivation of subsets of these neurons is insufficient to block color learning, suggesting that true color vision is mediated by multiple redundant pathways. It is hypothesized that flies represent color along multiple axes at the first synapse in the fly visual system. The apparent redundancy in learned color discrimination sharply contrasts with innate ultraviolet (UV) spectral preference, which is dominated by a single pathway from the amacrine neuron Dm8 to the Tm5c projection neurons (Melnattur, 2014).

The extraction of directional motion information from changing retinal images is one of the earliest and most important processing steps in any visual system. In the fly optic lobe, two parallel processing streams have been anatomically described, leading from two first-order interneurons, L1 and L2, via T4 and T5 cells onto large, wide-field motion-sensitive interneurons of the lobula plate. Therefore, T4 and T5 cells are thought to have a pivotal role in motion processing; however, owing to their small size, it is difficult to obtain electrical recordings of T4 and T5 cells, leaving their visual response properties largely unknown. This problem was circumvented by means of optical recording from these cells in Drosophila, using the genetically encoded calcium indicator GCaMP5. This study finds that specific subpopulations of T4 and T5 cells are directionally tuned to one of the four cardinal directions; that is, front-to-back, back-to-front, upwards and downwards. Depending on their preferred direction, T4 and T5 cells terminate in specific sublayers of the lobula plate. T4 and T5 functionally segregate with respect to contrast polarity: whereas T4 cells selectively respond to moving brightness increments (ON edges), T5 cells only respond to moving brightness decrements (OFF edges). When the output from T4 or T5 cells is blocked, the responses of postsynaptic lobula plate neurons to moving ON (T4 block) or OFF edges (T5 block) are selectively compromised. The same effects are seen in turning responses of tethered walking flies. Thus, starting with L1 and L2, the visual input is split into separate ON and OFF pathways, and motion along all four cardinal directions is computed separately within each pathway. The output of these eight different motion detectors is then sorted such that ON (T4) and OFF (T5) motion detectors with the same directional tuning converge in the same layer of the lobula plate, jointly providing the input to downstream circuits and motion-driven behaviours (Maisak, 2013).

Most of the neurons in the fly brain are dedicated to image processing. The respective part of the head ganglion, called the optic lobe, consists of several layers of neuropile called lamina, medulla, lobula and lobula plate, all built from repetitive columns arranged in a retinotopic way. Each column houses a set of identified neurons that, on the basis of Golgi staining, have been described anatomically, first by Santiago Ramon y Cajal, in great detail. Owing to their small size, however, most of these columnar neurons have never been recorded from electrophysiologically. Therefore, their specific functional role in visual processing is still largely unknown. This fact is contrasted by rather detailed functional models about visual processing inferred from behavioural studies and recordings from the large, electrophysiologically accessible output neurons of the fly lobula plate (tangential cells). As the most prominent example of such models, the Reichardt detector derives directional motion information from primary sensory signals by multiplying the output from adjacent photoreceptors after asymmetric temporal filtering. This model makes a number of rather counter-intuitive predictions all of which have been confirmed experimentally. Yet, the neurons corresponding to most of the circuit elements of the Reichardt detector have not been identified so far. This study focused on a set of neurons called T4 and T5 cells which, on the basis of circumstantial evidence, have long been speculated to be involved in motion detection. However, it is unclear to what extent T4 and T5 cells are directionally selective or whether direction selectivity is computed or enhanced within the dendrites of the tangential cells. Another important question concerns the functional separation between T4 and T5 cells; that is, whether they carry equivalent signals, maybe one being excitatory and the other inhibitory on the tangential cells, or whether they segregate into directional- and non-directional pathways or into separate ON- and OFF-motion channels (Maisak, 2013).

To answer these questions, Gal4-driver lines specific for T4 and T5 cells were combined with GCaMP5 and the visual response properties were optically recorded using two-photon fluorescence microscopy. In a first series of experiments, a driver line labelling both T4 and T5 cells was used. A confocal image revealed clear labelling in the medulla (T4 cell dendrites), in the lobula (T5 cell dendrites), as well as in four distinct layers of the lobula plate, representing the terminal arborizations of the four subpopulations of both T4 and T5 cells. These four layers of the lobula plate can also be seen in the two-photon microscope when the calcium indicator GCaMP5 is expressed. After stimulation of the fly with grating motion along four cardinal directions (front-to-back, back-to-front, upwards and downwards), activity is confined to mostly one of the four layers, depending on the direction in which the grating is moving. The outcome of all four stimulus conditions can be combined into a single image by assigning a particular colour to each pixel depending on the stimulus direction to which it responded most strongly. From these experiments it is clear that the four subpopulations of T4 and T5 cells produce selective calcium signals depending on the stimulus direction, in agreement with previous deoxyglucose labelling. Sudden changes of the overall luminance evokes no responses in any of the layers. However, gratings flickering in counter-phase lead to layer-specific responses, depending on the orientation of the grating (Maisak, 2013).

The retinotopic arrangement of this input to the lobula plate is demonstrated by experiments where a dark edge was moved within a small area of the visual field only. Depending on the position of this area, activity of T4 and T5 cells is confined to different positions within the lobula plate. Consequently, when moving a bright vertical edge horizontally from back to front, activity of T4 and T5 cells is elicited sequentially in layer 2 of the lobula plate. These two experiments also demonstrate that T4 and T5 cells indeed signal motion locally. Next, the question of where direction selectivity of T4 and T5 cells arises was investigated; that is, whether it is already present in the dendrite, or whether it is generated by synaptic interactions within the lobula plate. This question is hard to answer, as the dendrites of both T4 and T5 cells form a dense mesh within the proximal layer of the medulla (T4) and the lobula (T5), respectively. However, signals within the inner chiasm where individual processes of T4 and T5 cells can be resolved in some preparations show a clear selectivity for motion in one over the other directions. Such signals are as directionally selective as the ones measured within the lobula plate, demonstrating that the signals delivered from the dendrites of T4 and T5 cells are already directionally selective (Maisak, 2013).

To assess the particular contribution of T4 and T5 cells to the signals observed in the above experiments, driver lines specific for T4 and T5 cells, respectively, were used. Applying the same stimulus protocol and data evaluation as described above, identical results were obtained as before for both the T4- as well as the T5-specific driver line. It is concluded that T4 and T5 cells each provide directionally selective signals to the lobula plate. Thus, both T4 and T5 cells can be grouped, according to their preferred direction, into four subclasses covering all four cardinal directions, reminiscent of ON–OFF ganglion cells of the rabbit retina (Maisak, 2013).

Next whether T4 cells respond differently to T5 cells was addressed. To answer this question, moving edges instead of gratings were used with either positive (ON edge, brightness increment) or negative (OFF edge, brightness decrement) contrast polarity as visual stimuli. It was found that T4 cells strongly responded to moving ON edges, but showed little or no response to moving OFF edges. This is true for T4 cells terminating in each of the four layers. The opposite was found for T5 cells. T5 cells selectively responded to moving OFF edges and mostly failed to respond to moving ON edges. Again, this was found for T5 cells in each of the four layers. Next, whether there are any other differences in the response properties between T4 and T5 cells was addressed by testing the velocity tuning of both cell populations by means of stimulating flies with grating motion along the horizontal axis from the front to the back at various velocities covering two orders of magnitude. T4 cells revealed a maximum response at a stimulus velocity of 30° s-1, corresponding to a temporal frequency of 1 Hz. T5 cell responses showed a similar dependency on stimulus velocity, again with a peak at a temporal frequency of 1 Hz. Thus, there is no obvious difference in the velocity tuning between T4 and T5 cells. As another possibility, T4 cells might functionally differ from T5 cells with respect to their directional tuning width. To test this, flies were stimulated with gratings moving into 12 different directions, and the relative change of fluorescence was evaluated in all four layers of the lobula plate. Using the T4-specific driver line, an approximate half width of 60-90° of the tuning curve was found, with the peak responses in each layer shifted by 90°. No decrease of calcium was detectable for grating motion opposite to the preferred direction of the respective layer. When the experiments were repeated using the T5-specific driver line, a similar dependence of the relative change of fluorescence was found on the stimulus direction. It is concluded that T4 cells have the same velocity and orientation tuning as T5 cells. The only functional difference that was detected remains their selectivity for contrast polarity (Maisak, 2013).

The finding about the different preference of T4 and T5 cells for the polarity of a moving contrast makes the strong prediction that selective blockade of T4 or T5 cells should selectively compromise the responses of downstream lobula plate tangential cells to either ON or OFF edges. To test this prediction, the output of either T4 or T5 cells was blocked via expression of the light chain of tetanus toxin, and the responses of tangential cells via somatic whole-cell patch was recorded to moving ON and OFF edges. In response to moving ON edges, strong and reliable directional responses were observed in all control flies. However, T4-block flies showed a strongly reduced response to ON edges, whereas the responses of T5-block flies were at the level of control flies. When moving OFF edges were used, control flies again responded with a large amplitude. However, the responses of T4-block flies were at the level of control flies, whereas the responses of T5-block flies were strongly reduced. These findings are reminiscent of the phenotypes obtained from blocking lamina cells L1 and L2 and demonstrate that T4 and T5 cells are indeed the motion-coding intermediaries for these contrast polarities on their way to the tangential cells of the lobula plate. Whether the residual responses to ON edges in T4-block flies and to OFF edges in T5-block flies are due to an incomplete signal separation between the two pathways or due to an incomplete genetic block in both fly lines is currently unclear (Maisak, 2013).

To address the question of whether T4 and T5 cells are the only motion detectors of the fly visual system, or whether they represent one cell class, in parallel to other motion-sensitive elements, tethered flies walking on an air-suspended sphere were used, and and they were stimulated by ON and OFF edges moving in opposite directions. As in the previous experiments, T4 and T5 cells were blocked specifically by selective expression of the light chain of tetanus toxin. During balanced motion, control flies did not show significant turning responses to either side. T4-block flies, however, strongly followed the direction of the moving OFF edges, whereas T5-block flies followed the direction of the moving ON edges. In summary, the selective preference of T4-block flies for OFF edges and of T5-block flies for ON edges not only corroborates the findings about the selective preference of T4 and T5 cells for different contrast polarities, but also demonstrates that the signals of T4 and T5 cells are indeed the major, if not exclusive, inputs to downstream circuits and motion-driven behaviours (Maisak, 2013).

Almost a hundred years after T4 and T5 cells have been anatomically described, this study reports their functional properties in a systematic way. Using calcium as a proxy for membrane voltage, this study found that both T4 and T5 cells respond to visual motion in a directionally selective manner and provide these signals to each of the four layers of the lobula plate, depending on their preferred direction. Both cell types show identical velocity and orientation tuning which matches the one of the tangential cells. The strong direction selectivity of both T4 and T5 cells is unexpected, as previous studies had concluded that the high degree of direction selectivity of tangential cells is due to a push–pull configuration of weakly directional input with opposite preferred direction. Furthermore, as the preferred direction of T4 and T5 cells matches the preferred direction of the tangential cells branching within corresponding layers, it is currently unclear which neurons are responsible for the null-direction response of the tangential cells. As for the functional separation between T4 and T5 cells, this study found that T4 cells selectively respond to brightness increments, whereas T5 cells exclusively respond to moving brightness decrements. Interestingly, parallel ON and OFF motion pathways had been previously postulated on the basis of selective silencing of lamina neurons L1 and L2. Studies using apparent motion stimuli to probe the underlying computational structure arrived at controversial conclusions: whereas some studies concluded that there was a separate handling of ON and OFF events by motion detectors, others did not favour such a strict separation. The present study directly demonstrates the existence of separate ON and OFF motion detectors, as represented by T4 and T5 cells, respectively. Furthermore, the results anatomically confine the essential processing steps of elementary motion detection -- that is, asymmetric temporal filtering and nonlinear interaction -- to the neuropile between the axon terminals of lamina neurons L1 and L2 and the dendrites of directionally selective T4 and T5 cells. The dendrites of T4 and T5 cells might well be the place where signals from neighbouring columns interact in a nonlinear way, similar to the dendrites of starburst amacrine cells of the vertebrate retina (Maisak, 2013).

Detecting motion is a feature of all advanced visual systems, nowhere more so than in flying animals, like insects. In flies, an influential autocorrelation model for motion detection, the elementary motion detector circuit (EMD), compares visual signals from neighboring photoreceptors to derive information on motion direction and velocity. This information is fed by two types of interneuron, L1 and L2, in the first optic neuropile, or lamina, to downstream local motion detectors in columns of the second neuropile, the medulla. Despite receiving carefully matched photoreceptor inputs, L1 and L2 drive distinct, separable pathways responding preferentially to moving 'on' and 'off' edges, respectively. Serial electron microscopy (EM) identifies two types of transmedulla (Tm) target neurons, Tm1 and Tm2, that receive apparently matched synaptic inputs from L2. Tm2 neurons also receive inputs from two retinotopically posterior neighboring columns via L4, a third type of lamina neuron. Light microscopy reveals that the connections in these L2/L4/Tm2 circuits are highly determinate. Single-cell transcript profiling suggests that nicotinic acetylcholine receptors mediate transmission within the L2/L4/Tm2 circuits, whereas L1 is apparently glutamatergic. It is proposed that Tm2 integrates sign-conserving inputs from neighboring columns to mediate the detection of front-to-back motion generated during forward motion (Takemura, 2011).

Given that both L2 and L4 express Choline acetyltransferase (Cha) and are thus genotypically qualified to synthesize acetylcholine and provide cholinergic input to Tm2, the expression of acetylcholine receptors in Tm2 was profiled. This proved more complex than for L2 and L4. In addition to Dα7 and Dβ1 nAcR shared with L2 and L4, Tm2 also expressed Dα1/2 and Dβ2 nAcR. The exclusive expression of nicotinic rather than muscarinic receptors (nAcR not mAcR) in Tm2 suggests that both L2 and L4 provide fast excitatory inputs to Tm2. It was also found that Tm2 expressed Cha but not VGlut, indicating that, like L2 and L4, Tm2 is also genotypically cholinergic. In summary, these data predict that both synaptic connections in the L2/L4/Tm2 network are mediated by excitatory acetylcholine systems, and therefore sign-conserving (Takemura, 2011).

While either the L1 or L2 channel alone can mediate rudimentary motion detection, each also responds differentially in walking flies, and in head-yaw assays the L2 pathway is preferentially tuned to front-to-back motion. Although the connections between L4 and L2 along the anteroposterior direction might account for this front-to-back preference, these connections are reciprocal so that information also flows from posterior to anterior, while L2's activity fails to reveal asymmetrical responses. Between L2's two targets, only Tm2 receives two additional L4 inputs from neighboring posterior columns; Tm1 does not. These L2/L4/Tm2 connections are highly determinate, underscoring a critical role in connecting neighboring L2 channels along the AP direction, in what is arguably the most important motion direction for flies since it occurs during forward flight. Interestingly, other flies have a Tm neuron closely resembling Drosophila's Tm2 morphologically, for example Tm1 in the calliphorid Phaenicia. This is proposed to receive L2 inputs, suggesting that an L2/L4/Tm2 network might be conserved in higher Diptera (Takemura, 2011).

Tm2 could conceivably serve as half of the EMD's multiplier stage, comparing the temporally delayed input from collateral L4s with the cognate signal from L2. However, electrophysiological investigations on calliphorid 'Tm1' neurons, which resemble morphologically Drosophila's Tm2, have yet to provide strong evidence for this role. An alternative interpretation is that the L2/L4/Tm2 network serves instead as a prefilter in the preprocessing stage while Tm2's output feeds into the multiplier stage. The topology and sign-conserving nature of L4/Tm2 connections suggest the spatial summation of neighboring visual signals, which could increase light sensitivity at the expense of spatial acuity. It has been suggested that under low luminance conditions, neighboring visual signals are pooled prior to their interaction at the multiplier stage, while at higher luminance levels nearest-neighbor interactions dominate motion detection. Alternatively, the L4/Tm2 connections could convert visual signals sampled from the hexagonal ommatidial array into an orthogonal coordinate upon which motion signals can be derived. Differentiating between these possibilities must await future investigations that combine genetic and electrophysiological approaches (Takemura, 2011).

Motion vision is a major function of all visual systems, yet the underlying neural mechanisms and circuits are still elusive. In the lamina, the first optic neuropile of Drosophila melanogaster, photoreceptor signals split into five parallel pathways, L1-L5. This study examines how these pathways contribute to visual motion detection by combining genetic block and reconstitution of neural activity in different lamina cell types with whole-cell recordings from downstream motion-sensitive neurons. Reduced responses to moving gratings are found if L1 or L2 is blocked; however, reconstitution of photoreceptor input to only L1 or L2 results in wild-type responses. Thus, the first experiment indicates the necessity of both pathways, whereas the second indicates sufficiency of each single pathway. This contradiction can be explained by electrical coupling between L1 and L2, allowing for activation of both pathways even when only one of them receives photoreceptor input. A fundamental difference between the L1 pathway and the L2 pathway is uncovered when blocking L1 or L2 output while presenting moving edges of positive (ON) or negative (OFF) contrast polarity: blocking L1 eliminates the response to moving ON edges, whereas blocking L2 eliminates the response to moving OFF edges. Thus, similar to the segregation of photoreceptor signals in ON and OFF bipolar cell pathways in the vertebrate retina, photoreceptor signals segregate into ON-L1 and OFF-L2 channels in the lamina of Drosophila (Joesch, 2010).

Neurons responding to visual motion in a directionally selective way are found in a vast number of animals and brain regions, ranging from the retina of rabbits to the visual cortex of macaques. In flies, large-field motion-sensitive neurons are located in the third neuropile layer, the lobula plate, and are thought to be involved in visual flight control. These lobula plate tangential cells are preferentially sensitive to vertical (VS cells) and horizontal (HS cells) motion, respectively. They depolarize when stimulated by motion along their preferred direction (PD motion) and hyperpolarize during motion along the opposite, null direction (ND motion). In the first neuropile, the lamina, photoreceptors R1-R6 provide input, directly or indirectly, onto five different monopolar cells (L1-L5) using histamine as their transmitter. L1-L5 send their axons into the medulla where neurons compute the direction of motion in accordance with the Reichardt model. Such motion detectors then provide excitatory and inhibitory input onto the dendrites of lobula plate tangential cells. However, the neural circuitry presynaptic to the tangential cells represented by the Reichardt detectors has so far escaped a detailed analysis, because of the small size of the columnar neurons. This study set out to elucidate the cellular implementation of the Reichardt model of visual motion detection starting from the lamina, asking which of the various neurons provide input to the motion detection circuitry. Previous studies addressing this question in Drosophila used behavioural read-outs to test for effects of blocking and rescuing of specific lamina cells. To get closer to the circuit in question, the Gal4 or Split-Gal4/UASsystem were used and genetic intervention was combined in different lamina neurons with electrophysiological recordings from lobula plate tangential cells (Joesch, 2010).

Recordings were made from HS and VS cells, and the output of lamina neurons L1 and L2 was blocked by targeted expression of shibire^ts. Control flies always revealed strong and reliable responses to a moving grating, saturating for increasing contrast levels for both PD motion as well as ND motion. Blocking both L1 and L2 led to a complete loss of motion responses even at the highest pattern contrast. Blocking only L1 strongly reduced PD and ND responses for all contrasts tested. Blocking L2 using two different driver lines moderately reduced the responses at all contrast levels. To test whether the temperature shift alone could lead to altered motion responses, flies that had the same genotype as experimental flies except for the GAL4 driver gene were put to restrictive temperature 1 h before the experiment. The responses of these flies were indistinguishable from the ones of the other control flies. Together, these results indicate that L1 and L2 are necessary for wild-type responses to grating motion (Joesch, 2010).

In a complementary approach, photoreceptor input to lamina cells L1 and L2 was selectively rescued via targeted expression of the wild-type histamine receptor, encoded by the ort gene, in an ort-null mutant background. Given the above results from the blocking experiments, rescuing either L1 or L2 pathway should lead to only small motion responses at best. However, rescuing L2 led to wild-type motion responses at all contrasts tested, for PD motion as well as for ND motion. The same was true when lamina cells L1 were rescued: again, motion responses were nearly indistinguishable from the ones of 'positive control' flies. In these positive control flies, no L1- or L2-GAL4, but one wild-type ort-allele, was present, leading to wild-type motion responses as expected. In 'negative control' flies, where either no L1-GAL4 and L2-GAL4 or no UAS-ort was present in an ort-null mutant background, motion responses were literally zero. Thus, blocking L1 or L2 revealed that the output of both L1 and L2 is necessary for wild-type motion responses. Rescuing the pathway of either L1 or L2 indicates, however, that either L1 or L2 is sufficient for a wild-type motion response. This contradiction deserves further investigation (Joesch, 2010).

The blocking and rescuing experiments presented above differ in one important aspect: in one case, the synaptic output of L1 and L2 was blocked, in the other case, the synaptic input to the same cells was rescued. If L1 and L2 receive their input in parallel without any further interactions, both procedures should yield complementary results, which were not found. Thus, the existence of electrical connections between L1 and L2 was examined by immunolabelling of the innexin protein Shaking B, a member of the gap-junction-forming protein family in flies. Strong immunolabelling was found within the entire optic lobe including the lamina. Furthermore, the basal laminar processes of L1 and L2 appeared to co-localize with the Shaking B immunolabelling. Because some dipteran gap junctions were demonstrated to be permeable for neurobiotin, L1 cells were injected with neurobiotin and co-staining in L2 was looked for. When a single L1 cell was injected, a clear staining became visible in the adjacent L2 cell as well, identified by its characteristic terminal in medulla layer 2. Injecting L2 led to co-staining of the adjacent L1 cell, identified by its characteristic terminals in medulla layers 1 and 5. Therefore it is proposed that L1 and L2 are electrically coupled via gap junctions (Joesch, 2010).

Gap-junctional coupling between L1 and L2 could, in principle, explain the contradictory results obtained in blocking and rescuing experiments: through electrical coupling, rescuing the photoreceptor input to L1 restores the L2 pathway as well, and vice versa. This explanation, however, requires that the coupling between L1 and L2 provides a sufficiently large input to the respective partner cell. To investigate the strength of the coupling, an inwardly rectifying potassium channel (Kir2.1) was expressed in one of the two lamina cells. When the potassium channel was expressed in L1 alone, motion responses were completely abolished, comparable to the situation when L1 and L2 were blocked by shibire^ts. A similar finding was obtained when the potassium channel was expressed in L2 cells. These results indicate a strong electrical coupling between L1 and L2 and, thus, resolve the apparent discrepancy between blocking and rescuing experiments (Joesch, 2010).

So far, these data support the view that both L1 and L2 feed, with a somewhat different contribution, into the motion detection circuitry. However, no evidence is provided as to any functional specialization of each of the pathways. As one possibility, lamina cells L1 and L2 might be specifically involved in the analysis of either ON or OFF input signals, in analogy to the vertebrate retina. Because a grating stimulus is composed of many simultaneously moving dark-to-bright (ON edge) and bright-to-dark transitions (OFF edge), this would have escaped the analysis presented above. To investigate this possibility, moving edges of a single polarity were presented to flies in which the output of lamina cells L1 and L2 was blocked by shibire^ts. In control flies, moving ON and OFF edges elicited strong and reliable voltage responses in lobula plate tangential cells during PD and ND motion. When the output from L1 was blocked, the response to moving ON edges was literally zero whereas the response to moving OFF edges was still about 50% of the wild-type response. The opposite was true when the output from L2 was blocked by expressing shibire^ts using two different GAL4 driver lines: then, the response to moving ON edges was only mildly reduced whereas the response to moving OFF edges was nearly abolished (Joesch, 2010).

In a pioneering study, and consistent with these results, it was found that rescuing either the L1 or the L2 pathway led to wild-type optomotor responses at high pattern contrasts. For low contrasts (5%-10%), a functional specialization of the L1 and L2 pathway for back-to-front and front-to-back motion was suggested, which, however, does not match the current data on tangential cell responses in that contrast range. The first evidence for a role of the L2 pathway in transmitting light OFF signals was obtained in a study on freely walking flies, where blocking L2 impaired turning tendencies in response to contrast decrements. However, the current finding that photoreceptor signals in the fly segregate into ON and OFF pathways via L1 and L2 neurons is surprising in so far as, different from ON and OFF bipolar cells of the vertebrate retina, both lamina cell types posses the same transmitter receptor and produce similar light responses in their dendrite. This similarity is likely to be increased even further by the gap-junctional coupling between dendritic compartments of L1 and L2, which might help to average out uncorrelated noise both cells receive from photoreceptor R1-R6 input. Subsequently, however, these signals must become differentially rectified. For L2, this has been recently shown to occur already within the cell, as L2 axon terminals reveal pronounced calcium signals selectively in response to light OFF stimuli (Joesch, 2010).

Whether this also holds true for L1, or whether the selective responsiveness of the L1 pathway to light ON stimuli is only acquired further downstream in its postsynaptic neurons, is currently not known. On the basis of the co-stratification of columnar neurons as well as 2-deoxyglucose activity labelling, L1 and L2 have long been proposed to represent the entry points to two parallel motion pathways in the fly visual system, with L1 synapsing onto medulla intrinsic neuron Mi1 which in turn contacts T4 cells, L2 synapsing onto transmedullar neuron Tm1 which in turn contacts T5 cells, and with T4 and T5 cells finally converging on the dendrites of the lobula plate tangential cells (Joesch, 2010).

The results provide evidence that these two pathways deal specifically with the processing of ON and OFF stimuli. Moreover, splitting a positively and negatively going signal into separate ON and OFF channels alleviates the neural implementation of a multiplication, as postulated by the Reichardt detector. Whereas otherwise, the output signal of the multiplier had to increase in a supra-linear way when both inputs increase as well as when they decrease, dealing with positive signals only in separate multipliers seems to be less demanding with respect to the underlying biophysical mechanism. Whatever this mechanism will turn out to be, the finding about the splitting of the photoreceptor signal into ON and OFF pathways adds to the already described commonalities between the invertebrate and the vertebrate visual system. Obviously, the selection pressure for an energy-efficient way of encoding light increments and decrements led to rather similar implementations across distant phyla (Joesch, 2010).

In the visual system of Drosophila, photoreceptors R1-R6 relay achromatic brightness information to five parallel pathways. Two of them, the lamina monopolar cells L1 and L2, represent the major input lines to the motion detection circuitry. A new method was devised for optical recording of visually evoked changes in intracellular Ca2+ in neurons using targeted expression of a genetically encoded Ca2+ indicator. Ca2+ in single terminals of L2 neurons in the medulla carried no information about the direction of motion. However, this study found that brightness decrements (light-OFF) induced a strong increase in intracellular Ca2+ but brightness increments (light-ON) induced only small changes, suggesting that half-wave rectification of the input signal occurs. Thus, L2 predominantly transmits brightness decrements to downstream circuits that then compute the direction of image motion (Reiff, 2010).

The fly visual system continuously provides information about the motion of objects, conspecifics, predators and the three-dimensional structure of the environment. This information underlies the execution of notable visually driven behaviors. However, the manner in which small-scale neural networks accomplish such computational efficacy remains an open question, and the complete motion detection circuitry has not yet been determined in any animal. This study examined this question in Drosophila by analyzing how brightness changes become encoded in changes in the concentration of presynaptic Ca2+ in the axon terminals of L2 neurons, a major input channel to the motion detection circuitry (Reiff, 2010).

The processing of brightness changes underlies the detection of visual motion. On the basis of a detailed input-output analysis of the optomotor response in tethered beetles, the well-known Hassenstein-Reichardt model (HRM) of visual motion detection was derived. The HRM essentially performs a spatio-temporal cross-correlation of two luminance input signals by multiplying the signals derived from two neighboring image points after one of them has been temporally delayed. This operation is executed in each of two mirror-symmetrical half-detectors that operate with opposite sign. Summing the output of both half-detectors results in a directionally selective response of the full detector. Notably, the HRM precisely describes the observed optomotor behavior of walking beetles and walking and flying flies in algorithmic terms. Furthermore, the fundamental computations of the HRM can explain motion detection in different vertebrate species, including humans. In flies, directionally selective responses that closely match the predictions of the model are observed in the large tangential neurons of both large fly species and Drosophila. These cellular responses carry distinct signatures that derive from the correlative processing in the HRM12 (Reiff, 2010).

Because of the purely algorithmic nature of the HRM, no immediate conclusions about the underlying neuronal hardware can be drawn; different implementations of the model can result in similar output. To gain insight into the cellular implementation of the model, extracellular responses of the directionally selective H1 neuron where recorded while presenting apparent motion stimuli. Results of sequential stimulation of individual photoreceptor pairs (R1 and R6) of the same ommatidium led to the proposal that each input signal is split into an ON and an OFF channel that then fed into separate multipliers for the processing of brightness increments and decrements, respectively. However, interactions among brightness increments and decrements are inherent in the original HRM and have repeatedly been observed in behavioral optomotor responses and in the cellular responses of the H1 neuron21. Apparent motion stimuli with opposite polarity induce responses that report a reversal of the true direction of the stimulus, a phenomenon that is known in psychophysics as reverse-phi. If a neuron is assumed to perform a multiplication in a sign-correct manner, then this neuron's output signal should increase in a supra-linear way when both inputs increase as well as when both inputs decrease. No biologically plausible mechanism is known that could accomplish such a computation. A circuit was proposed that was inspired by the 'four-quadrant multiplier' used in analog signal processing: bipolar (both positive and negative signal components) input signals were half-wave rectified, resulting in only positive signals. These signals are subsequently processed in four separate multipliers accounting for all possible interactions (ON-ON, ON-OFF, OFF-ON and OFF-OFF). The output of the individual multipliers is then summed by a postsynaptic integrator in a sign-correct manner (from the perspective of this integrator). Thus, in contrast with a previous account, separate input channels for the processing of brightness increments and decrements cannot be excluded on the basis of responses of the integrating neuron to mixed input signals (Reiff, 2010).

In flies, the lamina monopolar neurons L1 and L2 are the largest and best-investigated second-order visual interneurons postsynaptic to the photoreceptors R1-R6. L1-L3 and one amacrine cell (amc) all express a chloride channel encoded by the ort gene, which is gated by the photoreceptor transmitter histamine. The processes of amacrine cells stay in the lamina, where they synapse onto L5 and where L4 receives input from L2 (and feeds back onto two more lateral L2 neurons). L4 and L5, as well as L1-L3, project to distinct layers in the medulla. Thus, five possible parallel processing streams (three direct channels, L1-L3; two indirect channels, L4 and L5) transmit information about brightness changes from the lamina to the medulla. Behavioral and genetic experiments suggest that L3 is involved in processing of ultraviolet light and in phototaxis. In contrast, L1 and L2 provide the major input to the motion detection circuitry in the medulla. Recordings of their dendritic membrane potential reveal nondirectional, strongly adapting responses in large fly species; dendritic voltage changes in L1 and L2 to a transient light pulse correspond to an inverted, high pass-filtered biphasic version of the voltage change recorded in photoreceptors that depolarize in response to light. The reported inhibitory current through histamine-gated chloride channels explains the hyperpolarizing ON response; however, it does not explain the excitatory OFF component at the end of a light pulse that has been observed in large flies. Depolarizing voltage responses to light-OFF have not been observed in Drosophila (Reiff, 2010).

Even though there have been electrophysiological studies on lamina monopolar cells and few other columnar neurons in Calliphora, the signals that are transmitted by lamina monopolar cells to neurons of the motion detection circuitry in the medulla could not be recorded so far for methodological reasons. This study addresses this problem in fly motion vision by investigating how L2 axon terminals in the medulla encode brightness changes in presynaptic intracellular calcium. Visually evoked Ca2+ is measured by a new method that employs optical recording of the genetically encoded calcium indicator TN-XXL targeted to L2 neurons and an interlaced visual stimulation technique (Reiff, 2010).

The data on L2 terminal Ca2+ corroborate the previously reported inversion and high pass-filtering and complete the processing by adding half-wave rectification of the brightness signal. Taking into account the role of Ca2+ in presynaptic vesicle release, it is proposed that L2 primarily transmits the information about brightness decrements to the motion detection circuit in the medulla (Reiff, 2010).

Dendritic recordings of L2 membrane potential in large flies show at least small depolarizing responses induced by light-OFF, suggesting that the underlying processing likely involves an amplification of the positive dendritic membrane potential and opening of voltage-activated Ca2+ channels in L2 terminals induced by light-OFF. Light-ON hyperpolarizes the L2 membrane potential, which might rapidly inactivate the calcium channels. Efficient calcium extrusion then likely mediates the observed rapid return of the calcium signal to baseline that is induced by light-ON (Reiff, 2010).

Nonlinear processing steps represent an important feature of second-order visual interneurons in flies and in the vertebrate retina; vertebrate ON- and OFF- bipolar cells preferentially relay either increments or decrements in brightness. However, half-wave rectification in bipolar cells is not complete and partly results from inhibitory interactions among ON and OFF channels. The increase of the time constant observed in L2 rescue flies suggests that interactions between different lamina cell types are involved in the generation of imperfectly half wave-rectified light-OFF calcium responses in L2 axon terminals. Such interactions are also suggested by the rich anatomical connections at the level of the dendrites in the lamina and at the level of the axon terminals in the medulla. Furthermore, given that L2 terminals transmit their main signal at light-OFF, other channels must exist for the signaling of brightness increments. Such ON and OFF signaling is a common motif in different animals and sensory modalities. Thus, although not necessary for Hassenstein-Reichardt-type computations, half-wave rectifying the input signals into parallel ON and OFF channels and multiplying each pair separately allows the outputs to be treated in a sign-correct manner. The devised imaging approach should pave the way for future studies that ultimately reveal the cellular implementation of the HRM of visual motion detection (Reiff, 2010).

In the eye, visual information is segregated into modalities such as color and motion, these being transferred to the central brain through separate channels. This study genetically dissected the achromatic motion channel in the Drosophila at the level of the first relay station in the brain, the lamina, where it is split into four parallel pathways (L1-L3, amc/T1). The functional relevance of this divergence is little understood. This study showed that the two most prominent pathways, L1 and L2, together are necessary and largely sufficient for motion-dependent behavior. At high pattern contrast, the two pathways are redundant. At intermediate contrast, they mediate motion stimuli of opposite polarity, L2 front-to-back, L1 back-to-front motion. At low contrast, L1 and L2 depend upon each other for motion processing. Of the two minor pathways, amc/T1 specifically enhances the L1 pathway at intermediate contrast. L3 appears not to contribute to motion but to orientation behavior (Rister, 2007).

Visual systems process the information from the environment in parallel neuronal subsystems. In higher vertebrates, for instance, the visual modalities of color, form, and motion are segregated at the level of the retina into separate channels. Similarly, insects have distinct sets of photoreceptors for motion and color. Investigating the motion channel in the Drosophila this study shows that at the next level below the eye, the lamina, the motion channel is again split into several functionally distinct parallel pathways (Rister, 2007).

Directional responses to visual motion have been intensely studied, predominantly in dipteran flies. They are provided by arrays of elementary movement detectors, the smallest motion-sensitive units that temporally compare the intensity fluctuations in neighboring visual elements (sampling units; see Anatomy of Peripheral Interneurons of the Fly's Visual System). Their neuronal implementation in flies is still unknown. In the rabbit retina, a candidate interneuron computing directional motion has been identified. The present study is confined to the input side of the movement-detection circuitry (Rister, 2007).

The compound eye of Drosophila is composed of about 750 ommatidia. Each of these contains eight photoreceptors (R1-8) that can be structurally and functionally grouped into two subsystems: six large photoreceptors (R1-6) mediate the detection of motion, whereas two small ones (R7, R8), together forming one rhabdomere in the center of the ommatidium, are required for color vision (Rister, 2007).

The lamina consists of corresponding units called neuro-ommatidia, or cartridges. These are the sampling units of the motion channel, whereas the color channel (R7, R8) bypasses the lamina cartridge to terminate in the second neuropil, the medulla. The lamina is anatomically and ultrastructurally known in exquisite detail. Its functional significance, however, is little understood (Rister, 2007).

In the lamina, the motion channel is split into four parallel pathways (see The Functional Role of L1 and L2 in Motion Detection). In each cartridge, the photoreceptor terminals are connected by tetradic synapses to four neurons, L1, L2, L3, and the amacrine cell α (amc; connecting to the medulla via the basket cell T1). The most prominent of these are the large monopolar cells L1 and L2. Their position in the center and their radially distributed dendrites throughout the depth of the cartridge suggest a key role in peripheral processing. This can be visualized by 3H-deoxyglucose activity labeling. Single-unit recordings of L1 and L2 in large flies so far have revealed only subtle differences between them. Their specific functional contribution to behavior is largely unknown (Rister, 2007).

Several hypotheses have been advanced. The loss of L1 and L2 and concomitantly of optomotor responses in the mutant Vacuolar medulla^KS74 had prompted a proposal that these cells were involved in motion detection. Later, however, it was claimed that L1 and L2 should be dispensable, because optomotor responses were still measured in flies that were assumed to have complete degeneration of L1 and L2 (Rister, 2007).

If indeed L1 and L2 mediate motion vision, are they functionally specialized or redundant? The latter is unlikely to be the whole answer, considering the differing synaptic relationships of the two neurons. For one, they have their terminals in separate layers of the medulla. Second, L2, but not L1, has feedback synapses onto R1-6. These might play a role in neuronal adaptation and could exert a modulatory influence on the photoreceptor output. It has also been suggested that L1 and L2 might be specialized to provide the respective inputs to the two branches of the elementary motion detector (EMD) (Rister, 2007 and references therein).

In Drosophila, L2 innervates and reciprocally receives input from a second-order interneuron, L4 that has two conspicuous backward oriented collaterals connecting its own cartridge to the neighboring ones along the x and y axes of the hexagonal array. In this network, the L2 neurons are connected to the L4 neurons of two adjacent cartridges, and the L4 neurons are directly connected to all six neighboring L4s. The significance of this circuitry is not yet understood. It has been speculated that the L4 network might be specialized for front-to-back motion, the prevalent direction in the visual flow-field of fast forward-moving animals (Rister, 2007).

Using the two-component UAS/GAL4 system for targeted transgene expression, single interneurons or combinations of them, were manipulated in all lamina cartridges. To study whether a particular pathway was necessary for a given behavioral task, their synaptic output was blocked using the temperature-sensitive allele of shibire, shi^ts1. In addition, the inverse strategy was adopted, studying whether single lamina pathways are sufficient for mediating the behavior in the same experimental context. Using a mutant of the histamine receptor gene outer rhabdomeres transientless (ort) that has all lamina pathways impaired, the wild-type ort-cDNA was expressed in chosen types of lamina interneurons known to receive histaminergic input from R1-6. Testing necessity and sufficiency it is now possible to start to relate the structural organization of the lamina to visually guided behavior (Rister, 2007).

This study reports the first steps into the genetic dissection of the neuronal circuitry mediating motion and position detection, the main perceptual processes of visual orientation behavior and gaze control. Some basic properties emerge: two subsystems, the L1 and L2 pathways, were identified that both mediate directional motion independently of each other. A third subsystem, the L3 pathway, may provide position information for orientation. The two motion pathways were remarkably redundant under a broad range of visual conditions, in line with the general observation that motion detection is a very robust phenomenon. To detect an impairment with only one of the pathways remaining intact, one had to drive the system to its operational limits (Rister, 2007).

Clearly, the L1 and L2 pathways play the principal role in motion detection. Flies without the L3 and amc/T1 pathways are fully motion competent, as far as the present analysis can reveal. In contrast, flies with both L1 and L2 blocked are motion blind using optomotor yaw torque responses, motion-driven head movements, and landing response as criteria. This result is based on three independent driver lines and is in line with findings on the unmapped mutant Vam^KS74. As the L2 pathway mediates optomotor responses at very low stimulus strengths, it would not be surprising if few functional L2 neurons were sufficient to have mediated the response, like when there are few residual ommatidia in sine oculis mutant flies (Rister, 2007).

The relation between the L1 and L2 pathways is of particular interest. Throughout most of the pattern contrast range either pathway alone provides full-sized motion responses. At high pattern contrast, the two pathways are redundant, while in the intermediate contrast range they are specialized for front-to-back and back-to-front motion, respectively. Only at the low end of the contrast range do the two pathways depend upon each other (Rister, 2007).

In natural habitats of insects, intermediate pattern contrasts prevail. It is in this contrast range where the L1 and L2 pathways show unidirectional sensitivity for back-to-front and, respectively, front-to-back movement. A specialization of L1 and L2 for these two directions of motion had been proposed almost four decades ago. Different strengths of the respective optomotor responses in large flies and reduced responses for only one of the two directions in Drosophila mutants had suggested separate arrays of EMDs for the two directions. The new data are compatible with at least two models. In the first one, which is the sparser one, either neuron would serve its array of unidirectional EMDs: L1 an array for back-to-front, L2 one for front-to-back motion. The model would entail crosstalk between the two pathways at high pattern contrast, most likely in the medulla, and a more complex interaction between them at the low end of the pattern contrast range. The second model envisages EMDs for both directions to be served by either pathway. In this case, no crosstalk would be required at high pattern contrast, but one would be in need of additional explanations for the unidirectional responses in the intermediate contrast range (Rister, 2007).

The asymmetry of the L4 collaterals and the close interaction between L4 and L2 are an intriguing structural correlate of the unidirectional contrast sensitivity of the L2 pathway. No equivalent network with opposite polarity has been detected in the lamina for the L1 pathway, but might still be found in the medulla. As long as no physiological data exist of L4 in Drosophila, it is not possible to tell whether the L4 network provides lateral inhibition, lateral pooling, or the second input pathway for an array of front-to-back EMDs (Rister, 2007).

The L2 pathway is more sensitive to pattern contrast and low light intensity than the L1 pathway. As this distinction was observed with three independent genetic variants, an artifact due to the genetic methods is unlikely. The enhanced contrast sensitivity of L2 might be attributed to the feedback synapses of L2 onto photoreceptors R1-6, possibly providing some kind of gain control, or also to the L4 network. Enhancing sensitivity for front-to-back motion could be useful for fast flying animals, as this type of flow field prevails during fast forward flight. How these differences between the two pathways at low light intensity and pattern contrast show in flight behavior when both pathways are operating remains to be investigated (Rister, 2007).

Somewhat surprisingly, the two lamina pathways seem not to be differentiated for speed or contrast frequency. Possibly, only one array of EMDs might exist for each direction (sparse model) and the two may have to be tuned the same. Genetic intervention in the lamina as examined in this study obviously does not affect the tuning of EMDs. This supports the view that motion processing is located proximal to the lamina (Rister, 2007).

At high pattern contrast, the L1 and L2 pathways are redundant. L1 and L2 both mediate motion sensitivity in both directions. Bidirectionality at high contrast can be interpreted as crosstalk between two unidirectional pathways. This could be a property of the regular circuitry or due to wiring defects in the absence of neural activity in one of the pathways during development. The latter explanation is rather unlikely. In the L2-GAL4/shi^ts1 flies, about equal back-to-front and front-to-back responses were observe at m = 10% pattern contrast, whereas the L1-GAL4 ort⁺ rescue flies at this pattern contrast respond only to back-to-front motion. Why should the permanently low neural activity in the L2 pathway during development (caused by the mutated histamine receptor) render an originally bidirectional L1 pathway more unidirectional (Rister, 2007)?

The anatomical differences between the L1 and L2 pathways had prompted the speculation that the splitting of the signal from R1-6 into two pathways could correspond to the delayed and nondelayed input channels of the EMD. The present analysis refutes this idea as an overall explanation of the duplicity of the large lamina monopolar neurons. Either pathway alone mediates motion stimuli at high and intermediate pattern contrast. Hence, both neurons can serve the delayed as well as the nondelayed branch of the EMD. Yet, at the low end of the pattern contrast range of wild-type this is different. Neither L1 nor L2 alone mediate optomotor responses. The two pathways need to interact to provide motion sensitivity. Conceivably, by combining two unidirectional EMDs of opposite polarity one can more than additively improve their signal-to-noise ratio. Indeed, the original motion-detector model contains a subtraction of the signals of the two antidirectional EMDs to eliminate the dependency of the output upon light intensity. Alternatively, at this very low pattern contrast L1 and L2 might, after all, specialize to serve the delayed and respectively nondelayed branch of the EMD (Rister, 2007 and references therein).

Finally, it is not yet clear whether the motion response based on the interaction of the L1 and L2 pathways operating at low pattern contrast is uni- or bidirectional. At the lowest contrast measuree (m = 5%), no directional preference was found in the control flies, although the overall response was already reduced to less than 50% (Rister, 2007).

The high sensitivity for pattern contrast of the L2 pathway is paralleled by a low threshold for light intensity. At the lowest intensity measured at which wild-type is still responsive, the L2 pathway is not only necessary but also fully sufficient, implying again that under these conditions the L2 neurons serve both input channels to the EMD. It remains open whether at even lower intensities an interaction between L1 and L2 might be found as is the case with low pattern contrast (Rister, 2007).

The data indicate that the special trade-off at low light intensity, whereby sensitivity is gained at the expense of acuity, can use the L1 pathway as input. The mechanism is supposed to pool the signals of many visual elements for the delayed as well as the nondelayed channels of an array of EMDs with large sampling base. In the current experiments, the L1 pathway at the broad pattern wavelength (λ = 36°) is about as sensitive as the L2 pathway at λ = 18°. This shows that the role of the L1 and L2 pathways in pooling is not yet understood well. Lower light intensities may reveal an involvement of also L2 in pooling (Rister, 2007).

Recently, it has been shown that the T1 neuron has no conventional chemical synaptic output sites in the medulla as judged by its ultrastructure. Hence, it is an open question whether and how shi^ts1 expression in T1 might block a presumed nonsynaptic output from T1. Expressing shi^ts1 at the restrictive temperature has, on the other hand, been found to perturb the organization of microtubules in the expressing photoreceptor cells. Moreover, it is likely that the processing of other membrane vesicles and hormone secretion at the Golgi apparatus are affected as well. The data consistently show an effect of shi^ts1 expression in T1 neurons at the restrictive temperature. Optomotor responses are reduced at intermediate pattern contrast, if L2 is blocked as well. The mechanism mediating this effect is not known (Rister, 2007).

Assuming shi^ts1 to block T1 output, it is concluded that the amc/T1 pathway supports the L1 pathway at intermediate pattern contrast, at which the response of the L1 pathway just reaches saturation. Under these conditions, disturbance of T1 reduces the gain of the system and shifts the saturation range to higher contrast levels. The finding that saturation is eventually reached could be explained by the assumption that neurons like L5 with a presumed higher response threshold might be added to the system at still higher pattern contrast. In line with this hypothesis is the finding that the on-off units in the outer chiasm of large flies, which might correspond to L5, did not respond to contrasts smaller than 10% in electrophysiological recordings. The rather subtle effect of blocking T1 is taken to indicate that the stimulus conditions for T1 function have not yet been properly defined. It is unlikely that T1 functions were not observed because shi^ts1 did not block T1 output. Expression of DTI and Kir^2.1 in T1 neurons did not show a more substantial effect (Rister, 2007).

In contrast to earlier assumptions, evidence has been accumulating that orientation toward landmarks does not necessarily require motion. In Musca, position-sensitive torque responses could be elicited in stationary flight, if the luminance of a stationary vertical stripe was sinusoidally modulated (local flicker). In Drosophila, torque responses toward stationary dark objects (δ = 5°) have directly been documented (Rister, 2007 and references therein).

In the present study, neuronal pathways mediating motion and position detection have been genetically separated. This study has shown that motion-blind animals are still able to approach landmarks, corroborating the notion that motion vision is not essential for the detection and fixation of a stationary object. In contrast, the data also suggest that motion detection improves the fixation of landmarks, especially when these are narrow or have a reduced contrast. Note, that in this paradigm testing freely walking flies motion vision was not excluded experimentally. Obviously, in visual orientation both neuronal subsystems are at work, and genetic dissection will help to unravel their interaction (Rister, 2007).

In flies having the entire motion channel (R1-6) blocked, the color channel (R7/R8) alone provides basic position information. With only L1 and L2 blocked, flies are still completely motion blind in all paradigms tested, but their orientation behavior is distinctly superior to that of flies with the entire motion channel blocked. Apparently, elements among the remaining lamina pathways improve landmark orientation as mediated by R7 and R8. Given that L5 was blocked in one of the driver lines without an additional impairment of orientation behavior, it is assumed that at the conditions of the paradigm L5 did not substantially contribute to orientation behavior (Rister, 2007).

Blocking T1 in addition to L1 and L2 caused no further reduction of the orientation response. Hence, the amc/T1 pathway seems not to contribute significantly to this behavior either. This means that the L3 pathway, possibly interacting with the R7 and R8 pathways in color vision, may mediate orientation behavior, since flies without functional L1, L2, and amc/T1 still show better orientation behavior than flies with the entire R1-6 channel blocked. The residual orientation behavior in flies without functional L1 and L2 is very sensitive to a reduction in object contrast. This suggests that the underlying phototactic or tropotactic orientation mechanism might integrate the visual input over large parts of the visual field, reducing the apparent pattern contrast of small targets below threshold. This spatial integration might occur at any level in the system (Rister, 2007).

In summary, genetic dissection indicates that position detection might be as robust and redundant as motion vision. The color channel (R7/R8), L1, L2, and L3 all contribute to position detection. Presumably, single pathways are sufficient for this task. Detecting a singularity in space may require a less sophisticated neural mechanism than motion detection based on a temporal comparison of signals from neighboring visual elements (Rister, 2007).

Applying circuit genetics, this study has found the peripheral neuronal network of the fly optic lobe is functionally more complex than what previous behavioral, anatomical, and electrophysiological studies on wild-type animals had revealed and, maybe, what the early pioneers of the 1950s and 1960s had envisaged. Still, with this new approach, the fly optic lobe once again proves to be a uniquely suited case study for gaining basic insights into the neuronal mechanisms of visual information processing and, more generally, for the comparison of structure and function in neural networks (Rister, 2007).

Recent work in Drosophila has uncovered several neighboring classes of sleep-regulatory neurons within the central complex. However, the logic of connectivity and network motifs remains limited by the incomplete examination of relevant cell types. Using a recent genetic-anatomic classification of ellipsoid body ring neurons, this study conducted a thermogenetic screen in female flies to assess sleep/wake behavior and identified two wake-promoting drivers that label ER3d neurons and two sleep-promoting drivers that express in ER3m cells. Intersectional genetics was used to refine driver expression patterns. Activation of ER3d cells shortened sleep bouts, suggesting a key role in sleep maintenance. While sleep-promoting drivers from the mini-screen label overlapping ER3m neurons, intersectional strategies cannot rule out sleep regulatory roles for additional neurons in their expression patterns. Suppressing GABA synthesis in ER3m neurons prevents postinjury sleep, and GABAergic ER3d cells are required for thermogenetically induced wakefulness. This study used an activity-dependent fluorescent reporter for putative synaptic contacts to embed these neurons within the known sleep-regulatory network. ER3m and ER3d neurons may receive connections from wake-active Helicon/ExR1 cells, and ER3m neurons likely inhibit ER3d neurons. Together, these data suggest a neural mechanism by which previously uncharacterized circuit elements stabilize sleep-wake states (Singh, 2023).

This study applied novel community detection methods to analyze the synapse-level reconstruction of an adult female Drosophila melanogaster brain containing >20,000 neurons and 10 million synapses. Using a machine-learning algorithm, the most densely connected communities of neurons were found by maximizing a generalized modularity density measure. The community structure was resolved at a range of scales, from large (on the order of thousands of neurons) to small (on the order of tens of neurons). The network was found to be organized hierarchically, and larger-scale communities are composed of smaller-scale structures. These methods identify well-known features of the fly brain, including its sensory pathways. Moreover, focusing on specific brain regions, it was possible to identify subnetworks with distinct connectivity types. For example, manual efforts have identified layered structures in the fan-shaped body. These methods not only automatically recover this layered structure, but also resolve finer connectivity patterns to downstream and upstream areas. A novel modular organization was found of the superior neuropil, with distinct clusters of upstream and downstream brain regions dividing the neuropil into several pathways. These methods show that the fine-scale, local network reconstruction made possible by modern experimental methods are sufficiently detailed to identify the organization of the brain across scales, and enable novel predictions about the structure and function of its parts (Kunin, 2023)

Understanding how neural circuits underlie behaviour is challenging even in the connectome era because it requires a combination of anatomical and functional analyses. This is exemplified in the circuit underlying the light avoidance behaviour displayed by Drosophila melanogaster larvae. While this behaviour is robust and the nervous system relatively simple, the circuit is only partially delineated with some contradictions among studies. This study devised trans-Tango MkII, an offshoot of the transsynaptic circuit tracing tool trans-Tango, and implement it in anatomical tracing together with functional analysis. Neuronal inhibition was used to test necessity of particular neuronal types in light avoidance and selective neuronal activation to examine sufficiency in rescuing light avoidance deficiencies exhibited by photoreceptor mutants. These studies reveal a four-order circuit for light avoidance connecting the light-detecting photoreceptors with a pair of neuroendocrine cells via two types of clock neurons. This approach can be readily expanded to studying other circuits (Sorkac, 2022).

Flexible behaviors over long timescales are thought to engage recurrent neural networks in deep brain regions, which are experimentally challenging to study. In insects, recurrent circuit dynamics in a brain region called the central complex (CX) enable directed locomotion, sleep, and context- and experience-dependent spatial navigation. This study describes the first complete electron-microscopy-based connectome of the Drosophila CX, including all its neurons and circuits at synaptic resolution. This study identified new CX neuron types, novel sensory and motor pathways, and network motifs that likely enable the CX to extract the fly's head-direction, maintain it with attractor dynamics, and combine it with other sensorimotor information to perform vector-based navigational computations. Numerous pathways were identified that may facilitate the selection of CX-driven behavioral patterns by context and internal state. The CX connectome provides a comprehensive blueprint necessary for a detailed understanding of network dynamics underlying sleep, flexible navigation, and state-dependent action selection (Hulse, 2021).

Recent physiological and anatomical studies at the light and EM level have highlighted strong links between circuit structure and function in the adult fly central brain. These links have proven to be valuable both for generating hypotheses and for experimentally testing them. This recent history gives reason to expect that connectomics will continue to accelerate studies of circuit function. CX circuits, in particular, are thought to be involved a wide variety of flexible, context-dependent behaviors. This work provides a detailed description of CX neuron types and circuits, with a particular focus on extracting and examining network motifs from the perspective of what is currently known about CX circuit function and CX-mediated behavior. Many repeating motifs were found, raising the possibility that an understanding of the computational roles of some of these may generalize to others. Some of these motifs match those that have been proposed previously to implement ring attractors for head direction computation. Others seem suitable for gain control in multiple structures. And still others seem to be ideal for vector computations that would be required for robust navigational behaviors. This study found that information from the CX's output neuron types is broadcast through fairly segregated pathways that are distributed across the brain, not just to premotor centers but to sensory regions and, importantly, back into the CX itself. In the sections that follow, some functional implications of these motifs and of other results from this analyses are described. These functional implications are derived not just from connectomic analyses and the historical precedent of structure predicting function in many different neural circuits, but also on published physiological and behavioral studies. Testing the hypotheses are outlined below will require a long series of functional experiments, but the connectome provides an invaluable guide for the design and prioritization of such experiments (Hulse, 2021).

The value of EM-level connectomes in understanding the function of neural circuits in small and large brains is widely appreciated. Although recent technical advances have made it possible to acquire larger EM volumes and improvements in machine learning have enabled high-throughput reconstruction of larger neural circuits, the step from acquiring a volume to obtaining a complete connectome still requires considerable human proofreading and tracing effort. As part of this analysis of the CX connectome, it was found that although increased proofreading led to an expected increase in the number of synaptic connections between neurons, it did not necessarily lead to significant changes in the relative weight of connections between different neuron types. While it is important to note that comparisons were made between the hemibrain connectome at fairly advanced stages of proofreading in the CX, the results do suggest that it may be possible to obtain an accurate picture of neural circuit connectivity from incomplete reconstructions. It may be useful for future large scale connectomics efforts to incorporate similar validation steps of smaller sample volumes into reconstruction pipelines to determine appropriate trade-offs between accuracy and cost of proofreading. Connectivity and neural processing beyond the typical synapse. Although this study provides a detailed description of the CX's hundreds of neuron types, recurrent networks and pathways, there is still more information that could be extracted from the CX connectome. The CX is innervated by a large number of modulatory and peptidergic neurons, many unidentified and almost all of unknown function. These neurons likely significantly modulate the function of recurrent networks in ways that few studies address. Knowing their identities - whether by matching LM images of known neuron types to their EM counterparts in the hemibrain or by advances in machine-learning based identification of neuromodulator/neuropeptide and receptor types - would help guide circuit studies into context- and internal-state-dependent processing in the CX (Hulse, 2021).

A large number of CX neuron types that make T-bar and E-bar synapses in CX structures also send projections to other structures in which they make no such synaptic connections. This study investigated these projections in more detail and consistently found dense core vesicles in these otherwise nearly synapse-free processes. Although the involvement of some of these neuron types, for example PFGs neurons, in sleep-wake circuits suggests a plausible scenario for state-dependent modulation of CX circuits, such explanations are not easily available in all cases (Hulse, 2021).

It is important to note that use of relative weights to assess synaptic strength was informed by observed correlations between synapse counts and the area of synaptic contact in larval Drosophila, and the dependence of synaptic strength on synaptic surface area, at least in the mammalian neocortex. It is expected that relative weights provides only an approximate measure of true functional strength. Further, synapses across the Drosophila brain undergo structural changes depending on the time of day, sleep, activity and the animal's specific experiences; properly accounting for the impact of such factors on connectivity patterns would require comparisons across multiple connectomes. Also, as previously discussed, the hemibrain connectome does not capture glial networks or gap junctions. Despite all these limitations, the identification of chemical synapses between CX neurons and examining their relative weight based on synapse counts allowed extraction of network motifs that make strong predictions about function (Hulse, 2021).

Many flexible, goal-driven behaviors unfold over longer durations than fast reflexive responses and are robust to the temporary loss of sensory cues directly associated with the goal. Desert ants, for example, use path integration to return to their nests after long foraging trips in relatively featureless landscapes, and mammals use working memory to perform delayed match-to-sample tasks. For such behaviors, brains are believed to rely on intermediate representations and neural dynamics that persist or update even in the absence of direct sensory inputs. Such persistent representations have long been believed to be generated, updated and maintained by recurrent attractor networks. These more abstract intermediate representations also enable disparate sensory and self-motion cues of different modalities to be registered to a shared reference frame. A path integrating ant, for example, may use such a representation to register cues from polarized light, visual optic flow and proprioception and all Diptera likely need to register visual and haltere input, as flesh flies do. Ultimately, information in these reference frames must still be dynamically converted to a body-centered reference frame for situation-appropriate action. Decades of experimental work in a variety of species have led theorists to propose gain fields for the implementation of such coordinate transformations, but the predicted neural circuit connectivity has not been directly identified. In addition, for an animal to learn from experience, any past associations of the current context with good or bad outcomes must be recalled and used to modify neural dynamics at the level of such intermediate representations, raising computational questions that have been explored in the field of reinforcement learning. The repertoire of flexible navigational behaviors that insects display suggests that their small brains may solve many of these computational challenges. Further, insect circuits may have evolved solutions to these problems that resemble those proposed by theorists to account for neural response properties in mammalian circuits (Hulse, 2021).

Flies in particular use short-term memory to orient towards the last-known positions of attractive visual beacons that have disappeared. They learn about their body size and use that information when attempting to cross gaps. They learn to avoid heat punishment by using visual patterns around them to orient to safety. Although they are not central place foragers like bees and ants, they are capable of returning to a spot of food even when exploring their surroundings in darkness, and of remembering visual landmarks to navigate to safe spots in an otherwise hostile open space. The CX is thought to be essential for many of these behaviors. In the sections that follow, how the patterns of connectivity revealed by the CX connectome may enable the neural dynamics, coordinate transformations, and learning-induced changes in action selection associated with meeting the computational challenges of some of these behaviors will be discussed (Hulse, 2021).

Head direction representations enable an animal to flexibly rely on a variety of different cues, including self-motion, to orient. Work in Drosophila and other insect species has established that the CX builds a stable head direction representation using information from ring neurons, which convey directional sensory cues, such as polarized light, visual landmarks, and wind direction. In Drosophila visual head direction information reaches the CX via the anterior visual pathway, which appears to convey different visual information in separate, parallel 'channels'). Some 'channels' of this pathway have been characterized functionally, while the function and sensory tuning of other groups of neurons remains elusive. For example, most of the ring neurons (and their inputs) in the superior bulb (BU) are spatiotemporally tuned to visual features with some degree of orientation preference and the pathway through the anterior BU appears to be dedicated to polarization signals. In contrast, little is known about the role of ring neurons that get their inputs in the inferior BU in informing the head direction representation. Wind stimuli reach the compass circuitry through separate input pathway via the LAL and it is unknown whether other sensory modalities are conveyed through this route (Hulse, 2021).

Connectivity-based analysis suggests that there are 20 ring neuron types, 18 of which receive inputs via the anterior visual pathway. In contrast, an anatomical and developmental characterization of ring neurons found only 11 distinct morphological types. Notably, the current connectome-based typing likely represents a subdivision of the previously suggested types rather than a drastic reorganization. Given that past neurophysiological studies have only tested tuning to a relatively small number of sensory stimuli, it remains to be seen how many functionally distinct input types exist (Hulse, 2021).

The connectome reveals mechanisms by which sensory stimuli are integrated to inform the fly's head direction estimate. The findings suggest that different cues exert differing levels of influence on the ellipsoid body-protocerebral bridge-gall neuron (EPG) neurons that carry the head direction representation. each EPG (compass) neuron arborizes in a single EB sector ('wedge'). A prioritization of certain sensory cues is reflected in the relative locations of synaptic input from different sensory streams onto the EPG dendrites in the EB, in the relative weight of those inputs, in the feedback that some ring neuron pathways receive from the EPG neurons, and in the relative weight of across-type inhibition from some ring neuron types onto others. The implicit hierarchy of ring neuron inputs to the fly compass indicates that the EPG head direction representation preferentially tethers to environmental references that are likely to indicate a global direction (Hulse, 2021).

Bright visual landmarks, for example, may originate from celestial bodies such as the sun, but they could also be generated by local terrestrial objects. By contrast, a polarization pattern in the sky, if available, represents a reliable global reference, which might explain the observed circuit motifs that suggest the preferential use of polarization cues to update the fly's head direction representation. However, the relatively high connection strength between ER4m and EPG neurons may also arise from this fly not being exposed to polarized light stimuli. Such deprivation could have prevented these connections from being subjected to the synaptic depression that other visual pathways may have experienced (but note that there is no evidence yet for long-term structural changes at any of these synapses) (Hulse, 2021).

Particularly when navigating over long distances, skylight cues allow the head direction representation to be tethered to global landmarks such as the sun and to the polarized light patterns of the sky. Indeed, polarized light e-vector information has long been thought to be important for the determination of sky-compass-based head direction in many insects. A dorsal band of the insect eye called the dorsal rim area is structurally specialized for the detection of polarized light e-vectors in the sky. Despite their comparatively small dorsal rim area, flies can also use polarized light cues to determine their heading. Sensory information about the celestial polarization pattern reaches the Drosophila CX1 via a dedicated pathway to the ER4m neurons. Although only 5 ER4m neurons from each hemisphere show strong tuning to e vector orientation, this tuning collectively covers a large part of the 180° range of possible e vector orientations. However, in contrast to the position of the sun, the 180° symmetric polarized light patterns do not immediately provide the ability to distinguish a specific direction from one directly opposite to it (Hulse, 2021).

The CX connectome suggests that the fly's compass may have evolved a solution to this problem. For the polarization-tuned ER4m neurons, it was observed that synapse numbers to EPG neurons varied smoothly along the circumference of the EB, but with mirror-symmetric profiles for ER4m neurons from the left and right hemisphere, respectively. If synapse counts correlated with synaptic strength, this would result in stronger connections from ER4m neurons of the left hemisphere for EB wedges on the right half of the EB and stronger connections from the right hemisphere to the left half of the EB. This structure was even more clearly revealed when the pairwise correlation of EPG neurons according to their ER4m inputs: all EPG neurons on the right side of the EB were positively correlated with each other, while being anticorrelated with those on the left side, and the inverse pattern was observed for the left EPG population. Given that polarized light has a 180° symmetry, this connectivity pattern0 may allow the fly to generate a complete, 360° head direction representation from polarized light input (Hulse, 2021).

One possible mechanism by which this could be achieved hinges on the geometry of the fly's polarization sensors in the dorsal rim area and how it interacts with the natural polarization5 pattern of the sky. The receptive fields of the fly's polarization sensors in the left and right eye face the contralateral celestial hemisphere and tile a small strip along the rostral-caudal axis of the fly. Along this strip tuning to e-vectors varies continuously and covers nearly the full 180° range of possible e-vector orientations. Given the naturalistic celestial polarization pattern, the geometry of the slightly curved receptive field 'strip' might act as a rough 'matched filter', such that neurons in the dorsal rim area on the side of the sun (facing the contralateral sky) are systematically more strongly activated than those on the side facing away from the sun. The all-to-all inhibition between left and right ring neurons in the EB may then systematically select either the left or the right ring neurons to tether the head direction depending on which direction the fly is facing relative to the current position of the sun, thus disambiguating the 180° mirror 06 symmetry in the polarization signal (Hulse, 2021).

In locusts, TL-neurons, homologs of the fly's ring neurons, and (protocerebral bridge) PB neurons have been shown to exhibit matched-filter like tuning to the full-sky polarization patterns generated by the sun. A corollary of these studies is that individual TL neurons in the locust have receptive fields that span large parts of the sky. Indeed, the area of the sky that is sampled by photoreceptors in the dorsal rim area is significantly larger in locusts compared to flies, and it is plausible that further sensory processing along the anterior visual pathway toward the CX differs between species as well, in which case different insects might employ different strategies for disambiguating polarized light stimuli (Hulse, 2021).

The mechanism described above would not require that the sun be directly visible, but it might still be beneficial to have ring neurons that have multimodal tuning to both polarized light and sun-like stimuli. Such cells have been described in other insects. While this has yet to be demonstrated experimentally, multimodal ring neurons tuned to both visual features and polarized light e-vector orientation may also exist in flies. A previous study reported polarization-tuned neurons in the superior bulb, where tuning to bright features has also been observed (Hulse, 2021).

Besides visual cues, mechanosensory wind stimuli can drive the fly's head direction system in the EB. Information about wind direction reaches the EB via ring neurons that arborize in the lateral accessory lobe (LAL). Wind tuning has been demonstrated in both ER3a and ER1 neurons, although only ER1 neurons were able to update the head direction estimate. Analysis of the connectome suggests that both the ER1 and ER3a neuron populations consist of multiple types with distinct inputs. Only ER1_b and ER3a_b neurons got strong inputs from cells that are believed to be the wind-sensitive LAL138 WL-L) and WPN neurons. The connectivity of these two ring neuron types onto EPG neurons, with strong connections from ER1_b but no connections from ER3a_b neurons, is consistent with the observation that ER1 but not ER3a neurons can drive the head direction representation. It is also noteworthy that in the EB, ER1_b neurons deviate from the within-type all-to-all inhibition motif that all other ring neurons show in the EB. A possible reason is that an accurate mapping from ER1 neuron activity to a head direction representation requires pooling information from multiple ring neurons at once. This analysis also suggests that ER1_b input to the EPG neurons is suppressed by ER1_a neurons, but it is presently unknown whether ER1_a neurons also encode wind direction or whether these neurons are tuned to a different stimulus. ER1_a and ER1_b inputs in the LAL are distinct and unfortunately little is known about the inputs ER1_a receives (Hulse, 2021).

Whether over short or long distances, olfactory cues are strong indicators of good food sources. Flies are known to fly upwind when they encounter an appetitive odor, a strategy also employed by other insects navigating to an odor source. A robust navigational strategy would allow an insect to maintain the same heading using other cues even if the wind were to transiently die down. Based on the proximity of different ring neuron inputs to the putative spike initiation sites of EPG neurons, the head direction representation is likely to be strongly tethered to wind direction by input from ER1_b neurons (and perhaps also ER1_a neurons, although their function is currently unknown). If visual cues are flexibly mapped onto head direction representation using this wind direction input as a reference, the EPG compass could allow the fly to preserve its heading using those cues even in the absence of wind (Hulse, 2021).

The relative importance of synapse location in determining the cues to which the EPG compass tethers will only be clear with in-depth investigations of EPG neuron biophysics. More broadly, future studies of ring neuron and EPG interactions should provide an implementation-level understanding of a variety of computations related to dynamic multisensory integration and the resolution of conflicts between cues of different reliability (Hulse, 2021).

Recent studies have proposed an important role for fast-timescale, short-term plasticity of synaptic connections between ring and EPG neurons in enabling the EPG compass to quickly adapt to different sensory settings. The connectome suggests that the ring neuron network may also preselect more salient cues for the compass through all-to-all inhibitory connectivity within each type. The precise impact of all-to-all inhibition on the ring neuron network's preprocessing of localizing cues that are used to generate the head direction representation will depend on the timescale of the inhibitory conductance, which is as yet unknown. If the inhibitory conductance is fast, all-to-all inhibition would create winner-take-all dynamics in which a few ring neurons receiving the strongest inputs effectively shut down all other ring neurons. In sensory settings characterized by a single dominant sensory cue, such as the sun or polarized light e-vector orientation in a desert landscape during the day, only a handful of ring neurons with appropriately tuned receptive fields would be active within each type for any particular head direction. Fast all-to-all inhibition in this setting would enhance the activity of the most dominant ring neuron within each type and would minimize the impact of noise from the others, which might otherwise disrupt the stability of the EPG compass (Hulse, 2021).

However, stable and unique heading representations are also generated within scenes with multiple strong cues, such as within a forest or when walking on the branches of a tree, as long as the two-dimensional arrangement of cues allows for a unique determination of heading. In the presence of multiple salient cues, it would be expected that multiple ring neurons would respond with comparable strength for any given heading of the fly and several to respond weakly to any additional visual cues ('clutter') in the scene. Although fast all-to-all inhibition in this scene would still filter out these weaker responses, it could allow multiple, strongly responsive ring neurons to remain active for each heading. A slower inhibitory conductance would, in this situation, induce oscillatory spiking dynamics between these multiple 'winners', a situation that has been referred to as 'winnerless competition', and that has been suggested to be useful for sequential memory (Hulse, 2021).

Ring neuron responses are not determined purely by sensory cues. These neurons appear to be modulated by state, maintain a baseline level of activity, and may be biophysically configured to support oscillatory population activity linked to sleep need. The many additional inputs that many ring neuron types receive in the BU provide clues as to how the activity of these neurons might be modulated by the fly's behavior and its internal state. The visually-tuned superior BU ring neurons primarily receive input from a large interhemispheric anterior optic tubercle (AOTU) neuron, which may mediate dynamic stimulus selection through delayed contralateral inhibition. The same group of ring neurons also shows changes in activity with the fly's behavioral state (flight versus walking) and indeed many of these neurons receive input from the dopaminergic ExR2 neuron that has been linked to changes in the fly's motor activity. A different set of ring neurons that receive their inputs in the inferior BU receives strong inputs from two ExR neurons -ExR1 and ExR3- that have been linked to the control of sleep, and may gate sensory stimuli according to the fly's behavioral state (Hulse, 2021).

The fly's head direction representation tethers to directional sensory cues conveyed by ring neurons, but is also updated by self-motion cues and is maintained across periods of immobility. Strong experimental and theoretical evidence suggests that the representation is maintained by a ring attractor network, which includes at least some of the recurrently connected columnar neurons that link the EB and the PB: the EPG, PEN_a, PEN_b and PEG 24 neuron types. The patterns of connectivity between individual neurons of these types are consistent around the entire EB and across the length of the PB. Similarly, the broad connectivity patterns of individual neurons within these types to tangential neurons -Δ7 neurons in the PB and different types of ring neurons in the EB- are similar across these structures. Notably, however, two distinct classes of neurons (EPGt neurons and the P6-8P9 neurons) only innervate the edges of the network; both innervate the outer glomeruli of the PB, and the EPGt neurons also innervate the corresponding wedges in the EB. These types may help to stitch together what might otherwise be a discontinuity in the ring attractor network. Indeed, the EPGt neurons in the left and right PB arborize in wedges in the EB that lie directly in between the wedges occupied by the EPG neurons on either side of the potential discontinuity (in PB glomeruli 1 and 8). The EPGt neurons may therefore represent angular positions halfway between the edge angles, bridging the gap. It is noted, however, that the Δ7 neurons and the P6-8P9 neurons that output in these outer glomeruli each receive unique input from different sets of EPG neurons, making it hard to assign a clear corresponding EB angle to glomerulus 9 (Hulse, 2021).

The hemibrain connectome further allowed identification of several neuron types and connectivity motifs that are likely involved in the network's function, but whose roles await experimental investigation. Many of these additional types are tangential neurons. Some of them appear to provide additional sources of inhibition, potentially regulating overall network activity. The ER6 neurons, for example, receive input from the EPG and PEG neurons in the GA and send outputs to the PEG, EL and PEN_b neurons in the EB, potentially modulating the EPG-to-PEG-to-PEN_b-to-EPG feedback loop. Furthermore, many of the ExR neurons make connections to and receive input from EB columnar neurons. The PB receives neuromodulatory input from the dopaminergic LPsP neurons and the octopaminergic P1-9 neurons. The LPsP neurons may enable changes in synaptic strength in the PB. Such plasticity in the PB has been suggested to allow flies to calibrate their directed movements to their body size. In sensory brain regions, octopaminergic neurons are known to modulate neuronal conductances based on the fly's behavioral state, and it is an open question whether the P1-9 neurons play a similar role in the PB. Notably, the Δ7 neurons connect recurrently to each other in the PB, but the function of this recurrence is unknown. One possibility is that recurrent Δ7 connections may increase the stability or robustness of the ring attractor network (Hulse, 2021).

Two additional classes of columnar neurons also contact the ring attractor network: EL and IbSpsP. The connectivity pattern of EL neurons in the EB is remarkably similar to that of the EPG neurons, but their function is unknown. In the PB, the IbSpsP neurons bring input into specific glomeruli from regions associated with premotor functions, potentially allowing them to exert an influence on the dynamics of the bump in the PB (Hulse, 2021).

The ring attractor network described above generates a single activity bump in the EB that encodes the fly's head direction. The connectome allowed following this activity bump through the CX as it gets duplicated, reformatted, recombined, and, finally, read out. In the process, network motifs were discovered that seem ideally suited for performing vector computations. These motifs place constraints on the network's computational capacity and inspire conceptual models for how the network might function. First is described how the activity bump is forwarded from the EB to the PB, where it is duplicated and reformatted into a sinusoidal profile. In subsequent sections it is considered how the fan-shaped body (FB) network may recombine these bumps 76 to perform vector computations in support of goal-directed behavior. 77 78 The EPG population divides the EB into 16 'wedges', suggesting that the fly's head direction system samples angular space at 22.5° intervals. Importantly, this does not mean that the system cannot resolve head directions at resolutions higher than 22.5°, since the differential activation of columnar neurons with distinct directional tunings can effectively represent any arbitrary angle within the 360° around the fly. From the EB, EPG neurons convey the hΔbump to both the left and right PB, generating two bumps that are sampled at approximately 45° intervals. Due to the EPG projection pattern, there is a 22.5° shift in the directional tuning between EPG neurons in left and right PB, as recently confirmed by physiological recordings. Importantly, the bumps in the left and right PB still encode the same head direction, but do so using sets of neurons whose sampling of angular space is shifted by 22.5° (Hulse, 2021).

Within the PB, FB columnar neurons inherit a head direction bump directly from EPG neurons and indirectly through Δ7 neurons. The Δ7 populations appear ideally suited to reformat the EB bump into a sinusoidal profile, regardless of its original shape (Hulse, 2021).

Individual Δ7 neurons provide output to 2-3 PB glomeruli spaced ~180° apart (that is, separated by 7 glomeruli). Between these axonal compartments are dendritic segments whose EPG input weight is well fit by a sinusoid, suggesting that individual Δ7 neurons should have a sinusoidal tuning curve. Assuming the Δ7 population uniformly samples angular space (for example, with a 45° sampling interval), this would manifest as two sinusoidal bumps across the PB, one in the left PB and one in the right PB. Furthermore, recurrent Δ7 connections may enforce a sinusoidal activity pattern on the Δ7 population itself, which could improve the ability of Δ7 neurons to reformat the activity bump into a sinusoidal profile before passing it on to PB-FB types. The Δ7 population provides input to ~10 types of PB-FB neurons, effectively duplicating the activity bump in the process. As discussed below, this duplication may allow the FB network to recombine bumps in a way that implements a compact vector calculator (Hulse, 2021).

Why might the Δ7 population reformat the activity bump into a sinusoidal shape? Perhaps because sinusoids are a particularly suitable representation for vector-based computations, since the sum of any two sinusoidal of equal frequency is also a sinusoid. One way of schematizing this process is to use phasor diagrams. Viewed in this way, the sinusoidal activity bumps become vectors whose magnitude reflects bump amplitude and whose angular position indicates bump phase, with each phase mapping to an allocentric direction. Adding vectors is equivalent to adding sinusoidal activity profiles (Hulse, 2021).

Path integration is a canonical vector-based navigation strategy used by a diverse array of both flying and walking animals, potentially including Drosophila. In its most basic form, 2D path integration requires that an animal keep track of its direction and distance relative to a stored goal location, such as a food source or nest, often without the use of external landmarks. The direction and distance to the goal location is thought to be computed through the integration of self-motion signals and stored as a 'home vector'. To return to the goal location, animals are thought to generate appropriate motor commands by comparing their current heading to the stored home vector. While many insects are thought to generate and use visual snapshots of their surroundings to guide return trips, and while such visual homing may involve the MB, this study focused on how a home vector might be constructed and read out in the CX using only a stable head direction signal, a situation that can arise in featureless landscapes or in darkness. Although there is as yet no definitive evidence that the CX is used for path integration, in the next few subsections, it is shown how a network built from FB-inspired circuit motifs could compute a translational velocity vector in an allocentric reference frame whose integration would yield a home vector. It is noted that the framework for vector computations that is describe below is likely to be useful for a much broader array of behaviors involving oriented action selection (Hulse, 2021).

The potential for vector computations in the FB PFN neurons serve as the major columnar input to the FB network. The +/- 45° phase shift that is characteristic of all PFN neuron types implies that activity bumps from the left and right PB would end up ~90° apart in the FB. The amplitude of these activity bumps is likely to be strongly influenced by the different inputs that PFN neurons receive through their lateralized projections in the NO , setting up the possibility of bump-based vector computations in the FB. The PEN_a neurons, which are conjunctively tuned to head direction and angular velocity, perform a similar computation in the EB by providing phase-shifted input to the EPG neurons, thereby updating the position of the EPG bump when the fly turns. Inside the FB, vector computations fed by phase-shifted PFN bumps whose amplitudes are controlled by different conjunctive signals could ultimately drive PFL neuron types to generate appropriate motor commands, an algorithmic idea for which an FB implementation was first proposed (Hulse, 2021).

Although the intra-FB columnar network is highly recurrent, much of it is built from a limited number of circuit motifs. These motifs serve as the backbone of a 2D grid in which activity bumps are constrained to either maintain their phase (that is, maintain their column) while moving across layers, or shift phase by 180° (that is, shift by half the width of the FB). While some pathways directly connect PFN neurons to output pathways, such as those involving PFL neurons, many more pathways run through this 2D grid. Thus, the network has depth, providing multiple layers with which to process activity bumps. In addition, a large number of tangential neuron types selectively innervate different layers of the FB, suggesting that the FB's vector computations are influenced by context and internal state. The sections that follow, draw from published experimental and theoretical work to explore the navigational implications of PB-FB neuron phase shifts. Importantly, for the purposes of discussion, it is assumed that the magnitude of PFN phase shift is precisely 90°, a simplifying assumption about symmetry in the circuit that ignores the type-to-type variability in estimated phase shifts across PFN types, the functional significance of which remains unknown. Similarly, all LNO types are assumed to be excitatory, but the proposed conceptual models could be built from inhibitory LNO types as well. Finally, while the columnar structure of the various dΔand hΔtypes show considerable variability, it is assumed that these neurons can either maintain the phase of an FB bump or shift it by 180° (Hulse, 2021).

Despite the PFN phase shifts, the two 90°-separated activity bumps arising from a single PFN type cannot propagate independently through the FB network, because nearly all single neurons and neuron types that are postsynaptic to PFN neurons sample from left and right populations equally. Instead, each postsynaptic FB type likely sums the bumps from the left and right PFN populations, indicating that each postsynaptic type represents a single, summed bump that will propagate through the FB network (Hulse, 2021).

Much like the PEN neurons, the PFN neurons innervating the left PB project to the right NO and neurons innervating the right PB project to the left NO, where they receive input from various LNO types. One potential function that this differential NO input to the left and right PFN populations could serve is to produce a new, transformed directional representation that could take on angles +/- 45° around the fly's instantaneous head direction. For example a strong excitatory input to the right nodulus would increase the bump amplitude of the left PFN population relative to the right PFN population. In turn, the summation of these two bumps by a postsynaptic neuron type in the FB would result in a new bump that lies closer to that of the left PFN population (Hulse, 2021).

Phasor diagrams of this process are given in the paper. Critically, because these vectors can only take on positive values (firing rates above 0), such differential input could only shift the resulting vector's phase by +/- 45° around the fly's instantaneous head direction. What might this transformed directional representation encode? The answer likely depends on the nature of the input that PFN neurons receive from LNO types. Recent work in Drosophila has shown that some PFN neurons show differential activity in the NO that reflects the fly's turning behavior during flight, but the nature of the rotational velocity signal remains to be determined. Two hypothetical scenarios are highlighted that differ in the specific information carried by PFN neurons (Hulse, 2021).

Directly wiring sensors to actuators in different ways can, in principle, allow a simple agent to display a variety of behaviors. But flies, like most animals, have to deal with an additional complication: some of their sense organs are on body parts that are different from those that enable them to move. PFN phase shifts could enable coordinate transformations, such as converting the allocentric head direction representation into an allocentric body direction representation. Flies make head movements that change their head-body angles by as much as 30° during both flight and walking. In this scenario, LNO neurons that provide input to PFN neurons arborizing in the left and right NO would encode how much the head is rotated to one or the other azimuthal direction of the body's axis, perhaps derived from proprioceptive information. When properly calibrated, such differential input could allow the PFN phase shift (which, at 45°, is sufficient to encode the entire range of head-body angles) to rotate the head direction vector by an angle equal to the head-body angle. This would effectively transform the fly's head direction vector into an allocentric body direction vector. To do so, the network could use gain fields, with an intermediate layer composed of PFN neurons whose head-direction tuning curves are gain-modulated by shared input related to the head-body angle. The neuron types downstream of PFN neurons would complete the coordinate transformation through their structured sampling of PFN neurons with distinct directional tunings. Coordinate transformations such as these may be useful when combining 2 allocentric directional representations with body-centered velocity estimates to estimate direction and distance. This idea is returned to in more detail in subsequent sections on vector computations related to path integration (Hulse, 2021).

A second scenario is presented: if it is assumed that an LNO type carries a motor efference copy of the fly's rotational velocity, as has been shown to exist in the fly visual system, then bump shifts driven by differential input to the left and right PFN populations could function as a forward model that encodes a prediction of the fly's future head direction or body direction. Why might a forward model of head or body direction be useful? Intracellular recordings from neurons in the ring attractor network have revealed that PEN activity is tuned to the fly's rotational velocity, but that this activity lags behavior by ~100 ms. Similar lags may result from propagation delays in neural processing, either along sensory pathways into the EB, or in passing the bump from EB to FB. In situations where flies might rely on the CX to direct their movements - especially in time-critical scenarios - such delays in updating the compass could be costly. One way to overcome this is for the PFN network to compute the fly's approximate future head or body direction so that navigational decisions can effectively be made in real-time, a strategy that dragonflies have been shown to use during rapid prey capture maneuvers. A forward model could also allow the fly to distinguish changes in its head or body direction associated with voluntary movements from those induced by external disturbances, such as changes in wind direction. More generally, matching the predicted head and body direction with the actual direction could enable the fly to fine-tune its movements to produce intended motor outputs. As described below, PFL neuron types have anatomical phase shifts that appear well suited to perform such comparisons (Hulse, 2021).

The PFN neuron types that the computations hypothesized above might involve is as yet unclear. However, the analysis of PFN inputs from LNO types does allow drawing some inferences about the sort of self-motion information that different PFN neuron types might carry. For example, considering that PFL neurons provide selective feedback to the LNO3 neuron type and considering that the PFL3 neurons feed the LCNOp neurons, it is hypothesized that the former may provide its downstream PFN neurons - the PFNv neurons - with efference information related to translational (and potentially forward) movement, and that the latter may provide its PFN targets - several PFNp sub-classes - with efference information related to rotational movements. In addition, many LNO types are downstream of pathways from vPNs, consistent with the use of optic 5flow-based self-motion signals. Furthermore, given that LNO types are the target of multiple input pathways, these neurons could carry combinations of sensory and motor signals to encode self-motion. Physiological recordings in behaving flies would be needed to test such hypotheses (Hulse, 2021).

Phase shifts of a single PFN neuron type could enable the generation of vectors that are within ±45° of the fly's instantaneous head direction. Intriguingly, the FB network appears to be wired to expand this range to allow for computations with vectors of arbitrary angles. he dΔ and hΔ neuron types could, in principle, allow for vector computations across arbitrary azimuthal angles. In this example, this study considered two hypothetical PFN populations (PFN1 and PFN2), both with 45° contralateral phase shifts. On their own, these PFN populations are limited to directional representations spanning 45° around the head direction signal that they inherit. However, an excitatory hΔ (or an inhibitory vΔ) would invert the PFN2 vectors, shifting them by 180°. Thus, if a postsynaptic neuron type were to sum the input from the non-inverted PFN1 neuron population and an inverted PFN2 neuron population, it could form a representation of any arbitrary vector over the full 360° range, even though the PFN1 and PFN2 populations are individually range-limited. A similar inversion could happen at the level of the PB, if one PFN population were to receive excitatory Δ7 input while the other received inhibitory Δ7 input, which is likely how the 180° separation of PEN_a and PEN_b population bumps is generated and maintained in the PB. Together, the non-inverted PFN1 and inverted PFN2 neuron populations form a basis set of four basis vectors, all separated by 90°. As mentioned previously, the requirement for PFN neurons to have a positive firing rate prevents any single PFN population from forming a basis set on its own; instead, forming a basis set requires four independent bumps located at 90° intervals. When this situation is achieved, independent NO input could alter the relative amplitudes of bumps carried by each of the four PFN populations (inverted and non-inverted PFN1 and PFN2 populations), enabling their sum to encode a vector with any angle and could thus represent such independent vectors. Importantly, during navigation, the orientation of this set of four vectors would be dynamically updated with the head direction representation, such that any computations derived from these vectors would be independent of the fly's current head direction (Hulse, 2021).

Could the intra-FB network support the construction of arbitrary vectors, which requires two layers beyond PFN input? A type-to-type network graph suggests that there are many pathways within the FB's 2D grid that could potentially implement a four-vector basis set. Two arbitrary PFN types, PFNd and PFNp_c were chosen, and their connectivity with two downstream dΔand hΔneuron types -vDK and hDA- to illustrate how this might work. It was also shown how a downstream neuron type -PFL3 in this case- could sum the input from the dΔ and hΔ to represent arbitrary vectors determined by independent NO inputs to the left and right PFNd and PFNp_c populations. Physiological investigations will be required to establish which of the FB's many pathways implement such computations, and whether or not the large number of these pathways is an indication of vector computations in different behavioral contexts (Hulse, 2021).

Having established that the FB network of Drosophila could, in principle, compute arbitrary vectors, the potential utility of PB-FB phase shifts and intra-FB connectivity motifs for path integration was explored. A variety of models have been proposed for path integration. These models have several differences, including whether the home vector is stored in an allocentric reference frame or an egocentric reference frame, and whether it is stored using a 'static vectorial basis' or a 'dynamic vectorial basis'. This study focused on models that store the home vector in an allocentric reference frame using a static vectorial basis, which has been shown to have several theoretical advantages and whose implementation is directly suggested by the FB's network architecture. Path integration models can be further divided into two groups according to whether the home vector is stored and read out as independent components or as a single vector (Hulse, 2021).

An example of the first type of path integration model, which stores the home vector as two independent components, was recently put forward in a previous study of bees. This work combined anatomical and functional data from bees, including physiological recordings of optic flow-sensitive LNO neurons and EM data, to build an anatomically inspired model of path integration based on the projection and innervation patterns of CX neurons, but without access to their synaptic connectivity. The model utilized PB-FB phase shifts to read out a home vector and, importantly, could also account for holonomic motion, which occurs when animals move in directions that are not aligned with their head/body axis, an issue returned to below. At its core, this model and those derived from it function by modulating the amplitude of left and right PFN bumps according to the insect's motion in the leftward or rightward direction, respectively. Integration of the left and right PFN activities can then store a home vector as two independent components. During readout, a population of PFL neurons is assumed to compare the insect's current head direction to that of directions 45° to the left and to the right to decide which direction is closer to the implicitly stored home vector. This 45° 'functional offset' (phase-shift) was derived from physiological recordings demonstrating that some LNO neurons function as optic flow sensors whose optimal expansion points are 45° to the left and right of the bee, a feature returned to below. While conceptually elegant, one major feature of this model is inconsistent with the anatomy and connectivity of the homologous neurons in the Drosophila connectome. In particular, the model requires that the left and right PFN bumps independently propagate to right and left PFL populations, respectively. This operation is unlikely to be supported by the FB columnar network, since every neuron and neuron type postsynaptic to PFN neurons receives input from both the left and right populations (Hulse, 2021).

In the next two sections, use the additional anatomical and connectivity information provided by the CX connectome to propose two conceptual models for computing an allocentric translational velocity (TV) vector whose integration could be stored as a single home vector. The first model builds on the work using bees and uses PFN offsets to simplify home vector computation. The second model is more relevant to walking insects and incorporates a head-to-body coordinate transformation to compute the fly's translational velocity vector. In both cases, the key computation performed by the FB network is a coordinate transformation that ensures that egocentric velocity signals and allocentric directional representations are directionally aligned (Hulse, 2021).

Flying insects are thought to perform visual odometry by relying on optic flow sensors to estimate their velocity relative to the ground, consistent with leg-based cues being of little use and motor signals being unreliable in the face of external perturbations, like gusts of wind. In addition, during flight many insects make banked turns involving body rolls that are accompanied by gaze-stabilizing head rotations that keep the head near the horizontal plane. Importantly, flight trajectories often contain a significant sideslip component as well, during which the insect's translational velocity is in a direction that is different from that of its head-body axis (Hulse, 2021).

The FB's recurrent circuitry described above could use self-motion information to compute a flying insect's allocentric translational velocity (TV) vector. One potential model is composed of two PFN neuron types that receive independent input from two hypothetical LNO neuron types, LN1 and LN2, for which there are multiple candidates. It exploits the FB network's ability to form a set of four basis vectors to compute a single TV vector. To do so, it employs the optic flow sensors described for bees - with their preferred expansion points spaced at 45° intervals around the fly's head - to modulate the amplitudes of the four basis vectors such that their sum encodes an instantaneous allocentric TV vector. Importantly, this model relies on the fact that the basis vectors and optic flow sensors are directionally aligned. That is, at every moment in time, each bump in the basis set has its amplitude modulated by a velocity input that senses movement in the same direction as encoded by the bump. Much like the model for bees this model can account for holonomic motion (that is, an animal's movements in directions not limited to its heading and head direction). Another feature of this model is that it should be insensitive to head movements in the yaw plane, since the optic flow sensors and FB basis vectors are both in head-centered coordinates. A recent study found that, similar to the optic flow sensors described above, some PFN neuron types and their LNO inputs are preferentially tuned to air flow oriented ~45° to left or right of the fly's head, providing for a second potential velocity estimate whose tuning is aligned to PFN basis vector. The next conceptual model explores how this might work in walking insects, when the velocity sensors may be in a body-centered reference frame while the directional representation is in a head centered allocentric reference frame (Hulse, 2021).

Could the FB network compute an instantaneous TV vector in cases where its velocity and directional estimates are in different reference frames? A model explores such a scenario using a hypothetical example of a walking fly whose velocity estimates are computed using cues that operate in an egocentric, body-centered reference frame. These velocity estimates could be derived from motor efference copies or proprioceptive cues, and the existence of estimates for both forward (parallel to the body axis) and sideslip (perpendicular to the body axis) velocity. Computing a TV vector in this scenario is more complicated than in the previous model because the direction of the head and that of the body are not necessarily aligned, which requires a head-to-body coordinate transformation. This model uses the head-body angle to compute the total TV vector as the sum of two components, which represent the distance traveled parallel and perpendicular to the fly's body axis. To compute the parallel and perpendicular components of the TV vector, the model uses two basis sets that receive NO input related to the head-body angle as well as either a forward or sideslip velocity signal (Hulse, 2021).

A circuit for computing the component of the TV vector parallel to the fly's body axis is shown that involves two calculations that occur in parallel. The circuit recruits two independent PFN populations, one to encode movement in the forward direction, and the other for movement in the backward direction. A velocity signal increases the amplitude of the two PFN vectors that point in the direction the fly is moving, resulting in a vector whose amplitude encodes velocity and whose direction is either the fly's head direction, or its reverse head direction, which would be captured by the PFN2 population. At the same time, an input related to the head-body angle transforms the head-centered vector into a body-centered vector. The result is a single vector encoding the component of the fly's movement parallel to the body's axis in either the forward or backward direction. As mentioned above, this sort of computation could employ gain fields, but with the transformed representation (that is, body direction) being scaled by the fly's velocity in the process (Hulse, 2021).

The component of the TV vector that is perpendicular to the fly's body axis could be computed using the same circuitry as above, but with right/left sideslip velocity signals instead of forward/reverse velocity signals. Such a circuit would work regardless of whether the fly is sideslipping right or left or whether its head is to the right or left of the body axis. The output of these two circuits could then be summed tocompute a single vector that encodes the fly's instantaneous translational velocity in an allocentric reference frame (Hulse, 2021).

The conceptual models described above -one for flight and the other for walking- could, in principle, compute an allocentric translational velocity vector whose integration would yield an exact home vector. To accomplish this, the models use coordinate transformations to ensure that allocentric vectors are directionally aligned with the egocentric velocity estimates that control their amplitudes. While these particular models highlight the general utility of such transformations, the FB circuitry could, in principle, accommodate many similar models. In addition, it is possible that animals structure their movements during outbound paths to simplify the computation of the translational velocity vector. For example, if an animal were to only move forward during outbound paths, then circuit components dedicated to encoding backward motions would not be needed by the path integration circuit, a feature explicitly used by the model for bees. Similarly, it is possible that an exact solution is not always required to perform path integration. For example, if a model generates errors that tend to cancel out during the integration process, the home vector can still effectively guide behavior. The local search behavior of foraging Drosophila, for example, involves relatively short loops that may not require a precise accounting of the goal location. In most situations, the fly should also be able to use local sensory cues in addition to path integration during such search behaviors. Finally, it is possible that egocentric velocity signals could come pre-aligned to FB bumps, assuming that the LAL could implement the trigonometric functions needed to scale velocity signals according to, for example, head-body angle. Taken together, these models highlight the connectome's ability to inspire novel, implementation-level hypotheses about network computation. They also provide a framework for generating many similar models, with specific implementations that likely depend on cell type, species, and behavioral need. Ultimately, evaluating models like these necessarily requires physiological recordings from animals in specific behavioral contexts. Indeed, two contemporaneous studies have discovered direct physiological evidence that FB circuits compute the fly's translational velocity, and have independently proposed theoretical models that are conceptually similar to those described above. Yet, it is currently unclear if the output of this computation encodes the fly's translational velocity vector or just the phase of this vector. Similarly, how the type-to-type variability in PFN phase shift magnitude affects these computations requires future study. Finally, it is also possible that tangential neurons carrying feedback or self-motion signals could scale the magnitude of these vectors within the FB network itself (Hulse, 2021).

Theoretical work has suggested several potential ways to integrate translational velocity vectors. First, translational velocity could be integrated and stored using two separate circuits: a ring attractor that encodes the angle of the home vector and a line attractor that encodes the length of the home vector. Rather than keeping track of the distance traveled in each allocentric direction, this network would shift the columnar location of an activity bump to encode the phase of the home vector. Solutions like these seem unlikely to be implemented by the FB, since they require FB-centered attractors with circuit motifs for shifting the bump, similar to those found in the EB-PB attractor, which no evidence is seen for in the FB network. Second, the model for bees employed 18 neurons per FB column and used structured recurrent connections between them to integrate and store a two-component home vector. The hemibrain connectome provides little evidence for such structured recurrent PFN connections, especially in the NO, where some PFN types show all-to-all connectivity. Finally, the conceptual models described above allow for the computation of a single TV vector, suggesting that the FB network could simply integrate the corresponding activity bump and store the resulting home vector directly. In doing so, this integration processes would function by keeping track of the distance traveled in each allocentric direction. How might the network integrate the TV vector and store the resulting home vector (Hulse, 2021)?

Integration and storage could occur through several complementary mechanisms operating at different scales, from changes in synaptic strength or the excitability of individual neurons to persistent activity at the single neuron or network level. In addition, integration and storage mechanisms may vary across species and environmental context depending on the animal's needs. For example, desert ants can maintain multi-day memories of multiple goal vectors to food sites and remember home vectors over 1 to 2 days. This sort of long-term maintenance would favor stable storage mechanisms, such as changes in synaptic strength. In contrast, Drosophila performing relatively brief local searches close to a food source may rely on short-term mechanisms that could involve persistent neural activity. The connectome alone does little to constrain the space of possible storage mechanisms, but it can provide information regarding the likely site of storage and inspire a few conceptual models for how the vector could be stored (Hulse, 2021).

Several considerations narrow the potential site of home vector storage in Drosophila. In the framework of the above conceptual models, the home vector would be stored downstream of the four vector basis sets used to compute the TV vector. The PFL2 and PFL3 neuron types are well positioned to read out the home vector by comparing it to the fly's instantaneous head direction, suggesting that the home vector is perhaps stored by neuron types that provide inputs to the PFL neurons. The PFL neuron types could also store the home vector themselves. Some insects are likely to maintain more than one goal vector, but the PFL neurons could store these different goal vectors through input-synapse-specific presynaptic or postsynaptic plasticity. In addition to direct PFN input, PFL2/3 neurons receive shared input from a handful of hΔtypes, several FC2 types, one or two dΔtypes, and many FB tangential neuron types, each of which could also potentially store a home vector (Hulse, 2021).

Several potential storage mechanisms seem plausible. Many hΔneuron types have within-type recurrent connections, forming small loops that connect pairs of hΔneurons that encode directions 180° apart. If the biophysical properties of these neurons allowed for graded, persistent activity, and hΔneurons have inhibitory connections, each column-pair could encode the direction traveled along one dimension. Alternatively, while the FC neurons providing input to PFL2/3 neurons largely lack within-type recurrent connections, they could maintain a vector in working memory through graded changes in their excitability or activity. Finally, the FB's tangential neurons could potentially store home vectors through column-specific plasticity, as is known to occur between ring neurons and EPG neurons. In general, some recurrent architectures may allow for the storage of home vectors, but an FB ring attractor, if it were to exist, would likely not allow for home vector storage, since these networks have the undesirable property of forming a single activity peak at the expense of the activity in distance columns that would be needed to fully encode the home vector (Hulse, 2021).

Overall, although the connectome can do no more than rule out some circuit implementations of how the home vector might be stored, it should prove useful in prioritizing a search for the likely neural targets for such a function. It is important to note that the entire circuitry described above must function in different modes depending on the animal's behavioral needs - integrating direction and distance traveled to update the home vector when the fly is searching, but switching to reading out the home vector when the fly is attempting to return to a previously visited spot. The likeliest candidates for such behavioral mode switching are the FB's tangential neurons (Hulse, 2021).

Once formed, how might an insect 'read out' the home vector to return to its goal location? In the current formulation, the home vector points from the nest to the insect's current location. Returning home, then, requires that an insect move in a direction opposite to the home vector. To accommodate the other behaviors and computations that these circuits are likely to be involved in, the home vector is referred to as the 'stored vector', which is read out to orient the insect along a 'goal vector'. However, unlike an ant or bee, the fly is not a central place forager. Thus, 'goal' in this context refers only to a spot that the fly is likely to return to during a local search, such as a food source. PFL neurons are generally regarded as the major columnar output of the FB network. Their PB-FB offsets strongly implicate them in reading out stored vectors in ways first proposed by theoretical work and then, at the implementation level, for bees. In particular, PFL neurons may use their PB-FB phase shifts to compare the fly's instantaneous head direction, which they receive in the PB, to that of the stored vector, which they may receive in the FB, to generate appropriate motor commands to guide the fly to its goal. In doing so, they effectively generate egocentric motorcommands based on allocentric directional variables. Interestingly, each of the three PFL types have characteristic phase shifts that strongly predict their involvement in generating distinct motor commands (Hulse, 2021).

PFL2 neurons may use their 180° phase shift and bilateral LAL projections to increase the fly's forward velocity when its heading is directly away from the stored vector, which in the formulation used in this study is towards the goal location. Unlike the other PFL types, PFL2 neurons receive only a single bump as input in the PB. This suggests that the population cannot make left versus right activity comparisons. In agreement with this, individual PFL2 neurons make bilateral projections to the left and right LAL. Because of their 180° phase shifts, the PFL2 population activity will be largest when the fly is heading directly towards its goal location. The above characteristic suggests that PFL2 neurons are ideally suited to generate a motor command related to forward velocity (Hulse, 2021).

PFL3 neurons may use their 90° phase shifts and lateralized LAL projections to orient the fly towards the goal. Their 90° offset predicts that the left and right PFL3 populations will have their maximum activity when the fly is 90° to the right or left of the goal direction, respectively. If the left PFL3 population generates left turns and the right PFL3 populations generated right turns, then the orienting behavior of the fly will have two equilibrium points: a stable equilibrium that occurs when the fly is oriented towards the goal direction and an unstable equilibrium when the fly is oriented in the opposite direction. This sort of read out would ensure that flies orient directly towards the goal location. It is additionally possible that across-column inhomogeneities in the EPG->PFL synaptic profile and in the PFL->LAL network may provide the fly with a 'default goal' in the absence of any FB input, similar to a hypothesis recently advanced in an independent study. The 45° offset of PFL1 neurons may serve a related function, although they target distinct downstream neurons compared to PFL2/3. One possibility is that the PFL2/3 neurons affect body orientation while the PFL1 population controls a separate variable, such as sideslip or head-body angle. Ultimately, it is also important to remember that brain regions like the LAL and CRE house complex recurrent networks with inter-hemispheric pathways that are likely to be inhibitory. These networks are likely to play a major role in the transformation of PFL population activity into motor commands for the fly, something that these hypotheses do not incorporate (Hulse, 2021).

Summary: vector computations in the FB The discussion above supports the notion that the FB network has the computational capacity to compute, store, and read out vectors in support of goal-directed navigational behaviors. While this study has focused on path integration as a canonical vector-based computation, Drosophila are known to perform several other behaviors that may rely on the formation of 97 goal vectors, including: local search, a path-integration-based foraging strategy; menotaxis, where a constant 99 heading is maintained relative to an arbitrary goal direction to generate straight trajectories that support long-distance dispersal; place learning, which requires associating visual cues with the presence of a cool spot in an otherwise hot 2D environment; and the detour paradigm, where flies orient towards directions associated with attractive landmarks even after they have disappeared (Neuser et al., 2008). In addition, ethologically-based studies in behaving insects have established a range of vector-based4 behaviors, from long distance migrations that require a time-compensated sun compass to the waggle dance that bees use to communicate the distance and direction of a food source. The ability of some insects to store multiple goal vectors and the fact that different insect species may use vector computations to support distinct behaviors has important implications for FB circuits. The FB may have evolved as a general vector calculator that can be co-opted, whether by evolution or in support of distinct behaviors, to support vector-based navigation strategies generally. In support of this idea, FB circuits, neuron types, and motifs are highly conserved across insects. Additionally, the ability of some insects to store multiple goal vectors requires mechanisms for switching between them, a function perhaps mediated by the large class of FB tangential neurons that could convey context and state information to the columnar networks involved in vector operations beyond navigation: the CX as a multifunctional network for context-based action selection. While this study has focused much of the discussion on column-specific computations supporting vector navigation, the CX also receives input from over 150 distinct tangential neuron types. In the sections below, these neurons' role in sensorimotor processing, memory-guided decision making, circadian rhythms, sleep-wake control, and nutrient homeostasis is briefly highlight. Together, these findings suggest that the CX operates as a multifunctional network supporting state- and context-dependent action selection for high-level behavioral control (Hulse, 2021).

Sensorimotor processing Consistent with the CX's involvement in navigation, several studies have implicated FB tangential neurons in sensorimotor processing. For example, ExFl1 neurons, which are likely FB2B_a and/or FB2B_b neurons, are strongly modulated by whether or not the fly is flying and are tuned to progressive optic flow, providing a potential indication of the fly's current sensory and motor state. Similar activity patterns may be expressed by several other FB types as well. In addition, a recent study focused on the LH identified an FB tangential neuron type called PV5k1 (FB2H_a, FB2H_b, and/or FB2I_b) whose activation during closed-loop visual conditions leads to a reduction in the fly's wingbeat frequency. Sensorimotor signals like these are well positioned to influence CX-driven motor commands based on the fly's immediate sensory environment and ongoing motor state (Hulse, 2021).

Memory-guided decision making Flexible behavior also requires animals to respond to their immediate sensory surroundings by evaluating past associations regarding the valence and novelty of available sensory cues. To investigate this, focus was placed on tracing pathways between the MB -the fly's main learning and memory center- and the CX. In agreement with results from a companion manuscript focusing on the MB and trans-Tango-based circuit mapping, extensive pathways were found leading from MBONs to FB tangential neurons. In the context of navigation, the MB is considered a potential source of visual snapshot memory, which may allow insects to base their navigation decisions on remembered panoramic views. Consistent with this general notion, some FB tangential neuron types in FB layers 2 and 8 have been proposed to play a major role in visual learning. In addition, recent studies have implicated MB-to-CX pathways in behaviors other than navigation. For example, MB-to-CX circuits may be important for experience-dependent alcohol preference. In addition, MB-to-CX circuits are involved in consolidating courtship experience into long-term memory (Dag et al., 2019). The sheer number of connections between MBONs and FB tangential neurons suggest this prominent pathway is involved in many behaviors that make use of valence and novelty signals extracted from past associations that the fly has made with 61 its current sensory surroundings (Hulse, 2021).

Circadian influence on the CX Animals also select their actions based on latent environmental variables, such as the time of day, which are predictive of environmental conditions like temperature and humidity. Flies are most active around dawn and dusk, and show consolidated periods of inactivity throughout the night and during a daytime siesta (Dubowy and Sehgal, 2017). This daily rhythm is imposed by outputs from the circadian network and functions to restrict behavior to appropriate times of day. Previous studies have identified a population of anterior-projection DN1 clock neurons that convey circadian information through TuBu neurons to EB ring neurons. Thus, CX circuits are likely to receive circadian information that could be used to select behaviors according to time of day. Whether circadian pathways target other regions of the CX requires further investigation. In addition to receiving circadian inputs that could affect rest-activity rhythms, considerable evidence suggest CX circuits are involved in tracking internal states, such as sleep need and nutritive state, which is discussed next (Hulse, 2021).

Sleep-wake control: While its functions remain largely unknown, sleep is associated with a variety of processes in Drosophila, including synaptic homeostasis, memory formation and consolidation, changes in gene expression, and several metabolic processes. Sleep in flies is behaviorally defined as a reversible state of immobility that is homeostatically regulated and associated with an increased arousal threshold. It is marked by drastic changes in brain-wide activity patterns. The neural circuits involved in tracking sleep need and inducing sleep are thought to partially reside in the CX. In particular, a heterogeneous population of FB tangential neurons labeled by the R23E10 GAL4 line induces sleep when activated and tracks sleep need through changes in baseline firing rate and intrinsic excitability. Similarly, ER5 ring neurons track sleep need, and reciprocal connections between the EB and dFB are hypothesized to form a core circuit for homeostatic control of sleep. Counteracting these sleep-promoting neurons are wake-promoting dopaminergic neurons in the dorsal FB that are thought to promote wakefulness by inhibiting R23E10 neurons. Connectomic analysis revealed extensive reciprocal connections between putative sleep- and wake-promoting populations within the dFB, which could function as a 'flip-flop' switch to ensure that only one population is active at a time. In addition, a large number of previously undescribed pathways were identified leading to and from sleep-wake neuron types whose potential involvement in sleep-wake control requires future investigation, including reciprocal pathways connecting neurons in the EB with those in the dorsal FB (Hulse, 2021).

Several limitations of the hemibrain dataset are notable in the context of sleep: neurons that that show structural changes as a function of the fly's sleep-wake history such as ER5, could have sleep-state-dependent connections different from those described in this study; similarly, at present, the hemibrain connectome does not include reconstructed glia, which are also known to be involved in sleep-wake control; lastly, the hemibrain dataset cannot resolve the presence of gap junctions, which may also be important for sleep-wake control (Hulse, 2021).

Nutrient homeostasis: Recent studies have suggested that the CX is involved in internal state-based action selection beyond sleep-wake control. Within the EB, a population of ring neurons allows flies to assess the nutritive value of sugars, independent of their taste. Similarly, tangential neuron types in the dorsal FB have been implicated in feeding decisions based on the nutritive value of foods, and they may incorporate past experience into these computations. And vΔA_a columnar neurons, which innervate the AB and dFB, show oscillatory dynamics that depend on hemolymph glucose levels, and altering vΔA_a activity levels affects fructose preference. Together, these studies implicate CX circuits in nutrient homeostasis, a process important for successful foraging based on the fly's metabolic needs (Hulse, 2021).

Circuit motifs for high-level behavioral control and action selection: The need for high-level behavior selection may explain the potential interactions of circuits related to navigation, feeding, circadian rhythms, and sleep. Hungry flies, for example, are known to forgo sleep in favor of foraging. Similarly, both sleep and feeding are known to be under circadian control, biasing their occurrence to appropriate times of day. Based on these considerations and the experimental evidence summarized above, it seems likely that the CX operates as a multifunctional network that can be dynamically reconfigured (Marder, 2012) to support a variety of goal-directed behaviors based on immediate sensorimotor variables, learned associations, time of day, sleep need, nutritive state, and other as-yet-unknown inputs. Such a view of the CX is consistent with the variety of neuromodulator and peptides released by FB neurons (Hulse, 2021).

This connectomic analysis identified circuit elements and motifs that may support appropriate action selection. Most notably, many tangential neuron types, including EB ring neurons, form dense recurrent connections, both within neurons of a type and across distinct neuron types. For example, the FB's tangential neurons in Layer 6 that have been implicated in sleep-wake control are highly recurrently connected. It is possible that some of these neurons or other neurons in their layer are involved in decision-making related to feeding. If so, inhibitory interactions between these different tangential neurons may -akin to the interactions of ring neurons for sensory control of the fly's compass- enable the fly to select appropriate actions based on internal need. Related to this, recent studies have reported oscillatory activity in ER5 ring neurons related to sleep-wake control, but how the highly recurrent networks in the EB and FB might support such oscillations remains to be determined. One possibility is that all-to-all inhibition between ring neurons in the EB could, with the appropriate inhibitory conductances, induce such patterns of activity. A different issue raised by the highly recurrent architecture of sleep-wake networks concerns how activity may propagate in these networks. Artificial stimulation of neurons within such potentially self-regulating networks may trigger downstream activity that is never seen in more naturalistic situations, confounding the interpretation of experimental results. Testing such ideas will require a finer-resolution analysis of the role that these neurons play in the action selection process (Hulse, 2021).

How might the CX's columnar architecture support these distinct behaviors? Links between the CX's role in sleep and navigation have begun to be explored both experimentally and computationally, but the CX connectome suggests that the number of pathways and neuron types that connect circuit elements known to be involved in these functions may have been underestimated. For example, the dFB tangential neurons involved in sleep-wake control contact many columnar neuron types. Although it is believed that this columnar structure -and the FB's 2D grid more generally- is convenient for vector computations, why this columnar structure may be needed for sleep-wake control or for feeding- and satiety-related computations remains mysterious. One possibility is that head direction or traveling direction signals may be used as proxies for tracking the quality or quantity of the fly's waking experience, perhaps to estimate sleep and/or nutritional need. Alternatively, the FB's navigational signals may be inherently activity-promoting since they likely drive premotor neurons in the FB. If so, these navigational signals may require suppression to establish a sleep state or to enable a hungry fly to stop on a patch of nutritive food. Another possibility is that tangential neurons may gate incoming sensory information, which could promote sleep or perhaps encourage a hungry fly to continue a feeding bout by ignoring distractors. Ultimately, if the columnar neurons are the main output of the CX, as seems likely, the FB's tangential neurons must impact behavior through them (Hulse, 2021).

Considering that the highest layers of the FB are associated with modulating the fly's activity depending on sleep state and satiety levels, the connectivity pattern within the FB suggests that information about the fly's current navigational state may enter the FB ventrally, that additional processing may happen in the middle layers, which receive considerable input from the MB, and that this processing may then determine the fly's next actions (or inaction) in the dorsal layers. An additional possibility suggested by the flow of bump information from ventral to dorsal layers of the FB, and by the diffusion of columns in the dorsal layers, is that the specificity of actions is organized along the vertical axis of the FB, with oriented actions modulated and signaled by output neurons originating in the middle layers and the fly's overall state of activity modulated in directionally non-specific ways by the highest layers (Hulse, 2021).

Compared to the vector computation models suggested by the CX's columnar structure, deriving connectome-inspired insights into the function of the CX's action selection networks proved more challenging. One reason for this is that most FB tangential neurons receive input from CX-associated regions whose function remains poorly understood, like the SMP/SIP/SLP, making it hard to assign specific circuit functions to these neurons based on their inputs alone. In contrast, the vector computation models relied on a considerable amount of prior experimental data that, when mapped to the connectome, provided physiological hooks for generating novel hypotheses regarding circuit function. In addition, FB tangential types often have extensive reciprocal connections to other tangential types, which, given the absence of functional data, is hard to interpret. Once some of these functions are better understood, it may be possible to derive internal state hierarchies, like those identified for directional sensory cues carried by EB ring neurons, which could suggest how the CX prioritizes different internal states. However, many of these internal states involve variables that evolve over time, such as nutritive state or sleep need, suggesting the underlying CX networks may undergo considerable plasticity that may not be apparent in connectome-level connectivity. The dynamic interaction of different internal state variables is likely also governed by neuropeptidergic signals that bathe the CX, but that this analysis did not capture. Finally, given the limited understanding of the variety of behavior the CX may support, understanding how internal state cues may factor into these behaviors is hard to predict at present. To better understand which behaviors the CX may be involved in, this study used the connectome to identify output pathways, a topic dealt with next (Hulse, 2021).

Directing and modulating movement based on the fly's current state: The outputs of the CX likely modulate the fly's actions in a variety of different behavioral contexts, including voluntary take-offs, negotiating uncertain terrain, feeding, oviposition and fighting. The structure of the FB, in particular, suggests that it could modify the head direction signal to orient the fly with respect to behaviorally specific 'goal' directions. Such goals could be a source of food or safety or, for female flies, a good site for oviposition (Hulse, 2021).

The FB's columnar output types (PFL, PFR, FR, FC and FS neurons) feed relatively independent output subnetworks, which may support, through unknown mechanisms, the maintenance of independent goal locations associated with different behaviors. Alternatively, these subnetworks could control independent sets of behaviors. If true, each subnetwork may carry the potential for the execution of actions towards independent goal locations, each specific for a given behavior and carried by a specific FB columnar type (or set of types). For example, some subnetworks could control behaviors related to goals in front of the animal, such as feeding or gap crossing. Some CX outputs contact a limited number of MBON-associated networks. These connections may allow the CX to modulate some behavioral responses to specific sensory contexts that have been associated with negative or positive valence through the MB. The fact that an oviposition neuron (oviIN) is associated with these MBON networks could mean that CX networks influence spatial decision-making during oviposition, which is known to be informed by several external factors. In contrast, how and why CX signals from the columnar FR1 neurons should directly influence MB neurons themselves (in the case of the FR1 neurons, the MBON30 neurons) is less clear. The variety of different interactions between the MB and CX suggest that investigations of memory-guided orientation and navigation may benefit from a study of both regions acting in concert. Consistent with such an idea, the atypical MBON, MBON27, targets the DNa03 neuron type, which is also targeted by PFL3 40 neurons (Hulse, 2021).

Another axis along which CX-mediated behaviors can be subdivided is the scale of orientation control. From the body to the head and legs, proboscis, abdomen or antenna, all body parts have an orientation relative to the environment. Each of these body parts could benefit from coordinated but independent control and could be individually targeted by CX outputs. The CX could, in the context oviposition, direct abdomen bending for egg laying in a manner that incorporates the fly's internal sense of its body size, posture and orientation relative to its surroundings. Hints for how the CX exerts such directional control may be found in the morphology of its outputs (Hulse, 2021).

Output neurons with bilateral innervation patterns in premotor regions such as the LAL and CRE are likely to modulate symmetric actions (for example, forward walking), while those with unilateral innervations in such regions likely control asymmetric actions (for example, turning). Examples of the former include the PFL2 and FS1-3 neurons, while PFL1, PFL3, PFR, FR, FS4 and 55 FC neurons all show unilateral innervations of premotor regions (Hulse, 2021).

These different output signals could also vary in how directly they control flies' behavior. CX outputs could themselves direct the animal's movements and/or orientation towards a desired location or away from one associated with danger. The PFL2 and PFL3 neurons provide the most direct link from the CX to the motor center, known as the ventral nerve cord (VNC). As the major output channel of the CX, they are prime candidates to guide orientation and/or movements to a CX-specified goal. However, these actions would need to be coordinated with movements of body parts that alter the sensed orientation, most notably head movement. The ExR8 neuron is a candidate to carry out some of those corrections, through connections both to DNs and to the visual system. The remainder of the CX's outputs act more indirectly, and may modulate and gate actions controlled by other brain regions rather than directly controlling them. This is well exemplified by the multiple points of convergence between visual pathways and CX output pathways. Some CX output neuron types (PFL1, ExR7 and most of the FC neurons) and much of their downstream circuitry are completely uncharacterized. THese underexplored brain regions, a more complete connectome, and genetically targeted imaging and perturbation experiments will help to identify the function of these pathways (Hulse, 2021).

Navigation with small networks and with numerical variation in columnar neurons: The remarkable behavioral repertoire of insects is still more remarkable when considering their small brains. The CX connectome suggests that part of the secret behind this wide-ranging repertoire lies in having evolved architectures that are precisely configured for sophisticated behavior, but -physiological and behavioral genetics studies suggest- with weights that are plastic to allow these behaviors to flexibly adapt to context and situational demand. It is likely that the impressive computational power of their brains may also derive from an underexplored aspect of their neurons: their capacity for arbor-specific local computations, possibly even subthreshold computations in which synaptic release does not require spiking, and molecular computations through signal transduction cascades. These issues will require further experiments, but the connectivity observeD in the EB, for example, hints at a rich potential for insights into subcellular computation in the CX (Hulse, 2021).

Regardless of the true computational capacity of single neurons, it is remarkable that the fly can navigate with a head direction system of directionally tuned columns (in the EB) and just a few thousand neurons performing vector computation (in the FB). In principle, such small networks should be exquisitely sensitive to any variations in the number of neurons encoding each direction. However, the CX connectome revealed a striking difference in the number of columnar neurons that innervate each of the 18 PB glomeruli (although these differences are mirror-symmetric). Although several studies have investigated the developmental origins of columnar CX neurons, none have noted or focused on this systematic mirror symmetry. There are indications of numerical variations in some columnar neurons, such as the EPG, PEN and PFL neurons, in the FAFB volume as well, and more complete EM reconstructions of that volume should be able to clarify whether these variations exactly match what is seen in the hemibrain volume (Hulse, 2021).

The functional consequences of the systematic variation of neuron numbers across columns are entirely unknown. It is possible that this variation builds redundancy into a critical navigational system, or that the increased numbers of neurons in specific glomeruli ensure a preferred 'default' location for the bump to occupy within the more central columns of CX structures, and perhaps also a 'default' heading for the fly to adopt, an idea that is similar to a suggestion advanced independently. A different possibility is that such variation is not stereotyped across flies, but is specific to individuals, and that this may account for locomotor biases across the population. Functional experiments with specific perturbations of neuron numbers in different columns may be necessary to further investigate this issue. Regardless of their functional role, how such mirror symmetric numerical variation is achieved may be an intriguing question for future studies of CX development. It is not known if asymmetries and mirror symmetries in columnar neuron numbers are also present in the CX of other insects, but parallel efforts in connectomics should soon make this clear. Some Diptera, including Drosophila, have a closed EB, in contrast with most other insects, whose CBLs have an open, FB-like (CBU-like) structure. The non-uniform distribution of the EPG at the base of the EB, EPGt innervation at that location, and systematic modifications to neuron number across the columnar neuron types may represent an evolutionary adjustment to the closing of this structure (Hulse, 2021).

The CX as a tractable deep recurrent neural network: Technical advances over the past several decades have enabled increasingly large-scale recordings of neural activity from the central brains of a wide range of animals. These recordings have, in turn, enabled high-throughput studies of neural response properties that have focused on relating patterns of neuronal activity to sensory, behavioral and internal state variables. However, the biophysical and circuit mechanisms underlying these response properties have been more challenging to access. Similarly, dramatic progress in the field of machine learning has enabled the creation of sophisticated artificial agents that can solve a variety of different cognitive tasks, including flexible navigation. Some units in these deep networks develop response properties broadly similar to those observed in real brains (Hulse, 2021).

Insights into how such artificial neural networks generate the representations observed in their units - something that could, in principle, guide mechanistic hypotheses for the function of natural neural networks - have been slower to come for progress in uncovering the architectural basis of navigational responses in these networks). In this era of deep learning, a broader question concerns the level of understanding that is appropriate or even possible for the function of large and complex neural networks. What seems achievable is an understanding of learning rules and objective functions that can, in principle, generate networks with realistic population responses for specific cognitive tasks. The conservation of the CX's structure across arthropods perhaps highlights the extent to which the region has, in practice, been shaped by such rules over evolutionary timescales in the service of flexible behavior. But what of an understanding of the actual network implementation itself? Some have argued against the necessity or desirability of such a level of understanding. The fly's relatively brief history in systems neuroscience provides an increasingly compelling counterargument and may eventually offer a roadmap for implementation-level understanding that could scale to much larger brains and more complex cognitive functions (Hulse, 2021).

The fly displays a wide repertoire of flexible behaviors, and some of its recurrent neural circuits show dynamics that have been linked to associative learning and navigation across animals. Its 100,000-neuron brain circuits may appear complex, but they also feature modularity, type-specific connectivity and topography that is genetically pre-specified and has been refined over its evolutionary history. Some of these features apply to much larger brains as well, although there is likely greater flexibility in the wiring of mammalian circuits and greater heterogeneity within cell types in the mammalian brain. It is possible that developmentally-driven organizational features of natural brains may actually make them more tractable than artificial neural networks for an understanding of their function. The connectivity of a small fraction of the fly CX's many neuron types arranged by layers. Taking a single-neuron-resolution view of this subnetwork shows just how densely recurrent it is, even at a small scale. Indeed, if the types and connectivity of these neurons were unknown, extracting network structure from population responses would be a challenge. However, sorting the neurons into types -in this case, inhibitory types- makes the logic of the network clearer. Combining this circuit connectivity with physiological studies has enabled not only the generation of hypotheses for the computations that may be carried out by subnetworks at each layer, but, increasingly, tests of these hypotheses. As a result, it is possible to establish circuit-level mechanisms underlying the generation of different response 80 properties. Importantly, fly circuit connectivity is not always structured, many synaptic connections are plastic, and information from one part of the network often flows to all other parts of it. Nevertheless, the developmentally pre-specified organization of these networks makes them experimentally tractable. Although the computational capacity of the morphologically and biophysically complex neurons in these networks has likely been vastly underestimated, the connectome thus raises the prospects for at least a circuit-level understanding of how the fly's CX generates many of this small animal's flexible behaviors (Hulse, 2021).

Hardcastle, B. J., Omoto, J. J., Kandimalla, P., Nguyen, B. M., Keleş, M. F., Boyd, N. K., Hartenstein, V. and Frye, M. A. (2021). A visual pathway for skylight polarization processing in Drosophila. Elife 10. PubMed ID: 33755020

Many insects use patterns of polarized light in the sky to orient and navigate. This study functionally characterized neural circuitry in the fruit fly, Drosophila melanogaster, that conveys polarized light signals from the eye to the central complex, a brain region essential for the fly's sense of direction. Neurons tuned to the angle of polarization of ultraviolet light are found throughout the anterior visual pathway, connecting the optic lobes with the central complex via the anterior optic tubercle and bulb, in a homologous organization to the 'sky compass' pathways described in other insects. This study detailed how a consistent, map-like organization of neural tunings in the peripheral visual system is transformed into a reduced representation suited to flexible processing in the central brain. This study identifies computational motifs of the transformation, enabling mechanistic comparisons of multisensory integration and central processing for navigation in the brains of insects (Hardcastle, 2021).

A critical challenge of active locomotion is knowing the right way to go. Sensorimotor reflexes can influence momentary changes in direction to hold a course or to avoid looming threats, but goal-directed behaviors, such as returning to a previous location from unfamiliar surroundings, require additional information and processing. External sensory cues must be transformed into an internal representation of position and orientation within the environment, Yhis can also be modified by past experience. In Dipteran flies, as in other invertebrates, a collection of neuropils known as the central complex (CX) is believed to coordinate such behaviors and plays a role in spatial memory, object memory, and action selection, in addition to homeostatic processes including hunger and sleep (Hardcastle, 2021).

Recent studies in Drosophila have revealed that activity in a network of CX neurons encodes and maintains a representation of the animal's angular heading relative to its environment, with similarity to head-direction cells in vertebrates. This neural representation of heading can be updated by internal, proprioceptive estimates of self-motion during locomotion, and by external cues, such as moving visual patterns and directional airflow. In other insects, including locusts, crickets, bees, butterflies, and beetles, the functional organization of the CX has frequently been studied in the context of navigation via celestial cues, particularly polarized light. The nearly ever-present pattern of polarization in the sky, formed by scattering of light in the atmosphere, offers an indicator of orientation to organisms able to detect and interpret it, and may be more stable than terrestrial landmarks. In these non-Dipteran insects, a multimodal neural circuit transmits polarization signals from the eyes to the central complex. This circuit is known as the 'sky compass' pathway for its proposed role in processing skylight polarization patterns and information about the position of the sun to bestow an animal with a sense of direction. In Drosophila, the anterior visual pathway (AVP), which comprises neurons connecting the medulla, anterior optic tubercle, bulb, and ellipsoid body, has been postulated to represent the homologue of the sky compass pathway. Visual processing in the AVP appears to be segregated into three topographically organized, parallel streams, of which two have been shown to encode distinct small-field, unpolarized stimuli. The neurons involved in polarization processing in Drosophila have not been identified beyond peripheral circuits of the dorsal rim area, a specialized region of the eye for detecting skylight polarization (Hardcastle, 2021).

A detailed mapping of the relevant polarization-sensitive neurons would allow the exquisite genetic tools and connectomic studies available in Drosophila to be leveraged to understand the workings of the CX and its integration of multiple sensory modalities. Behavioral experiments have demonstrated that Drosophila orient relative to polarization patterns while walking and in tethered-flight. A comparative approach would therefore provide insight into the processing strategies employed across taxa as well as species-specific adaptations. Furthermore, it may be possible to reconcile the existing evidence of a common, fixed representation of polarization patterns in the CX of non-Dipteran insects with the emerging model of a flexible representation of both visual information and heading direction in the Drosophila CX. Alternatively, fundamental differences in the organization and processing of polarized light signals between species may reflect specialized navigational requirements (Hardcastle, 2021).

This study set out to test the hypothesis that the anterior visual pathway conveys polarized light signals from the eye to the central complex in Drosophila. Neurogenetic tracing techniques and in vivo calcium imaging were used to characterize the organization of the neurons at each stage and their coding and transformation of visual features. Parallel circuitry in the medulla conducts polarization signals from photoreceptors in the dorsal rim area to a stereotyped domain of the anterior optic tubercle. From there, a postsynaptic population of neurons projecting to the anterior bulb relays polarization signals to ring neurons of the ellipsoid body, and in turn, the 'compass neurons' of the central complex. The superior bulb multiplexes polarized and unpolarized light signals, while the inferior bulb does not appear to be involved in polarization processing. Finally, population responses in the central complex were examined, and hallmarks were found of a flexible encoding of a single angle of polarization which could be used to direct motor output for navigation behavior (Hardcastle, 2021).

This study demonstrated that each section of the Drosophila anterior visual pathway (AVP) contains polarization-tuned neurons. Together, they provide a circuit to convey polarized light signals from the specialized dorsal rim area of the eye to the compass neurons of the central complex, via the anterior optic tubercle and bulb. This study shows that the pathway also conveys information about unpolarized visual features. The encoding of multiple visual modalities, the similarities in the constituent neurons, and the organization of the neuropils which accommodate them, support the view that the AVP in Drosophila is homologous to the sky compass pathway described in locusts, bees, butterflies, and beetles, among other insects (Hardcastle, 2021).

The approach of this study to investigating the neural processing of polarization vision offered a number of advantages over traditional intracellular electrophysiology. Firstly, it allowed simultaneous recording from whole populations of neurons, which would otherwise be technically challenging. This study investigates the spatial organization of polarization responses in an individual animal. This may be key in understanding the central complex, where dynamic responses reflect circuit plasticity and depend on numerous factors, such as proprioceptive inputs, internal states and goal-direction. Next, targeted expression of calcium indicators allowed isolation of specific anatomical groups of neurons, such as specific TuBu or ring neuron populations, greatly increasing the repeatability of functional characterizations. Crucially, the identification of corresponding genetic drivers will enable silencing experiments, optogenetic stimulation, and multi-population recordings to probe circuit function in the future. Imaging of calcium indicators also facilitated the characterization of neurons whose axons are prohibitively thin for recording intracellularly. MeTu-like neurons, for example, have long been assumed to deliver polarization signals from the medulla to the anterior optic tubercle, and this study was able to confirm this by direct observation for the first time (Hardcastle, 2021).

Since each detector for polarized light in the dorsal rim area (DRA) essentially has a different field of view, the success of this approach depended on the ability to stimulate a sizable number of DRA ommatidia. Surprisingly, almost the full extent of the DRA was stimulated by polarized light originating from a single point in the visual field with a common angle of polarization. A wide range of polarization tunings was subsequently revealed in downstream neurons, supporting the idea that the Drosophila medulla dorsal rim area (MEDRA) analyzes the overall pattern of polarized light in the sky and extracts a predominant angle of polarization (AoP), rather than performing many local AoP estimates. During the morning and evening when D. melanogaster are most active, the pattern of polarization in the sky can be well approximated by a single, predominant AoP. DmDRA1 neurons appear to spatially integrate polarization signals from multiple columns of the MEDRA, and individual neurons heavily overlap each other. This could provide an additional robustness to occlusions of the sky or of the DRA itself and average out inconsistencies in the available light (Hardcastle, 2021).

The parallel circuitry between DRA R7, DmDRA1, and MeTu neurons in MEDRA columns, resembles the color-processing pathway found in non-DRA columns involving R7, Dm8, and Tm5c. MeTu neurons in the MEDRA may also integrate color signals, as their dendritic fields extend into the non-DRA medulla, indicating that color and polarization processing are compatible. Parallel circuits may support antagonistic processing of the color and polarization pathways downstream, potentially providing a means to selectively process polarization cues for navigation from the anti-solar hemisphere where they are strongest. The responses of DmDRA2 neurons that contact R8 in this study have not been functionally described, and these neurons may be differently integrated with color processing. Both parallel functions will likely need to be incorporated to build a complete conceptual model of skylight polarization processing in the medulla (Hardcastle, 2021).

In the anterior optic tubercle (AOTU), polarization-sensitive neuron populations were found entering and leaving the tubercle via the intermediate-lateral domain. Polarization responses in the lateral domain were also observed, although it is unclear whether this is a result of separate polarization-sensitive MeTu types projecting from the MEDRA to different AOTU domains. Alternatively, since MeTu neurons are also postsynaptic in the AOTU, signals from a single polarization input channel could be redistributed to different regions of the AOTU for integration with other visual modalities or bilateral interactions. The AOTU in Drosophila is also likely to be a site for modulation of signals depending on time or internal states, and a capacity to modify responses may explain why multiple polarotopic organizations were observed in a MeTu neuron population in the AOTU. However, there may also be multiple functional subtypes within the population that more tailored experiments may be able to distinguish (Hardcastle, 2021).

Intriguingly, none of the polarotopies found in presynaptic MeTu neurons matched the polarotopy of postsynaptic TuBu dendrites in the AOTU, which was extremely consistent across animals. The findings suggest that TuBu neurons extract a processed form of the signals in the AOTU, encoding visual features within fewer neurons than the MeTu populations. TuBu neurons appear to divide signals into functional groups, and the anterior bulb-projecting TuBua group in every fly contained a set of around six tunings covering -90° to +90° of polarization space in approximately 30° steps, tightly packed in a micro-glomerular structure with no apparent polarotopy. The question remains open as to whether a sun position system and skylight polarization system are independent in the bulb. Unlike the TuLAL neurons in locusts (homologous to TuBu), where there is convergence on the dendrites of postsynaptic neurons, TuBu neurons appear to form one-to-one contact with individual ring neurons. Hence, it is posited that the site of integration of celestial cues is not at the synapse between TuBu and ring neurons. Although evidence was found that angles of polarization are represented in the superior bulb, where unpolarized cues are also known to be represented, the populations that were recorded contained a limited range of tunings and resembled a system for detecting visual features with a particular polarization signature, such as horizontally polarized light reflected from surfaces like water, rather than a system for accurate estimation of orientation. Such responses would likely be mediated by more ventral regions of the eye than the DRA. It should be noted that the polarized light stimulus broadly illuminated the eye from a dorsal position and, although attempts were made to minimize reflections, whether reflected polarized light fell on the ventral eye during these experiments was not measured (Hardcastle, 2021).

By recording the ensemble response of a population of ER4m ring neurons, both in the anterior bulb and ellipsoid body (EB), it was determined that they do not simply relay the responses of presynaptic TuBua neurons to the EB. Instead, they appear to deliver a subset of signals more prominently than others, bestowing the population with an ensemble response tuned to a specific angle of polarization. Furthermore, this study found that this population tuning conveys a different angle of polarization in individual animals, and one exciting possibility is that this represents a flexible heading signal relative to polarized light cues, which could direct behavior). A question to address in future work is whether the preferred angle of polarization of an individual ring neuron is itself fixed, in which case this study may have observed the result of a winner-take-all competition among the ER4m population in the EB, or if the whole population flexibly re-tunes to preferentially respond to a common AoP. Recordings from individual neurons will be required to resolve this (Hardcastle, 2021).

It is clear that among ER4m and E-PG neurons, polarization tunings are not represented with a retinotopic map in the EB or PB that is common between individual animals. This is in contrast with the consistent polarotopic organizations found upstream in the MEDRA or AOTU, but in agreement with a previous study which showed that the azimuthal position of unpolarized visual stimuli is also not represented retinotopically in E-PG neurons. The lack of organization in E-PG responses also matches previous findings in the corresponding CL1a neurons in locusts, but contrasts with the polarotopic organization found in other columnar neurons in the locust CX, such as CPU1, and the tangential TB1 neurons. A potential explanation for the lack of consistent polarotopy in CL1a, or indeed E-PG neurons, was offered in a previous study: at least two of each neuron type innervates an individual glomerulus in the PB. Could each of these have differential responses to polarized light to enable different configurations across the PB? Intriguingly, the TB1-like Δ7 neurons in the Drosophila PB appear to synapse onto only a subset of the E-PG neurons in a single glomerulus, perhaps indicating independent functional groups. Therefore, a polarotopic organization of responses in the Drosophila CX might yet be found. Alternatively, such an organization may reflect a common, genetically pre-programmed directional goal to facilitate migration, which flies may lack, instead using polarization cues to follow a fixed course and disperse along idiosyncratic heading (Hardcastle, 2021).

The current data suggests that in a given fly, E-PG neurons may respond to one of two approximately orthogonal angles of polarization, effectively dividing the population into two groups. Interestingly, when data from locust CPU1 neurons (likely homologues of P-F-R neurons in Drosophila) were pooled with tunings obtained from a number of other polarization-sensitive columnar CX neuron types, including CL1b (P-EG), CL2 (P-EN), CPU2, and CPU4 (P-FN), the organization of tunings in the locust PB could be interpreted as clustering around two orthogonal preferred angles. A binary system such as this would be well suited to influence downstream processes in a motor-centered coordinate frame. For example, the eventual output of the compass network may be a command signal to activate one descending neuron of a bilateral pair to initiate a turn to either the left or right, and thus maintain a heading specified by polarization patterns in the sky (Hardcastle, 2021).

An important next step will be to understand how polarized light influences the activity bump in columnar neurons and whether the activity of columnar neurons reciprocally influences the tunings of ER4m neurons. No activity bump was observed in E-PG neurons in the PB, likely due to the open-loop stimulus presentation as well as recordings performed in immobilized animals, although evidence was seen of flexible encoding of polarization information. According to these mappings of E-PG responses in the PB, the influence of a rotating polarized light stimulus might be to move the activity bump discontinuously between two positions, not dissimilar to observations in a recent investigation of the influence of airflow on the bump in E-PG neurons (see A neural network for wind-guided compass navigation). However, a limitation of the polarization stimulus used in this study is that the intensity gradient and position of the light source did not change as the angle of polarization rotated, as they would be seen by an animal turning under a natural sky. If the ambiguity between 0/180° polarization cues is resolved by integrating light intensity information, then the stimulus used in this study presented contradictory, unnatural changes (Hardcastle, 2021).

Behavioral studies in ants and dung beetles have demonstrated that skylight polarization cues can have a greater influence than other visual features in guidance and navigation behaviors. In Drosophila, intensity gradients have been shown to have a greater behavioral significance than polarized light, yet recent connectome analysis of the Drosophila CX highlights the polarization-sensitive ring neurons that were identified in this study as potentially being at the top of a hierarchy of sensory inputs. Furthermore, the unique pattern of asymmetrical connectivity between the ER4m populations from each brain hemisphere and the E-PG network hints at an attractively simple system for obtaining 360° heading information from ambiguous 0/180° polarization cues, by using signals from one population or the other depending on which side of the animal the sun is on. A key challenge for future studies will be to uncover such mechanisms for integrating and selecting from the multiple sensory modalities and visual qualities represented in the central brain in order to navigate complex environments (Hardcastle, 2021).

Insects are ectothermal animals that are constrained in their survival and reproduction by external temperature fluctuations which require either active avoidance of or movement towards a given heat source. In Drosophila, different thermoreceptors and neurons have been identified that mediate temperature sensation to maintain the animal's thermal preference. However, less is known how thermosensory information is integrated to gate thermoresponsive motor behavior. This study used transsynaptic tracing together with calcium imaging, electrophysiology and thermogenetic manipulations in freely moving Drosophila exposed to elevated temperature and identify different functions of ellipsoid body ring neurons, R1-R4, in thermoresponsive motor behavior. The results show that warming of the external surroundings elicits calcium influx specifically in R2-R4 but not in R1, which evokes threshold-dependent neural activity in the outer layer ring neurons. In contrast to R2, R3 and R4d neurons, thermogenetic inactivation of R4m and R1 neurons expressing the temperature-sensitive mutant allele of dynamin, shibire(TS), results in impaired thermoresponsive motor behavior at elevated 31 °C. trans-Tango mediated transsynaptic tracing together with physiological and behavioral analyses indicate that integrated sensory information of warming is registered by neural activity of R4m as input layer of the ellipsoid body ring neuropil and relayed on to R1 output neurons that gate an adaptive motor response. Together these findings imply that segregated activities of central complex ring neurons mediate sensory-motor transformation of external temperature changes and gate thermoresponsive motor behavior in Drosophila (Buhl, 2021).

Neural representations of head direction (HD) have been discovered in many species. Theoretical work has proposed that the dynamics associated with these representations are generated, maintained, and updated by recurrent network structures called ring attractors. This theorized structure-function relationship was evaluated by performing electron-microscopy-based circuit reconstruction and RNA profiling of identified cell types in the HD system of Drosophila melanogaster. Motifs were identified that have been hypothesized to maintain the HD representation in darkness, update it when the animal turns, and tether it to visual cues. Functional studies provided support for the proposed roles of individual excitatory or inhibitory circuit elements in shaping activity. This study also discovered recurrent connections between neuronal arbors with mixed pre- and postsynaptic specializations. These results confirm that the Drosophila HD network contains the core components of a ring attractor while also revealing unpredicted structural features that might enhance the network's computational power (Turner-Evans, 2020).

Mammalian head direction (HD) cells provide one of the clearest examples of an internal representation of an animal's relationship to its surroundings. The HD representation uses visual cues in the environment as a reference but persists in darkness, where it is updated by self-motion cues. This internal representation likely guides navigation behaviors. Indeed, perturbations to the system in rats induce errors in path integration. Theoretical studies have proposed that the neuronal population dynamics associated with HD representations are maintained by network structures called ring attractors. These recurrent networks are often schematized as a ring of neurons whose connectivity depends on their directional tuning preferences. In most models, neurons with similar directional tuning excite each other and those with different tuning inhibit each other, thereby enabling the generation of a stable pattern of localized activity in any part of the network. Recurrent loops with a shift move the activity 'bump' around the ring as the animal turns, and the compass-like representation uses visual inputs as a reference. Experimental support for this general theoretical formulation has come from analyses of HD cell population activity under a variety of different conditions. However, different network implementations of this general formulation make distinct assumptions about the connectivity of their constituent neurons. Importantly, such assumptions, which dictate exactly how the circuit functions, have been difficult to test in the large mammalian brain. Mammalian HD cells are distributed across many brain regions and are as yet not well classified into subtypes, making it challenging to identify and target them reliably (Turner-Evans, 2020).

An internal representation similar to the mammalian HD representation has also been discovered in the insect brain (Seelig, 2015; Schematic of fly central brain and CX: ellipsoid body (EB), fan-shaped body (FB), protocerebral bridge (PB), paired noduli (NO), lateral accessory lobe (LAL) and gall (Gall). MB: mushroom body). Further, there is strong evidence that this HD representation is implemented by a ring attractor. A ring attractor model that assumes recurrent connectivity between heading neurons and neurons that encode both angular velocity and heading maintains an accurate heading representation when driven by realistic velocity inputs. This model replicates the fly HD network's dynamics in darkness. Other models have invoked plasticity between visual inputs and heading neurons to show how visual and angular velocity information might update the representation in a mutually consistent manner (Turner-Evans, 2020).

Importantly, however, all fly ring attractor models have assumed the circuit's connectivity based on relatively indirect evidence. For example, the location of pre- and postsynaptic arbors has been inferred from whether neural processes visible in light-microscopic images seem spiny or bouton-like in specific substructures. The hypothesized connectivity of the circuit has then been derived from light-level overlap between the putatively pre- and postsynaptic arbors of neurons, in some cases further supported by GFP reconstitution across synaptic partners (GRASP) and trans-Tango experiments, although the reliability and accuracy of these methods to estimate pairwise connectivity is known to be limited. Similarly, measurements of functional connectivity by optogenetic stimulation of a population of one type of neurons and calcium imaging of another can be difficult to interpret within recurrent networks (Turner-Evans, 2020 and references therein).

In the current study, reconstructions were used based on serial transmission electron microscopy (EM). to determine synaptic connectivity within the neural network underlying the fly's HD representation. Neural connectivity matrices were compared to those extracted from the recently released fly hemibrain connectome, which was obtained by using focused ion beam scanning electron microscopy (FIBSEM). Cell-type-specific RNA sequencing (RNA-seq) and fluorescence in situ hybridization (FISH) allowed characterization of the expression profiles of the key cellular components of the ring attractor network. This integrated information was used to assess the role of each of the constituent cell types in the ring attractor's dynamics. It was found that the fly HD network contains motifs similar to those proposed in theoretical models. These motifs were hypothesized to maintain HD activity and update it both in the dark and when visual features are present. These ideas were tested using targeted two-photon calcium imaging and thermogenetic perturbations of the constituent neuron types in behaving, head-fixed Drosophila. It was also found that many neurons have mixed pre- and postsynaptic specializations within their innervations to single brain structures, creating 'hyper-local' recurrent loops that may allow local computations to supplement the role of recurrence at the network level. Moreover, although many ring attractor models rely on distinct units that provide local excitation and long-range inhibition to shape activity into one stable bump, consistent with the biological results, apparent redundancy was found in these structural elements. Taken together, these results provide new structural and functional insights into how a small biological ring attractor network allows an animal to maintain an accurate internal sense of direction (Turner-Evans, 2020).

Efforts to model the dynamics of HD networks have long focused on ring attractors. Although this conceptual framework has been very influential, testing the validity of the structural assumptions and functional predictions made by ring attractor models in biological circuits has been challenging. This study has bridged this gap by characterizing the neurons that compose the fly HD circuit and examining how they together produce ring attractor dynamics. A previous study proposed a ring attractor model that captured the observed dynamics of the fly HD network. The weights of this model were tuned without any knowledge of the actual synaptic connections. If the synapse counts measured in this work are extrapolated to the entire biological network, the theoretical and the experimentally derived connectivity matrices are remarkably similar (Turner-Evans, 2020).

Some of these results will require follow-up experimental and theoretical work. The long-range inhibition that has been invoked to form a stable bump of activity appears to be split across multiple brain regions and multiple classes of neurons. In addition to the Δ7s, GABAergic R4d (and other) visual ring neurons also inhibit much of the ellipsoid body-protocerebral bridge-gall neuron (E-PG) population in the presence of visual stimuli. Further, the gall-EB ring neuron population provides a source of inhibition onto the E-PGs. Indeed, the cumulative inhibition from multiple classes of ring neurons may account for some of the mutual suppression between E-PGs (Turner-Evans, 2020).

Local excitation also appears to be implemented through multiple classes of neurons. The redundancy of local excitation and long-range inhibition may allow the network to maintain HD activity in the absence of external input. These network features may also provide a means to stabilize the heading representation in the presence of noisy inputs and inhomogeneous synaptic weights, which can disrupt the function of continuous attractor networks (Turner-Evans, 2020).

Synapses between visual ring neurons were also observed, consistent with previous observations from EM in locusts and trans-Tango experiments in flies. Visual scenes are often complex and dynamic, and the observed inter-ring neuron connections may either form a winner-take-all network that leads one visual feature to dominate, preventing the bump from moving erratically as the scene shifts over time, or provide a mechanism for gain control that normalizes the total level of inhibition from those ring neurons onto the E-PGs (Turner-Evans, 2020).

At a methodological level, the recently completed FIBSEM-based connectome reinforces the conclusions from FAFB. Although there are sometimes more connections detected between specific types in one or the other dataset, only rarely are connections in one dataset not seen in the other (Turner-Evans, 2020).

Finally, the results highlight the challenges of linking structure to function. Connectomics is often perceived as providing strong constraints on models of circuit function. However, although this was partly true for the fly HD network, the results also identified unexpected computational mechanisms that the circuit may employ. For example, the profusion of mixed pre- and postsynaptic specializations within single compartments of individual neurons creates many more locally recurrent loops between similar and different neuron classes, and a functional account for such dense recurrence cannot yet be provide. Future theoretical and experimental work to probe their function may well reveal that fly brains are more powerful than their numerical simplicity might suggest (Turner-Evans, 2020).

Recent studies of the Central Complex in the brain of the fruit fly have identified neurons with activity that tracks the animal's heading direction. These neurons are part of a neuronal circuit with dynamics resembling those of a ring attractor. The homologous circuit in other insects has similar topographic structure but with significant structural and connectivity differences. This study modeled the connectivity patterns of two insect species to investigate the effect of these differences on the dynamics of the circuit. The circuit found in locusts can also operate as a ring attractor but differences in the inhibition pattern enable the fruit fly circuit to respond faster to heading changes while additional recurrent connections render the locust circuit more tolerant to noise. These findings demonstrate that subtle differences in neuronal projection patterns can have a significant effect on circuit performance and illustrate the need for a comparative approach in neuroscience (Pisokas, 2020).

The availability of tools for the study of insect brains at the single neuron level has opened the way to deciphering the neuronal organisation and principles of the underlying circuit's behaviour. However, even where there is progress towards a complete connectome, the lack of data on synaptic strengths, neurotransmitter identity, neuronal conductances, etc. leave many parameters of the circuit unspecified. Exploring these parameters via computational modelling can help to illuminate the functional significance of identified neural elements. This study has applied this approach to gain greater insight into the nature of the heading encoding circuit in the insect central complex (CX), including the consequences of differences in circuit connectivity across two insect species (Pisokas, 2020).

This study has focused on a subset of neurons in the PB and EB which have been hypothesised to operate as a ring attractor, with a 'bump' of neuronal activity moving across columns consistently with the changing heading direction of the animal. The neuronal projection patterns and columnar organisation differ between the two insect species that were analysed, the fruit fly and the locust. There are additional morphological columns in the PB of flies (9 vs. 8), resulting in a different number of functional units that could influence the symmetry of the underlying neural circuits. Also, the EB in the fruit fly forms a physical ring, while the homologous region in the locust is an open structure. This analysis of the connectivity as a directed graph has revealed, surprisingly, that the circuits are nevertheless equivalent in their effective structure, forming a closed ring attractor in both species with an identical functional role for each neuron class. The preservation of this circuit across 400 million years of evolutionary divergence suggests that it is an essential, potentially fundamental, part of the insect brain (Pisokas, 2020).

It is worth noting that an essential part of the circuit, namely the functionally closed ring that was found in both species, is achieved with two different solutions. In the fruit fly, the torus-shaped EB provides an anatomical solution to the closure of the ring via overlapping projections from E-PG neurons innervating the innermost and outermost PB glomeruli. In contrast, in the locust the midline spanning output fibers of the E-PG neurons in the medial protocerebral bridge (PB) glomeruli serve this function in combination with a slightly different projection pattern that results in the P-EN (protocerebral bridge-ellipsoid body-noduli neurons) forming reciprocal connections back to E-PG neurons in the same octant. In this context, it is interesting to note that neither solution to this problem is possible for insects of a different order, the lepidoptera (moths and butterflies). These insects have an almost straight EB, their PB is split along the midline, and right-left connections between the two halves are realised by a neuropil-free fiber bundle. Neither midline crossing E-PG fibers within the PB, nor local connections around the ring of the EB are therefore morphologically possible, suggesting that a functional closure of the heading direction circuit is either not required or achieved via other means in these species. The notion that there are many solutions to the same problem is further highlighted by data from bumblebees showing the existence of a ninth E-PG neuron that connects the medialmost PB glomerulus to the outermost ipsilateral EB wedge, closing the ring in yet another way. Exploring these different solutions across many species could provide key insights into the evolution of this circuit under a multitude of evolutionary history constraints (Pisokas, 2020).

In combination, these findings underline that the large-scale anatomical differences at the level of neuropils and projection patterns do not necessarily affect the core functions of the circuit. Rather, the functional constraints appear significant enough that even in those parts of the circuit that clearly differ between species convergent solutions have evolved that solve similar problems, albeit in slightly different ways Surprisingly, more subtle differences in the morphology between the two species have significant effects on the dynamical response of the heading direction circuit. First, the shape of the dendritic arborizations of one type of CX neuron determines how quickly the model circuit tracks rotational movements. Second, a difference in the overlap of neuronal projections in the EB results in an extra feedback loop between the P-EN and E-PG neurons in the locust circuit that makes it more robust to synaptic noise (Pisokas, 2020).

It is suggested that the effects of these differences are consistent with the behavioural ecology of the two species. On the one hand, the faster response of the ring attractor circuit in the fruit fly accommodates the fast body saccades that fruit flies are known to perform. On the other hand, the locust is a migratory species, so its behaviour demands maintenance of a defined heading for a long period of time. This requirement for heading stability might have provided the selective pressure needed to drive the evolution of a more noise resilient head direction circuit (Pisokas, 2020).

As any model, the circuits are necessarily condensed and simplified versions of the real circuits in the insect brain. In comparison to previous models, the work presented in this study has been more precisely constrained by the latest anatomical evidence. The models were additionally constrained to use plausible values for the biophysical properties of neurons (membrane conductance and capacitance) as well as spiking rates (background activity) supported by electrophysiological evidence. Furthermore, in building these models it was not assumed that the underlying circuits must be ring attractors, but rather whether, given the available connectivity data, they can be was asked and investigated. This was especially the case for the locust model since this work represents the first model of this circuit to date. Nevertheless, it is important to outline those areas where the assumptions cannot be fully justified from the existing data and to identify the potential consequences for the modelling results (Pisokas, 2020).

In the model of the fruit fly heading tracking circuit, a uniform distribution was assumed of dendrites across the Delta7 neurons. Imaging of these neurons suggests that there might be a subtle variation of the dendritic density along their length. However, it is unclear how this subtle variation might be related to synaptic density and efficacy. Therefore, the simplifying assumption was initially made that these neurons have uniform synaptic efficacy across the PB. However, this study also explored the effect of varying the degree of synaptic uniformity, showing that there is a range of synaptic efficacy distributions that still can produce the fly-like rapidity in the circuit response (Pisokas, 2020).

In general, arborization trees of neurons in the CX can be very complex, as they are not only confined to specific slices, but also to one or several layers, especially within the EB. In Drosophila, the spiny terminal arbors of E-PG neurons extend to the width of single wedges in the EB, occupying both the posterior and medial layers. In contrast, P-EG and P-EN neurons arborize in tiles, hence innervating only the posterior surface volume of the EB. Therefore, it is assumed that presynaptic terminals of P-EG and P-EN neurons form synapses with E-PG postsynaptic terminals in the posterior layer of the EB. In locusts, the E-PG arborizations are more complex, as these cells innervate a single wedge of the anterior and medial EB layers, but extend at least twice this width to either side in the posterior layer that provides overlap with the P-EN neurons. Additionally, the wider fibres have a different morphological appearance. P-EG neurons in this species innervate all layers evenly. Although these detailed differences likely have consequences for connectivity, these arborizations were simplified to their most essential components, aiding the extraction of the core features. With the advance of comparative connectomics, these aspects will become accessible for investigation (Pisokas, 2020).

Several assumptions were made while deriving the neuronal connectivity in these models. Well delineated borders of synaptic domains were assumed, which is clearly not always the case. Especially in the EB, some overlapping of neighbouring synaptic domains due to stray terminals is to be expected. The circumferential extent of arbors in wedges and tiles may affect the integrity of the resulting circuit and its properties. However, due to lack of adequate data about the extent of such overlap it is not possible to currently model this aspect in a sensible way (Pisokas, 2020).

Furthermore, neuronal connectivity was mostly inferred by co-location of neuronal arbors, that is, projection patterns. A functional connectivity study has reported that stimulation of E-PG neurons triggered significant responses to Delta7 neurons but no columnar neurons. However, as that study noted, the lack of response might be due to the limitations of the method used. Alternatively, such connections might be mediated by interneurons instead of being monosynaptic. Future work using electron microscopy data will elucidate which of the overlapping arborizations correspond to functional connections and allow augmentation of these models (Pisokas, 2020).

Further assumptions were made about neuronal polarity, type of synapses and synaptic efficacies. One study hsd characterise the EB arbor of E-PG neurons in Drosophila as having both presynaptic and postsynaptic domains; however, another study reports that using anti-synaptotagamin is inconclusive for presynaptic terminals. In the models presented in this studyfor both the fruit fly and the locust it was thus assumed that E-PG neurons are purely postsynaptic in the EB, following the most parsimonious polarity estimate. Connectomics data from a recent preprint demonstrate that in Drosophila synapses exist that directly link Delta7 to E-PG neurons in the PB. These synapses are most likely inhibitory and would thus inhibit the distal portion of the ring and thus would not alter the location of the activity 'bump'. For simplicity and because they do not affect the functional layout of the circuit, these synapses were not included in the model (Pisokas, 2020).

Furthermore, the Delta7 neurons are assumed to have inhibitory effect on their postsynaptic neurons, as a previous study proposed. However, there is evidence that Delta7 neurons make both inhibitory and excitatory synapses to other neurons. Indeed, these cells were recently shown to be glutamatergic, enabling both inhibitory and excitatory effects on postsynaptic cells via different glutamate receptors (Turner-Evans, 2019). As the downstream neurons with demonstrated excitatory responses (P-FN neurons) are not part of the current model, the simplifying assumption was made that Delta7 neurons have exclusively inhibitory effect on their postsynaptic neurons, both in flies and locusts. It is also possible that there are other sources of inhibition in the circuit, for example mediated by the GABAergic ring neurons originating in the bulbs, as suggested by one study, or via GABAergic Gall-EB ring neurons. This study did not explore this possibility (Pisokas, 2020).

It was additionally assumed that the synaptic strengths of all synapses of each class are identical. This might not be the case in the actual animals, especially considering that one of the EB tiles (T1) is innervated by twice as many neurons as other tiles in fruit flies (see Projection patterns of the excitatory portion of the fruit fly circuit). Neurons innervating this tile might have reduced synaptic efficacy in order to maintain the radial symmetry of the circuit intact. Similarly, the synaptic strengths of the neurons closing the ring in locusts would be expected to be different than those of other synapses if the ring does not have a functional 'seam'. Such a synaptic efficacy variation is suggested by the fact that the arborization density of E-PG neurons innervating the two medial PB glomeruli (G9 and G1) is not the same in both of them. There is certainly space for further exploration of the effect of synaptic efficacy in those segments of the ring in both species. Finally, synaptic strength variation might exist for the two Delta7 neurons that have presynaptic terminals in three glomeruli instead of two (Pisokas, 2020).

All types of neurons in these models were assumed to have the same nominal biophysical properties even though anatomical evidence has shown that their morphology, somata size and main neurite thickness differ. To relax this assumption, the effect of heterogeneity in the biophysical properties of the neuronal population was explored. The conclusions were corroborated using both rate-based and Leaky Integrate and Fire neurons with refractory period. This allowed highlighting of the significance of the neuronal connectivity on the circuit dynamics. The point spiking neuron model was sufficient for investigating the performance characteristics, spike timing dynamics and potential spike synchronisation effects in the ring attractors when exposed to neuronal noise, but clearly is highly abstracted in comparison to real neurons. However, most of the necessary detail is lacking to constrain more complex neural models. One caveat is that intrinsic properties of neurons could provide short-term memory that would radically alter the circuit response. It is not possible to explore this possibility with the models that were used, but it can be concluded that such properties do not appear to be necessary for generating basic ring attractor dynamics. Furthermore, it will be interesting to study how differences in the biophysical properties of neurons between the two species might be affecting performance. This possibility was not explored in this study (Pisokas, 2020).

This work has compared the hypothetical heading tracking circuit of two evolutionary distant species. The study beyond mere simulation of neuronal projection data by analysing and deriving the effective underlying circuit structure of the two ring attractors. This analysis and derivation of the complete effective neuronal circuits revealed not only differences in dynamics but also the construction principles of these circuits. This approach allowed identification of elements that differ in several ways from the 'canonical' ring attractor described in earlier theoretical models (Pisokas, 2020).

For example, the circuit found in the two insect species combines two functionalities in the P-EN neurons that are typically assigned to separate neural populations in computational models of ring attractors. Such computational models use one set of neurons to provide the lateral excitation to nearest neighbours and a different set of neurons that receive angular velocity input to drive the left-right rotation of the heading signal. In the insect circuit, the P-EN cells are part of the lateral excitation circuit, providing excitation to their two nearest neighbours, but they also receive angular velocity input. This difference is suggestive of a more efficient use of neuronal resources than the typical computational models of ring attractors. Another novel element found in the insect ring attractors is the presence of local feedback loops within each octant of the circuit structure (P-EG to E-PG and P-EN to E-PG). Both of these feedback loops increase the tolerance of the ring attractors to noise (Pisokas, 2020).

Another unique aspect of this modelling work is the comparison of related, but not identical, circuits found in two species. Indeed, using computational modelling allows investigation of 'hybrid' circuits, combining features of each, in order to try to understand the functional significance of each observed difference independently. Nevertheless, some differences between these circuits are not explained by the current model, and may require additional work to fully explicate (Pisokas, 2020).

One question is what is the role, if any, of the ninth PB glomeruli found so far only in Drosophila? In particular, the existence of the innermost glomeruli that are not innervated by the P-EN neurons seems perplexing. The same signals from tile 1 of the EB are sent to both ends of each hemisphere of the PB (glomeruli 1 and 9) and from there action potentials propagate along the Delta7 neurons along the PB length. It is speculated that this may be a mechanism to reduce the distance and time these signals have to travel to cover the full PB, that is, the maximum distance any signal must travel is only half of the distance it would need to propagate from one end of the PB to the other as in other species, such as the locust. If this is the case, it would constitute one more specialisation in Drosophila that reduces the response time of the ring attractor. It therefore seems that several specialisations have been orchestrated in minimising the response delays in fruit flies. Testing this idea would require multi-compartmental models to capture the action potential transmission time along neurites; as argued above, this may be contingent upon first obtaining detailed biophysical characterisation of the Delta7 neurons (Pisokas, 2020).

Another remaining question is what is the role of the closed ring-shaped EB in D. melanogaster. One possibility is that such a closed ring topology would allow local reciprocal connections between P-EN and E-PG neurons all around the EB ring, as reported in Turner-Evans (2019). This would allow direct propagation of signals between these neurons within the EB instead of requiring them to travel via the PB, as in the current model, again increasing the speed with which the heading direction can be tracked and allowing smoother transition between neighbouring tiles. Note that such direct reciprocal connections within the EB can only span the full ring with a closed ring anatomy and would not be possible between the two ends of the EB in the locust. To investigate the potential effect of such hypothetical reciprocal connections within the EB, further studies are required. Possibly blocking signal transmission via the PB to isolate functional connectivity within the EB would allow comparison of signal transmission time measurements within the EB versus via the PB. Such measurements would determine how different and hence significant those two pathways might be in the ring attractor performance (Pisokas, 2020).

A further hypothesis relates to the evolutionary lineage of these two features in the Drosophila CX. It will be of interest to study whether the ring shaped EB appeared before or after the appearance of the ninth glomeruli. One possibility is that the EB evolved into a ring shape after the appearance of the ninth glomeruli in the PB, allowing connections from one common tile to both glomeruli 1 and 9 and hence providing such a common driving signal. Alternatively, a pre-existing ring-shaped EB might have allowed the evolution of usable ninth glomeruli that resulted in faster propagation. Similarly, the P-EN to E-PG recurrency found only in the locust might be an acquired adaptation of the locust that increases robustness to noise, or an ancestral feature that has been lost in fruit flies (Pisokas, 2020).

Comparison of different species could potentially elucidate such questions. Individual species would be expected to have a selective subset of the specialisations that wer found, endowing them with brain circuits supporting the behavioural repertoire suiting their ecological niche. It will, therefore, be informative to analyse the effective heading direction circuit of other species, spanning evolutionary history, in order to get insights into how such adaptations relate to and accommodate behaviour. The results emphasise the importance of comparative studies if it is possible to derive general principles about neuronal processing, even in systems that appear highly conserved such as the CX head direction circuit in insects. Many of the circuit properties observed in Drosophila appear to reflect specific evolutionary adaptations related to tracking rapid flight manoeuvres. Despite the many strengths of Drosophila as an experimental model, it therefore remains important to ground conclusions about the insect brain in comparison with other species (Pisokas, 2020).

The genome versus experience dichotomy has dominated understanding of behavioral individuality. By contrast, the role of nonheritable noise during brain development in behavioral variation is understudied. Using Drosophila melanogaster, this study demonstrated a link between stochastic variation in brain wiring and behavioral individuality. A visual system circuit called the dorsal cluster neurons (DCN; ~40 clustered neurons located in the dorso-lateral central brain) shows nonheritable, interindividual variation in right/left wiring asymmetry and controls object orientation in freely walking flies. DCN wiring asymmetry instructs an individual's object responses: The greater the asymmetry, the better the individual orients toward a visual object. Silencing DCNs abolishes correlations between anatomy and behavior, whereas inducing DCN asymmetry suffices to improve object responses (Linneweber, 2020).

Individual variability in morphology is abundant, including among human identical twins and species that reproduce by parthenogenesis. In this regard, the brain is no exception. Examples of individual brain variation include differences of size, weight, and neuroanatomical parcellations of human brains. In invertebrates, where individual neurons can be identified across animals, single neurons show variability in morphology, wiring, synaptic connectivity, and molecular composition across individuals (Linneweber, 2020).

Similarly, innate behaviors, such as selective attention to stimuli, show individual variation even among genetically identical individuals. The stability of individual differences over time defines behavioral idiosyncrasies as animal individuality. It has been proposed that variability in innate behavior is due to neuromodulation of anatomically hardwired circuits. By contrast, there is evidence for developmental plasticity resulting in a range of possible circuit diagrams among individuals, but whether nonheritable individual anatomical differences in brain wiring can predict distinct behavioral outcomes is unexplored (Linneweber, 2020).

To test whether stochastic wiring of neural circuits affects behavioral variation, Drosophila contralateral visual interneurons called the dorsal cluster neurons (DCNs) (also known as LC14). DCNs exhibit up to 30% wiring variability of their axonal projections between individuals and between the left and right hemispheres of the same brain. DCN axons innervate two alternative target areas in the fly visual system called the medulla (M-DCNs) and the lobula (L-DCNs). The decision whether any given DCN becomes a M-DCN or L-DCN is determined by an intrinsically stochastic lateral inhibition mechanism mediated by the Notch signaling pathway (Langen, 2013). To test the link between wiring variation and behavioral variation, a visual behavioral assay called Buridan's paradigm was used. In this assay, a fly is placed between two identical high-contrast stripes at 180° from each other in a uniformly illuminated arena. The stripes are unreachable, inducing the fly to walk back and forth between them during the assay (Linneweber, 2020).

This study reports that flies show behavioral individuality that is nonheritable and is not reduced through inbreeding. The degree in left-right DCN wiring asymmetry in the medulla is a predictor of behavioral performance of individual flies. The more asymmetric the DCN medulla innervation is, the narrower the path a fly walks between the two stripes. DCN activity is necessary for this correlation, and reengineering DCN asymmetry suffices to change an individual's behavior (Linneweber, 2020).

While analyzing object orientation responses in wild-type Canton S (CS) flies, sex-independent interindividual variability in their trajectories was noted. This study focused on a parameter called absolute stripe deviation (henceforth aSD), measuring the deviation from the narrowest possible path between the stripes. Although males tend to walk narrower paths, the degree of interindividual variation in aSD is the same between males and females. This study therefore continued studies with combined populations (Linneweber, 2020).

To test whether behavioral variability correlates with genetic diversity, a subset of the Drosophila genomic reference panel (DGRP) was for genetically homogeneous strains with extreme object orientation responses. This identified two strains with opposing behavioral phenotypes: DGRP-639 showed low aSD, whereas DGRP-859 showed high aSD. Similar behavioral differences were found in seven other representative behavioral parameters. However, despite the extreme reduction of genetic diversity, the degree of individual variation in aSD was not reduced. On the contrary, DGRP-639 showed increased behavioral variability, hinting at the nonheritability of this variability (Linneweber, 2020).

If the genotype of an individual determines its behavior, repeated breeding of parental animals with a specific behavioral trait should select for a specific behavior, creating a behaviorally homogeneous population. Three pairs with the lowest and highest aSD scores, respectively were mated, and object orientation responses were measured in their offspring. No differences were found between the two sets of offspring in aSD scores as well as six other parameters tested. The same was true for the offspring of a single pair with low and high aSD. The same breeding schemes were repeated with the near-isogenic DGRP-639 and DGRP-859 for seven generations. For most parameters, a breeding pair reproduces the full range of variability in the population at every generation (Linneweber, 2020).

An individual's idiosyncratic behavioral profile may not be heritable either because it is driven by internal-state modulations, or because it is driven by nonheritable developmental mechanisms. To distinguish these possibilities, the same individual CS flies were tested once every other day for 3 days; an individual's behavior was virtually identical over the three trials. Statistical analysis of aSD showed that the individual responses of CS flies on different days were correlated. The same was true for path details like left- or right-shifted angles, distance, full walks, meander, absolute horizon deviation, absolute angle deviation, angle deviation, and center deviation. This analysis was extended over a 4-week period. The object responses of individuals were stable over this extended period. This stability argued against state modulations and in favor of individual properties. Indeed, starvation followed by refeeding over a period of 3 days failed to reduce stability of individual performances despite obvious changes in mean population behavior. Finally, it was asked whether reduced genetic diversity affects behavioral stability. Repeated testing of DGRP-639 and DGRP-859 individual flies was performed; both inbred strains showed temporally stable individual responses (Linneweber, 2020).

Together, these data show that individual variability in object orientation is a nonheritable, temporally stable trait that is independent of sex, genetic background, and genetic diversity. Where in the brain might such individuality in visual behavior originate (Linneweber, 2020)?

It has been suggest that object position processing in Drosophila requires qualitative asymmetry of the visual percept of an object. However, direct evidence for this notion is lacking, especially that the sizes of the left and right eyes of the same fly are highly correlated. It has been suggested that binocular interactions, through higher-order commissural visual interneurons, are required for object orientation). Putting the two predictions together this study hypothesized that variation in object orientation responses is regulated by the variation in the asymmetry of a higher-order contralateral visual circuit innervating the frontal visual field. The DCNs match this predicted circuit (Linneweber, 2020 and references therein).

To obtain a comprehensive description of DCN wiring, this study extended the previous analyses of DCNs that were based on 16 female flies, to 103 males and females. The number of DCNs varied from 22 to 68 cells, with a range of 11 to 55 L-DCNs and 6 to 23 M-DCNs. In addition, a distribution of variation was observed in medulla-targeting asymmetry by M-DCNs. The distribution of all DCN asymmetries showed a peak of low asymmetries, although extreme asymmetries were present but rare. Finally, three-dimensional reconstruction showed that M-DCN axons terminate in the posterior medulla, where visual columns from the frontal visual field are located, and the DCN wiring pattern in the medulla does not change in the adult (Linneweber, 2020).

DCNs represent an ideal candidate for an intrinsically asymmetric population of contralateral higher-order interneurons to mediate object responses. To test this hypothesis, it was first asked whether the DCNs were required for object orientation. Inactivating either all DCNs or only M-DCNs resulted in a strong increase in aSD. Next, the relationship between individual variability in object orientation behavior and individual variability in DCN wiring (N = 103) was queried. Unbiased correlation analysis between 36 behavioral parameters and 37 prominent DCN anatomical features showed that left-right asymmetry in M-DCN innervation correlated with an individual's aSD and other interdependent parameters. Individuals with high M-DCN asymmetry have a low aSD, whereas individuals with symmetric M-DCN have a high aSD. To test if DCN wiring asymmetry is a functional driver of individual object orientation behavior, DCNs were silenced and the analyses was repeated. This abolished the correlation between M-DCN asymmetry and aSD, but not stripe detection per se (Linneweber, 2020).

The data show that nonheritable developmental variation in DCN wiring asymmetry is necessary for creating variability in object orientation behavior across individuals. It was therefore asked whether the changed object orientation responses in the DGRP strains reflect DCN asymmetry alterations. The low aSD strain DGRP-639 displayed more DCN wiring asymmetry, and the high aSD strain DGRP-859 less DCN wiring asymmetry, consistent with the hypothesis. Next, the DCNs were developmentally rewired either by blocking endocytosis to inhibit developmental signaling among DCNs or by activating the Notch pathway, both in a DCN-specific fashion. This resulted in reduced DCN wiring asymmetry and a correspondingly higher aSD, while preserving the correlation between wiring and behavior. Finally, flies were genetically engineered to generate one-sided DCN clones expressing the neuronal silencer Kir2.1. Animals with asymmetrically silenced clones showed lower aSD scores than controls with unsilenced clones or no clones at all. Together, these data causally link DCN wiring asymmetry to object orientation responses (Linneweber, 2020).

Finally, to test the hypothesis further, it was asked if generating any asymmetry in visual processing is sufficient to override high stripe deviation. Among 79 CS flies tested, the 20 with the highest aSD indices (>40) were selected, monocular deprivation was performed, and they were tested again. This resulted in a reduction of aSD in these flies, as well as in the entire population (Linneweber, 2020).

This study has establish a link between variability in the development of the brain and the emergence of individuality of animal behavior. The work shows that intrinsically stochastic mechanisms of brain wiring give rise to intraindividual variation of left-right asymmetry in the innervation of the fly visual areas, which explains the individuality of behavioral differences in object responses. The amenability of the relatively complex Drosophila brain to multiscale analysis, from the molecular to the behavioral, at single-animal resolution makes it a model for understanding the emergence of individuality at each of these scales. It is speculated that similar mechanisms and consequences will hold true in other species, including humans (Linneweber, 2020).

Previous work in Drosophila visual behavioral neuroscience led to the proposal that asymmetry in visual information processing influences object responses. Where such functional asymmetry lay and how it might arise has, until now, remained unclear. Independently, the study of object responses in motion-blind mutants led others to propose a hypothetical contralateral circuit dedicated to object responses in the frontal visual field. The discovery that DCN asymmetry drives object orientation responses in individuals is an elegant solution combining both predictions: a contralateral asymmetric visual circuit that regulates object orientation in the frontal visual field. Future work will reveal the exact physiological consequences of morphological asymmetry, such as whether wiring asymmetry induces timing differences as in auditory navigation or whether the absolute differences are simply summed up (Linneweber, 2020).

This work provides evidence for the generation of multiple brain and behavior phenotypes from the same genotype via developmental stochasticity and noise. This can serve as a robustness factor for both the individual and the population by increasing the chances of survival of any given genome in case of strong selection pressure (Linneweber, 2020).

The brain perceives visual information and controls behavior depending on its underlying neural circuits. How UV information is represented and processed in the brain remains poorly understood. In Drosophila melanogaster, UV light is detected by the R7 photoreceptor that project exclusively into the medulla layer 6 (M(6)). This study imaged 28,768 single neurons and identified 238 visual projection neurons linking M(6) to the central brain. Based on morphology and connectivity, these visual projection neurons were systematically classified into 94 cell types belonging to 12 families. Three tracts connected M(6) in each optic lobe to the central brain: One dorsal tract linking to the ipsilateral lateral anterior optic tubercle (L-AOTU) and two medial tracts linking to the ipsilateral ventral medial protocerebrum (VMP) and the contralateral VMP. The M(6) information was primarily represented in the L-AOTU. Each L-AOTU consisted of four columns that each contained three glomeruli. Each L-AOTU glomerulus received inputs from M(6) subdomains and gave outputs to a glomerulus within the ellipsoid body dendritic region, suggesting specific processing of spatial information through the dorsal pathway. Furthermore, the middle columns of the L-AOTUs of both hemispheres were connected via the intertubercle tract, suggesting information integration between the two eyes. In contrast, an ascending neuron linked each VMP to all glomeruli in the bulb and the L-AOTU, bilaterally, suggesting general processing of information through the ventral pathway. Altogether, these diverse morphologies of the visual projection neurons suggested multi-dimensional processing of UV information through parallel and bilateral circuits in the Drosophila brain (Tai, 2020).

Sensory systems rely on neuromodulators, such as serotonin, to provide flexibility for information processing as stimuli vary, such as light intensity throughout the day. Serotonergic neurons broadly innervate the optic ganglia of Drosophila. This study mapped of patterns of serotonin receptors in the visual system, focusing on a subset of cells with processes in the first optic ganglion, the lamina. Serotonin receptor expression was found in several types of columnar cells in the lamina including 5-HT2B in lamina monopolar cell L2, required for spatiotemporal luminance contrast, and both 5-HT1A and 5-HT1B in T1 cells, whose function is unknown. Subcellular mapping with GFP-tagged 5-HT2B and 5-HT1A constructs indicated that these receptors localize to layer M2 of the medulla, proximal to serotonergic boutons, suggesting that the medulla neuropil is the primary site of serotonergic regulation for these neurons. Exogenous serotonin increased basal intracellular calcium in L2 terminals in layer M2 and modestly decreased the duration of visually induced calcium transients in L2 neurons following repeated dark flashes, but otherwise did not alter the calcium transients. Flies without functional 5-HT2B failed to show an increase in basal calcium in response to serotonin. 5-HT2B mutants also failed to show a change in amplitude in their response to repeated light flashes but other calcium transient parameters were relatively unaffected. While serotonin receptor expression in L1 neurons was not detected, they, like L2, underwent serotonin-induced changes in basal calcium, presumably via interactions with other cells. These data demonstrate that serotonin modulates the physiology of interneurons involved in early visual processing in Drosophila (Sampson, 2020).

Navigation in many animals involves an internal sense of heading direction. Such a sense of heading is thought to be mediated by neurons that are specifically active when the animal is orienting toward the neuron's preferred direction. These head direction cells reside in multiple brain regions and interact with each other as well as with other neurons, including cells encoding angular velocity (Shiozaki, 2020).

In Drosophila melanogaster, two types of columnar neurons innervating the ellipsoid body (EB) and the protocerebral bridge (PB), subregions of the central complex (CX), encode heading direction. As with head direction cells in mammalian brains, columnar neurons encode heading as the identity of active neurons among the population. This heading representation can be updated by visual and self-motion cues via interactions between specific types of columnar neurons. These cell type-specific analyses have provided support for computational models of heading direction encoding, originally proposed for rodent neurons, where angular information is integrated through recurrent excitation. However, the CX is comprised of four highly interconnected subregions that likely function coordinately, and how information related to heading is represented and further processed in other subregions, especially in downstream areas, remains unexplored (Shiozaki, 2020).

The fan-shaped body (FB), another subregion of the CX, is anatomically considered to be an area downstream of columnar neurons in the EB. A functional connectivity study suggested that columnar neurons in the EB, albeit indirectly, influence the activity of certain FB neurons. In addition, the FB directly receives input from outside the CX and is involved in various aspects of navigation, such as visual memory, locomotor handedness, and processing of optic flow. Therefore, the FB is well positioned to receive heading signals from columnar neurons in the EB and integrate them with other types of signals to guide navigation. However, it remains unclear whether and how neurons in the FB encode information about directional heading in behaving animals (Shiozaki, 2020).

Two-photon calcium imaging in the FB was carried out while flies were flying in visual virtual reality as well as in darkness. Specific types of columnar neurons in the FB show characteristic population dynamics that are prominent during flight but not in quiescence. These dynamics multiplexed information about ongoing turning behavior and heading direction. Activity of these FB neurons was coordinated with that of columnar neurons in the EB, which also encoded turning behavior and heading direction. Despite these similarities, columnar neurons in the FB and EB showed distinct activity in their branches in the noduli (NO), a subregion of the CX, where FB but not EB neurons flipped turn preference depending on the visual environment. These results therefore suggest that the heading direction system of Drosophila is composed of columnar neuron networks spanning the EB and the FB, where heading and angular signals from the EB are combined with information about visual context in FB neurons (Shiozaki, 2020).

This study analyzed neural dynamics in the navigation system of Drosophila by performing cell type-specific calcium imaging during flight. Columnar neurons in the FB and the EB showed coordinated population dynamics that encode ongoing turning behavior and, in short timescales, heading direction. A group of FB neurons flipped the preference for turn direction depending on the visual environment, whereas analogous neurons in the EB showed invariant turn tuning. These data suggest that the heading direction system in Drosophila is composed of multiple interacting circuits that distinctly integrate visual and self-motion information (Shiozaki, 2020).

Silencing of neurons in the FB impairs aspects of navigation, such as memory-guided flight orientation. However, little was known about the information encoded by FB neurons aside from flight-dependent change in visual and baseline calcium signals. The current data show that a population of columnar neurons in the FB encodes flight turning behavior as circular dynamics. The position and velocity of the activity bump in the FB were correlated with a fly's turning behavior. Moreover, the bump position was also correlated with a fly's heading in short timescales. Thus, columnar neurons in the FB multiplex information about two aspects of animal navigation (Shiozaki, 2020).

This population activity in the FB was coordinated with that in the EB, suggesting that they carry similar information. Indeed, columnar neurons in the EB encoded turning behavior and heading as in the FB. Unlike a previous study reporting that the movement of the bump in E-PG neurons is uncorrelated with flight turns in darkness, this study found that turning behavior is correlated with the position and velocity of activity bump in E-PG neurons even in darkness. This apparent discrepancy likely resulted in part from differences in prior experience in closed-loop flight because E-PG neurons did not show turn-related activity in darkness when the flies performed flight with a visual pattern and a bar in advance, as in the previous study. This study also found that the bump position was influenced more by turning behavior than visual cues, unlike in walking flies, where visual but not motor cues dictate the bump position. Together, the activity bump of columnar neurons encodes different types of information depending on experience and behavioral states. On a relevant note, experimental and computational studies have proposed that the association between bump position and visual landmarks is formed through experience-dependent synaptic plasticity. For these proposed models to work, turning behavior must drive the movement of the bump, which was observed in the CX of flying flies as in walking flies (Shiozaki, 2020).

The bump position of columnar neurons in the FB and the EB was correlated with heading only for several seconds during flight with a visual pattern, suggesting that the offset between bump position and heading drifted away over the course of the recordings. Such a strong drift was not observed for E-PG neurons in walking flies. A previous study has shown that the offset can be unstable when the visual scene contains multiple identical objects. Thus, the stronger drift in the current study might stem from the visual pattern, which contained an array of identical bars. Consistent with this idea, the drift was weaker during flight with a single bar. It is likely that heading representations are stable in flies navigating in natural environments where visual scenes contain more features to identify heading than the scenes in the current study (Shiozaki, 2020).

In the FB and EB, the bump position was correlated with turning behavior with a delay of around 1 s. Thus, these turn signals potentially contribute to flight behavior operating on a timescale of seconds. Notably, menotaxis, a behavior that requires columnar neurons in the EB, involves changes in heading on such a timescale. However, it remains to be investigated whether and how the turn signals encoded as the bump position contribute to guiding flight (Shiozaki, 2020).

How are heading circuits organized in the CX? The data show that columnar neurons in the FB and EB coordinately encode heading and turning behavior during flight. It is proposed that this coordination originates in the communication from columnar neurons in the EB to those in the FB for two reasons. First, P-F-R and P-FN neurons have dendrites in the PB, where axons of E-PG neurons reside. Second, E-PG neurons can activate a type of local neuron in the PB, Δ7 neurons, whose activation can influence the activity of P-FN neurons. Therefore, columnar neurons in the FB might inherit heading signals from the EB, where heading is thought to be computed through recurrent connections among columnar neurons (Shiozaki, 2020).

However, information encoded by columnar neurons in the FB and EB are not identical. This study found that P-FN but not P-EN neurons flipped the preference for turning behavior between flight in darkness and with a visual pattern, suggesting that two types of neurons are differentially influenced by visual input. This difference might reflect signals conveyed from visually responsive neurons that have dendrites outside of the CX and axons either in the EB (ring neurons) or in the FB (e.g., ExFl1 neurons). In addition, P-FN and P-EN neurons innervate different compartments in the NO, where axons of different sets of neurons projecting from the lateral accessory lobe terminate. Although the mechanism underlying this difference awaits further investigation, the current results suggest that columnar neurons in the FB likely do not just relay heading signals but integrate them with other inputs. Consistent with this idea, silencing of neurons in the FB and the EB have different effects on navigation (Shiozaki, 2020).

Although P-FN neurons flipped their turn preference in the NO depending on the visual environment, their branches in the FB did not (i.e., the sign of the correlation between turn direction and bump rotation was invariant). This suggests that calcium signals in the NO branches of P-FN neurons are modulated sub-cellularly, as in the EB branches of P-EN neurons (Shiozaki, 2020).

Columnar neurons showed turn-related activity in darkness, indicating that it derives from non-visual cues. One possibility is that the activity represents an efference copy because the CX receives input from putative pre-motor regions, including the lateral accessory lobe and the posterior slope. Alternatively, turn signals might reflect sensory feedback from, for example, mechanosensory neurons in the antennae, which are active during flight. Although the columnar neurons in the EB also encode turns during walking in darkness, its neural underpinnings likely differ from those during flight because the two modes of locomotion involve distinct patterns of neural activation in motor and sensory systems (Shiozaki, 2020).

Activity of P-F-R neurons increased during flight as in other types of CX neurons. This activity might be inherited from neurons in the optic lobe, whose activity increases during flight. Alternatively, this phenomenon might be mediated by octopaminergic neurons innervating the CX because octopamine regulates flight-dependent neural modulation in the optic lobe. Beyond mechanisms, it will be important to determine whether and how flight-dependent neural modulation contributes to sensory processing and behavior (Shiozaki, 2020).

Networks of columnar neurons in the FB are upstream of the motor system; a type of columnar neurons in the FB, PF-LCre neurons, send axons to the lateral accessory lobe, where dendrites of descending neurons reside. Because PF-LCre neurons are considered to be a primary output channel of the CX, the FB likely plays a unique role in modulating locomotion. In fact, suppressing the activity of columnar neurons in the FB influences walking behaviors in Drosophila, and electrical stimulation of the FB modifies walking in cockroaches. Because E-PG neurons are necessary for adopting arbitrary heading relative to a visual landmark but are dispensable for phototaxis, the FB might modulate elementary motor programs based on spatial information computed within the CX (Shiozaki, 2020).

By gathering the knowledge acquired in various insects, recent work proposed a computational model of the CX network that is capable of performing path integration. In this model, E-PG and P-FN neurons have distinct functions; E-PG neurons encode heading, whereas P-FN neurons integrate heading and speed signals over time to calculate the direction and distance an animal has traveled. Although this study has analyzed just one of a few types of P-FN neurons, the results suggest that P-FN neurons encode heading and turning behavior in a manner similar to E-PG neurons. Monitoring the activity of various types of P-FN neurons in flies engaged in behaviors that require path integration would provide data for direct tests of the model (Shiozaki, 2020).

Neurons innervating the dorsal part of the FB promote sleep, during which locomotion is suppressed. In the current study, neurons in this part did not show circular dynamics, suggesting that these sleep-promoting neurons do not encode spatial information in the same way as FB columnar neurons. It is tempting to speculate that the sleep circuit might rather inhibit locomotion by downregulating the activity of FB columnar neurons; for example, through the EB. Similar processing may shape other behaviors requiring the FB, such as memory-guided flight orientatio, courtship memory, protein seeking, nociceptive avoidance, and aggression, because locomotion is a key building block of these behaviors. The FB might mediate the interaction between spatial representations and the internal state to dictate when and in which direction an animal moves (Shiozaki, 2020).

One hallmark of the visual system is a strict retinotopic organization from the periphery toward the central brain, where functional imaging in Drosophila revealed a spatially accurate representation of visual cues in the central complex. This raised the question how, on a circuit level, the topographic features are implemented, as the majority of visual neurons enter the central brain converge in optic glomeruli. This study discovered a spatial segregation of topographic versus nontopographic projections of distinct classes of medullo-tubercular (MeTu) neurons into a specific visual glomerulus, the anterior optic tubercle (AOTU). These parallel channels synapse onto different tubercular-bulbar (TuBu) neurons, which in turn relay visual information onto specific central complex ring neurons in the bulb neuropil. Hence, these results provide the circuit basis for spatially accurate representation of visual information and highlight the AOTU's role as a prominent relay station for spatial information from the retina to the central brain (Timaeus, 2020).

Like various other sensory modalities for which spatial information is critical, neural circuits in the visual system of many animals are organized in a topographic fashion to maintain the neighboring relationship of adjacent pixels detected by photoreceptors in the periphery, along the visual pathways into the central brain. The topographic representation of different kinds of sensory information within the central brain of Drosophila is currently being investigated using molecular genetic tools in combination with cell-type-specific driver lines. Although it is well known that spatially patterned visual stimuli induce coherent activity bumps in the Drosophila CX, the pathway translating peripheral visual information into central activity patterns remains poorly understood (Timaeus, 2020).

This study has shown that medulla inputs to the AOTU fall into two morphological types regarding their arborization patterns: broad innervation versus spatially restricted axon terminals. In both cases, only a single domain within the AOTU is targeted. Although the topographic representation from the lobula neuropil is mostly lost in the broad innervation pattern of converging and intermingling LC projection neurons onto the majority of optic glomeruli, a unique spatial organization could be identified for the output channel from the medulla. Topographic representation of the medulla (at least its dorsal half, where most driver lines used in this study are expressed) is maintained in the SU of the AOTU, which is spatially separated from lobula representation within the AOTU (the LU). Interestingly, a strict topographic correlation only exists between the a-p position of the dendritic fields of MeTu projection neurons in the medulla and their restricted axon termination along the d-v axis within distinct domains of the SU in the AOTU. No such topography exists along the d-v axis in the medulla. These neurons are therefore well suited for filtering out specific visual information (such as landmarks or celestial bodies) for guiding heading decisions during visually guided navigation (Timaeus, 2020).

Based on their morphology, as well as their molecular identity, three principle types of MeTu neurons provide input into the AOTU, with overlapping dendritic fields within the medulla but segregated axon terminals to distinct AOTU (sub-)domains. MeTu-l and -c classes have a similar neuronal morphology with dendrite arborization restricted to a single medulla layer (M6) and spatially narrow axon termination areas in four separate AOTU subdomains (SU-la, -lp, -ca, and -cp), thereby building several pathways arranged in parallel. The nomenclature of the SU subdomain organization differs slightly from previous studies, because it is now based on the expression patterns of different cell surface molecules, which might reflect the functional organization of these structures. Because of this new classification, both lateral and central domains (but not the medial domain) of the SU become further subdivided into anterior (SU-la and SU-cp) and posterior halves (SU-l_p and SU-cp). Nevertheless, it should be noted that the total number of subdomains remains the same in both nomenclatures, with the major difference being the posterior-lateral subdomain ('lp'), which has been attributed to the central domain (SU-cp) in this study, as part of the Connectin-positive central neuropil. Based on the connectome reconstruction of the hemibrain dataset, which reports in a total number of 347 MeTu neurons ('MC61-type'), it is estimated ∼60 MeTu neurons per topographic class (and twice that for MeTu-m cells), assuming an equal innervation of SU subdomains of similar volume. Because 8-12 TuBu neurons were counted from three independent expression lines, a convergence ratio from MeTu to TuBu neurons of about 8:1 to 5:1. Only the organization of the MeTu-l and MeTu-c neurons clearly enables a spatial projection of visual information from the columnar organization in the medulla to the corresponding AOTU domains, which seems well suited to relay topographic information along one spatial axis toward the central brain (Timaeus, 2020).

The borders of the SU compartments are respected by molecularly defined populations of TuBu neurons, thereby defining the next synaptic elements in the parallel pathways toward the bulb neuropil. Although this neuropil with its afferent (TuBu) and efferent (R neurons) channels has been intensively studied in recent years, there still remains a gap in knowledge concerning how precise synaptic connections convey topographic information to the central complex. Four major findings of the TuBu->EB circuit are revealed by study. First, the topographic position of TuBu dendrites in the SU is not translated into a defined position within the bulb but instead exhibits a targeting plasticity within a restricted bulb area. Secondly, although the recent dissection of the AOTU->EB pathways described the bulb as a tripartite structure including both afferent and efferent neurons, this picture can now be refined by highlighting that although this analysis of TuBu-neurons is mainly restricted to only two representative TuBu classes (one in the SU-l_p and the other in the SU-ca domain), both these classes target to areas within the superior bulb (BU_s). More broadly expressed driver lines revealed exclusive TuBu neuron innervation of the BU_s, indicating that additional TuBu classes target to this bulb area. Thus, at least four different classes of TuBu neurons are expected to exclusively innervate the BU_s (TuBu-l_a, TuBu-l_p, TuBu-c_a, and TuBu-c_s), each of them connecting to a different set of output neurons, indicating an even more complex organization of the bulb, in particular the BU_s. Thirdly, TuBu classes project onto dendritic areas of R neuron classes (so called “sectors”) within the bulb, and specific connections are formed between TuBu neurons and R neuron classes. Although it was not possible to identify three R neuron classes within the BU_s, there probably exists a much higher diversity of connections within this small area of the bulb, reaching beyond the scope of this study. For instance, the postsynaptic partners of one subset of TuBu-ca neurons as well as neurons contacted by R2 and R5 dendrites remain to be identified. Additional postsynaptic partners other than R neurons are contacted by TuBu neurons, like contralaterally projecting neurons described in the locust and the bumblebee, which connect the AOTU units of both hemispheres (TuTu neurons) (Timaeus, 2020).

It appears therefore that topography is conserved within the AOTU output neuron projections toward the bulb and ring neurons, which is in good agreement with their physiological responses to visual stimuli, like bright objects. All ring neurons of the same type occupy the same ring layer within the ellipsoid body, raising the question of how topographic information is integrated within central complex neuropils. Interestingly, different MeTu neuron types with similar receptive fields may innervate different AOTU domains and thereby connect to different TuBu neuron populations forming parallel channels that then diverge within the bulb regions, where SU-l_p and SU-ca efferents were found mapping onto separate ring neurons (R4d versus R2). Hence, at least two distinct topographic MeTu channels into the central brain could be defined. Although functional differences between the BUi and BU_s have been described, functional studies have not yet compared the physiological responses of different TuBu classes or the responses of R neurons within the BU_s. Based on the data presented in this study, it would be expected that retinotopic information in the BU_s remains represented in the respective sector that is associated with their TuBu class (Timaeus, 2020).

A morphologically distinct class of MeTu cells is formed by MeTu-m cells. One distinguishing feature in respect to other MeTu cell types is that many cells arborize broadly in their respective AOTU domain. Axon terminals of single MeTu-m neurons invariably spread across the a-p axis of their SU-domain, whereas in the d-v axis they either covered their domain completely or partially-the former case being reminiscent of the afferent organization of LC neurons from the lobula within optic glomeruli in the PVLP regions, whereas the latter case is similarly described for lobula neurons innervating the AOTU's large unit (LU), where the topography of LC10 neurons in the LU has been analyzed, resulting in the distinction of four different LC10-classes. It remains to be seen whether MeTu-m neurons also could be divided into such classes. Those cells innervating the complete SU-m are well suited to form a nontopographic channel to the central brain. Interestingly, although topographic MeTu-l and -c neurons form dendritic fields within a single medulla layer, MeTu-m neurons integrate from three different medulla layers, reminiscent and in fact similar to some lobular LC neuron types, the main afferents of the AOTU large unit, for which a comparable rough topography along the dorsoventral axis has previously been found. Furthermore, only MeTu-m neurons form a collateral arborization in the lobula, indicating that this pathway could directly integrate visual information from both the medulla and lobula. The observation that MeTu-m neurons contact a population of TuBu neurons that projects into the inferior bulb area (Bui) separated from other TuBu neurons further suggests a different role for this pathway. A contralateral inhibition mediated by the Bui has been described, supporting a model in which the SU-m pathway is involved in suppressing ipsilateral stimuli with the expense of reduced spatial resolution (Timaeus, 2020).

Taken together, topographic and nontopographic afferents generate an interesting assembly of adjacent domains within the AOTU, from exclusively topographic medulla input in SU-l and SU-c domains, nontopographic medullar (and potentially also lobular) input in SU-m, and another large area of nontopographic input exclusively from the lobula in the LU. Thus, this study has identified multiple parallel topographic pathways separated from a parallel nontopographic channel (Timaeus, 2020).

This principle visual pathway involving the AOTU as a central relay station between medullar/lobular inputs and the central brain is widely shared among different insect taxa, where homologous structures can be found, e.g., orthopterans, hymenopterans, and beetles. The stimuli conveyed by this 'anterior visual pathway' have been addressed in only a few insect species so far. Most prominently, the AOTU has been associated with celestial orientation using polarized skylight in several species or in chromatic processing. Dorsal rim ommatidia harboring polarization-sensitive photoreceptors for polarized light vision are crucial for the sky-compass orientation and exist in most insects analyzed, such as locusts. However, it remains unknown whether MeTu neurons receive direct or indirect input from modality-specific cell types located in the DRA. In addition, processing of chromatic information was also shown to be accomplished via the AOTU in several insects. This study has now identified inputs to this pathway, by identifying direct connections between MeTu cells and UV-sensitive R7 photoreceptor cells in medulla layer M6 (Timaeus, 2020).

Furthermore, the molecular markers used in this study can serve as future tools to reveal the molecular mechanisms that underlie the formation of the LC-optic glomeruli network across species. Because Drosophila is among the smallest species for which the AOTU has been characterized and is believed to be a behavioral generalist, even more sophisticated architectures of the SU-homologue could exist in other insect taxa. On the anatomical and functional level, optic glomeruli share many features with the synaptic neuropil within the antennal lobe, which led to the postulation that the glomerular organization in the protocerebrum (optic glomeruli) and the deutocerebrum (olfactory glomeruli) are in fact homologous structures. Indeed, molecular characteristics in the PVLP and AOTU were found that resemble the combinatorial code of cell-surface proteins in the olfactory system (e.g. expression patterns of Ten-m, Con, Caps, and Sema1a in both systems). However, future developmental studies of mutant LC and MeTu neurons are needed to test to what extent common mechanisms of glomerular circuit assembly exist in both sensory systems. Although the idea of a serial homology of glomerular organized neural system is far from being resolved, it will be intriguing for further studies to analyze the developmental mechanisms that underlie the circuit formation of these parallel AOTU pathways and optic glomeruli circuits as well as to compare them with known molecular functions during olfactory system maturation (Timaeus, 2020).

The fact cannot be excluded that the SU of the AOTU might consist of additional functional units that so far have not been identified and that neurons were missed in this analysis due to the lack of expression lines to visualize them. Populations of neurons that were classified as a single type might turn out to be different enough (by morphology and/or synapse partners) to justify the establishment of further pathways, and these cell types might have been missed in the single cell labeling experiment, as this method involves random events where scarcer neurons can easily remain unnoticed. In vivo experiments measuring neuronal activity and responses to visual stimuli were beyond the scope of ths study but will be an essential part for understanding the functional features of the circuit. The wealth of genetic tools and their manifold combinations in Drosophila certainly provide capabilities of detailed analyses. As the driver lines used for this study to unravel the components of the visual pathway are publicly available and could be used to measure and manipulate neuronal activity, it is hoped that this study paves the way for future studies of components of this visual circuit (Timaeus, 2020).

The insect central complex (CX) is thought to underlie goal-oriented navigation but its functional organization is not fully understood. This study recorded from genetically-identified CX cell types in Drosophila and presented directional visual, olfactory, and airflow cues known to elicit orienting behavior. A group of neurons targeting the ventral fan-shaped body (ventral P-FNs) was found to be robustly tuned for airflow direction. Ventral P-FNs did not generate a 'map' of airflow direction. Instead, cells in each hemisphere were tuned to 45° ipsilateral, forming a pair of orthogonal bases. Imaging experiments suggest that ventral P-FNs inherit their airflow tuning from neurons that provide input from the lateral accessory lobe (LAL) to the noduli (NO). Silencing ventral P-FNs prevented flies from selecting appropriate corrective turns following changes in airflow direction. These results identify a group of CX neurons that robustly encode airflow direction and are required for proper orientation to this stimulus (Currier, 2020).

Many animals rely on an internal heading representation when navigating in varied environments. How this representation is linked to the sensory cues that define different surroundings is unclear. In the fly brain, heading is represented by 'compass' neurons that innervate a ring-shaped structure known as the ellipsoid body. Each compass neuron receives inputs from 'ring' neurons that are selective for particular visual features; this combination provides an ideal substrate for the extraction of directional information from a visual scene. This study combines two-photon calcium imaging and optogenetics in tethered flying flies with circuit modelling and shows how the correlated activity of compass and visual neurons drives plasticity, which flexibly transforms two-dimensional visual cues into a stable heading representation. This study describes how this plasticity enables the fly to convert a partial heading representation, established from orienting within part of a novel setting, into a complete heading representation. The results provide mechanistic insight into the memory-related computations that are essential for flexible navigation in varied surroundings (Kim, 2019).

Many animals rely on an internal heading representation when navigating in varied environments. How this representation is linked to the sensory cues that define different surroundings is unclear. In the fly brain, heading is represented by 'compass' neurons that innervate a ring-shaped structure known as the ellipsoid body. Each compass neuron receives inputs from 'ring' neurons that are selective for particular visual features; this combination provides an ideal substrate for the extraction of directional information from a visual scene. This study combines two-photon calcium imaging and optogenetics in tethered flying flies with circuit modelling, and shows how the correlated activity of compass and visual neurons drives plasticity, which flexibly transforms two-dimensional visual cues into a stable heading representation. This study also describes how this plasticity enables the fly to convert a partial heading representation, established from orienting within part of a novel setting, into a complete heading representation. These results provide mechanistic insight into the memory-related computations that are essential for flexible navigation in varied surroundings (Kim, 2019).

This study has shown how inhibitory Hebbian plasticity can rapidly transform visual feature information into an attractor-driven internal representation. Angular velocity input to the attractor converts an emerging mapping on the basis of limited views of a scene into a complete and consistent heading representation, a potentially critical function in animal navigation. The induction of inverse maps emphasizes the notable flexibility of the system. A key issue that remains unresolved is the nature of bump dynamics during translation in a two-dimensional environment. Mammalian head-direction cells are unaffected by translation1, but this Drosophila model suggests that the compass circuit tracks the angle between the orientation of the fly and an object in the visual scene without correcting for translation-potentially making it a local compass. However, the plasticity that this study has identified required only a few minutes, and may be even faster under natural conditions when the system can co-opt an existing mapping from ring to compass neurons. In simulations, this timescale prevented nearby objects and transient stimuli-such as neighbouring conspecifics that would not move coherently with the bearing of the fly-from being mapped, but tethered the compass to distant objects that moved coherently with the turns of the fly (Kim, 2019).

The locus of plasticity is likely to be synapses between ring and compass neurons; An accompanying article (Fisher, 2019), presents electrophysiological evidence that is consistent with plasticity altering inhibitory visual inputs to individual compass neurons. At a synaptic and biophysical level, it remains to be seen how the Hebbian mechanism that is proposed in this study relates to, and interacts with, other forms of plasticity such as spike-timing-dependent plasticity, or with plasticity-inducing mechanisms such as nitric oxide signalling in the ellipsoid body, dopaminergic modulation (as seen in the fly mushroom body) or plateau potentials (as seen during remapping of hippocampal place cells) (Kim, 2019).

The results support a model in which plasticity is constantly active to allow rapid adaptation to new settings, enabling the ring attractor to generate a single heading direction even in a complex environment. Such stable sensorimotor representations probably enable animals to overcome transient uncertainties in their surroundings as they pursue diverse behavioural goals (Kim, 2019).

In the Drosophila brain, 'compass' neurons track the orientation of the body and head (the fly's heading) during navigation. In the absence of visual cues, the compass neuron network estimates heading by integrating self-movement signals over time. When a visual cue is present, the estimate of the network is more accurate. Visual inputs to compass neurons are thought to originate from inhibitory neurons called R neurons (also known as ring neurons); the receptive fields of R neurons tile visual space. The axon of each R neuron overlaps with the dendrites of every compass neuron, raising the question of how visual cues are integrated into the compass. Using in vivo whole-cell recordings, this study shows that a visual cue can evoke synaptic inhibition in compass neurons and that R neurons mediate this inhibition. Each compass neuron is inhibited only by specific visual cue positions, indicating that many potential connections from R neurons onto compass neurons are actually weak or silent. It was also shown that the pattern of visually evoked inhibition can reorganize over minutes as the fly explores an altered virtual-reality environment. Using ensemble calcium imaging, it was demonstrated that this reorganization causes persistent changes in the compass coordinate frame. Taken together, these data suggest a model in which correlated pre- and postsynaptic activity triggers associative long-term synaptic depression of visually evoked inhibition in compass neurons. These findings provide evidence for the theoretical proposal that associative plasticity of sensory inputs, when combined with attractor dynamics, can reconcile self-movement information with changing external cues to generate a coherent sense of direction (Fisher, 2019).

The compass neurons in the Drosophila brain exhibit some resemblance to the head-direction cells of the mammalian brain. Visual cues stabilize the tuning preferences of mammalian head-direction cells, and when a visual cue is rotated to a new horizontal position, the preferences of all of the head-direction neurons rotate together. It has been proposed that the mammalian head-direction system represents a ring attractor-a network in which global dynamics exhibit multiple stable states that unfold in a repeated sequence in response to an input. However, it is not known how visual cues anchor the mammalian head-direction system at a mechanistic level. It has been suggested that Hebbian synaptic plasticity of visual inputs enforces the correct mapping between sensory cues and attractor network states (Fisher, 2019).

Similar to mammalian head-direction cells, Drosophila compass neurons (called E-PG neurons) have properties of a ring attractor. Indeed, the dendrites of E-PG neurons are arranged in a ring in the brain. At any point in time, there is one 'bump' of activity in the E-PG ensemble, which rotates as the fly turns. This network receives continuous input from brain regions that track the rotational velocity of the fly via optic flow signals, proprioceptive signals and/or motor efference signals. These rotational velocity inputs push the bump around the circle. Visual cues make the position of the bump more accurate and stable. It is not known whether visual inputs to E-PG neurons are plastic: the offset between the E-PG bump and the visual world is different in different individuals and it can occasionally change unpredictably within an individual; however, network instability alone does not provide evidence for synaptic plasticity (Fisher, 2019).

It is proposed that correlated pre- and postsynaptic activity triggers associative long-term synaptic depression of R-to-E-PG inhibition. This learning rule would explain why visual receptive fields and heading tuning are typically aligned in E-PG neurons. When an individual R neuron is activated by a visual cue, it should push the bump of activity towards the E-PG neurons that it inhibits most weakly. If the full ring attractor network agrees with this outcome, then long-term synaptic depression will occur and those weak R-to-E-PG synapses will become even weaker, further reinforcing this outcome. To ensure network stability, long-term synaptic depression should be balanced by long-term potentiation at R-to-E-PG synapses; the co-existence of long-term synaptic depression and long-term potentiation would also explain why bidirectional changes are found in visual receptive fields after training. These learning rules should produce a doubled pattern of R-to-E-PG synaptic weights after training in a two-cue world, reflecting the twofold symmetry of visuomotor correlations (Fisher, 2019).

The key result of this study-that visual inputs to E-PG neurons are plastic-supports theoretical models that describe how a network can progressively establish a spatial map of the world by incorporating information about consistent sensory cues during exploration. In robotics, this process is called simultaneous localization and mapping. Ther results provide direct experimental evidence for this type of unsupervised learning at the level of synaptic potentials in vivo (Fisher, 2019).

In a simultaneous localization and mapping framework, visual cues are often local, meaning that they can change in size and apparent angle as they are approached; by contrast, this study chose to use visual cues that could not be approached, simplifying the relationship between heading and visual cues. This choice was motivated by the known receptive field properties of R2 or R4d neurons, which seem adapted to detect the position of the Sun (or Moon). Specifically, R2 or R4d neurons have large inhibitory surrounds, meaning that they only respond robustly to isolated visual objects such as the Sun. The Sun is an ideal compass cue because it is effectively at infinity (Fisher, 2019).

It is proposed that plasticity at R-to-E-PG synapses allows the position of the Sun to be flexibly associated with other compass cues, such as the pattern of linearly polarized light in the sky, sky-wide chromatic and intensity gradients, and wind. In other insects, the E-PG network responds to multiple sorts of compass cues, and navigation behaviour can depend on arbitrary learned associations between compass cues. In a companion study, Kim (2019) provides evidence in favour of the idea that plasticity could be used to learn a complex conjunction of visual objects; in the future, to test this idea, it will be interesting to see whether any complex scene can generate a progressively more-stable heading representation (offset) during training. It will also be important to extend the approach that was taken in this study to simulate a more naturalistic virtual world, to study how multiple types of cues influence the behaviour of this network and the organism (Fisher, 2019).

ON and OFF selectivity in visual processing is encoded by parallel pathways that respond to either light increments or decrements. Despite lacking the anatomical features to support split channels, Drosophila larvae effectively perform visually-guided behaviors. To understand principles guiding visual computation in this simple circuit, focus was placed on investigating the physiological properties and behavioral relevance of larval visual interneurons. The ON vs. OFF discrimination in the larval visual circuit emerges through light-elicited cholinergic signaling that depolarizes a cholinergic interneuron (cha-lOLP) and hyperpolarizes a glutamatergic interneuron (glu-lOLP). Genetic studies further indicate that muscarinic acetylcholine receptor (mAchR)/Galphao signaling produces the sign-inversion required for OFF detection in glu-lOLP, the disruption of which strongly impacts both physiological responses of downstream projection neurons and dark-induced pausing behavior. Together, these studies identify the molecular and circuit mechanisms underlying ON vs. OFF discrimination in the Drosophila larval visual system (Qin, 2019).

ON and OFF selectivity, the differential neuronal responses elicited by signal increments or decrements, is an essential component of visual computation and a fundamental property of visual systems across species. Extensive studies of adult Drosophila optic ganglia and vertebrate retinae suggest that the construction principles of ON and OFF selective pathways are shared among visual systems, albeit with circuit-specific implementations. Anatomically, dedicated neuronal pathways for ON vs. OFF responses are key features in visual circuit construction. Specific synaptic contacts are precisely built and maintained in laminar and columnar structures during development to ensure proper segregation of signals for parallel processing. Molecularly, light stimuli elicit opposite responses in ON and OFF pathways through signaling events mediated by differentially expressed neurotransmitter receptors in target neurons postsynaptic to the photoreceptor cells (PRs). This has been clearly demonstrated in the mammalian retina, where light-induced changes in glutamatergic transmission activate ON-bipolar cells via metabotropic glutamate receptor 6 (mGluR6) signaling and inhibit OFF-bipolar cells through the actions of ionotropic AMPA or kainate receptors. In the adult Drosophila visual system, functional imaging indicates that ON vs. OFF selectivity emerges from visual interneurons in the medulla. However, despite recent efforts in transcriptome profiling and genetic analyses, the molecular machinery mediating signal transformation within the ON and OFF pathways has not yet been clearly identified (Qin, 2019).

Unlike the ~6000 PRs in the adult visual system, larval Drosophila eyes consist of only 12 PRs on each side. Larval PRs make synaptic connections with a pair of visual local interneurons (VLNs) and approximately ten visual projection neurons (VPNs) in the larval optic neuropil (LON). VPNs relay signals to higher brain regions that process multiple sensory modalities. Despite this simple anatomy, larvae rely on vision for negative phototaxis, social clustering, and form associative memories based on visual cues. How the larval visual circuit effectively processes information and supports visually guided behaviors is not understood (Qin, 2019).

Recent connectome studies mapped synaptic interactions within the LON in the first instar larval brain, revealing two separate visual pathways using either blue-tuned Rhodopsin 5 (Rh5-PRs) or green-tuned Rhodopsin 6 (Rh6-PRs). Rh5-PRs project to the proximal layer of the LON (LONp) and form direct synaptic connections with all VPNs, whereas Rh6-PRs project to the distal layer of the LON (LONd) and predominantly target one cholinergic (cha-lOLP) and one glutamatergic (glu-lOLP) local interneurons. The two PR pathways then converge at the level of VPNs (Qin, 2019).

These connectome studies also revealed potential functions for cha- and glu-lOLP. The pair of lOLPs, together with one of the VPNs, the pOLP, are the earliest differentiated neurons in the larval optic lobe and are thus collectively known as optic lobe pioneer neurons (OLPs). Besides relaying visual information from Rh6-PRs to downstream VPNs, the lOLPs also form synaptic connections with each other and receive neuromodulatory inputs from serotonergic and octopaminergic neurons, suggesting that they may act as ON and OFF detectors. This proposal is further supported by recent studies on the role of the Rh6-PR/lOLP pathway in larval movement detection and social clustering behaviors. However, it remains unclear how the lOLPs support differential coding for ON and OFF signals without anatomical separation at either the input or output level (Qin, 2019).

This study investigated the lOLPs' physiological properties and determined the molecular machinery underlying their information processing abilities. Functional imaging studies revealed differential physiological responses towards light increments and decrements in cha-lOLP and glu-lOLP, indicating their functions in detecting ON and OFF signals. Furthermore, it was found that light-induced inhibition on glu-lOLP is mediated by mAchR-B/Gαo signaling, which generates the sign inversion required for the OFF response and encodes temporal information between the cholinergic and glutamatergic transmissions received by downstream VPNs. Lastly, genetic manipulations of glu-lOLP strongly modified the physiological responses of VPNs and eliminated dark-induced pausing behaviors. Together, these studies identify specific cellular and molecular pathways that mediate OFF detection in Drosophila larvae, reveal functional interactions among key components of the larval visual system, and establish a circuit mechanism for ON vs. OFF discrimination in this simple circuit (Qin, 2019).

The Drosophila larval visual circuit, with its small number of components and complete wiring diagram, provides a powerful model to study how specific synaptic interactions support visual computation. Built on knowledge obtained from connectome and behavioral analyses, the current physiological and genetic studies revealed unique computational strategies utilized by this simple circuit for processing complex outputs. Specifically, the results indicate that ON vs. OFF discrimination emerges at the level of the lOLPs, a pair of second-order visual interneurons. In addition, the essential role is demonstrated of glu-lOLP, a single glutamatergic interneuron, in meditating OFF detection at both the cellular and behavior levels and identify mAchR-B/Gαo signaling as the molecular machinery regulating its physiological properties (Qin, 2019).

Functional imaging studies using genetically encoded calcium and voltage indicators provide valuable information regarding the physiological properties of synaptic interactions among larval visual interneurons and projection neurons. However, optical recording approaches have certain technical limitations, including the kinetics and sensitivities of the voltage and calcium sensors, as well as the imaging and visual stimulation protocols. In addition, although glu-lOLP displays a biphasic response towards the light stimulation, calcium reductions and increases for only the initial set of physiological characterizations were quantified. Compared to the delayed calcium rise, the light-induced calcium reductions have low amplitudes and high variabilities, possibly due to the half-wave rectification of the intracellular calcium previously described in adult visual interneurons. For the genetic experiments, focus was placed on evaluating the activation of glu-lOLP, which is reflected by the increase of intracellular calcium signals that lead to neurotransmitter release (Qin, 2019).

To process light and dark information in parallel, both mammalian and adult fly visual systems utilize anatomical segregation to reinforce split ON and OFF pathways. In the larval visual circuit, however, almost all VPNs receive direct inputs from both cha-lOLP and glu-lOLP as well as the Rh5-PRs. Therefore, the response signs of the VPNs cannot be predicted by their anatomical connectivity to ON and OFF detectors. Based on the cumulative evidence obtained through genetic, anatomical, and physiological studies, it is proposed that temporal control of inhibition potentially contributes to ON vs. OFF discrimination in larvae. While cha-lOLP displays clear ON selectivity, the OFF selectivity in glu-lOLP is strengthened by the extended suppression of its light response by mAchR-B/Gαo signaling. This temporal control may also produce a window of heightened responsiveness in cha-lOLP and ON-VPNs towards light signals, similar to the case in mammalian sensory systems where the temporal delay of input-evoked inhibition relative to excitation sharpens the tuning to preferred stimuli. Together, the temporal separation between cholinergic and glutamatergic transmission could reinforce the functional segregation in the VPNs and lead to distinct transmissions of ON and OFF signals. Although further functional validations are needed, temporal control of inhibition provides an elegant solution that may be of general use in similar contexts where parallel processing is achieved without anatomically split pathways (Qin, 2019).

The connectome study identified ten larval VPNs which receive both direct and filtered inputs from two types of PRs and transmit visual information to higher brain regions, including four LNvs (PDF-LaNs), five LaN, nc-LaN1, and two pVL09, VPLN, and pOLP17. Based on these studies on LNvs and pOLP, it is expected the functional diversity in VPNs generated by differential expression of neurotransmitter receptors or molecules involved in electric coupling will be observed. Besides basic ON vs. OFF discrimination, VPNs are also involved in a variety of visually guided behaviors. The temporal regulation of their glutamatergic and cholinergic inputs as well as the local computation within the LON are among potential cellular mechanisms that increase the VPNs' capability to process complex visual information. Further physiological and molecular studies of the VPNs and behavioral experiments targeting specific visual tasks are needed to elucidate their specific functions (Qin, 2019).

Besides the similarities observed between larval lOLPs and the visual interneurons in the adult fly visual ganglia, an analogy can be drawn between lOLPs and interneurons in mammalian retinae based on their roles in visual processing. Cha-lOLP and glu-lOLP carry sign-conserving or sign-inverting functions and activate ON- or OFF-VPNs, respectively, performing similar functions as bipolar cells in mammalian retinae. At the same time, lOLPs also provide inhibitory inputs to either ON- or OFF-VPNs and thus exhibit the characteristics of inhibitory amacrine cells. The dual role of lOLPs is the key feature of larval ON and OFF selectivity, which likely evolved to fulfill the need for parallel processing using limited cellular resources (Qin, 2019).

Lastly, these studies reveal signaling pathways shared between mammalian retinae and the larval visual circuit. Although the two systems are constructed using different neurochemicals, Gαo signaling is responsible for producing sign inversion in both glu-lOLP and the ON-bipolar cell. In mGluR6-expressing ON-bipolar cells, light increments trigger Gαo deactivation, the opening of TrpM1 channels, and depolarization. In larval glu-lOLP, how light induces voltage and calcium responses via mAchR-B signaling has yet to be determined. Gαo is known to have functional interactions with a diverse group of signaling molecules including potassium and calcium channels that could directly link the light-elicited physiological changes in glu-lOLP. Genetic and physiological studies in the larval visual circuit will facilitate the discovery of these target molecules and contribute to the mechanistic understanding of visual computation (Qin, 2019).

The ellipsoid body (EB) of the Drosophila central complex mediates sensorimotor integration and action selection for adaptive behaviours. Insights into its physiological function are steadily accumulating, however the developmental origin and genetic specification have remained largely elusive. This study identified two stem cells in the embryonic neuroectoderm as precursor cells of neuronal progeny that establish EB circuits in the adult brain. Genetic tracing of embryonic neuroblasts ppd5 and mosaic analysis with a repressible cell marker identified lineage-related progeny as Pox neuro (Poxn)-expressing EB ring neurons, R1-R4. During embryonic brain development, engrailed function is required for the initial formation of Poxn-expressing ppd5-derived progeny. Postembryonic determination of R1-R4 identity depends on lineage-specific Poxn function that separates neuronal subtypes of ppd5-derived progeny into hemi-lineages with projections either terminating in the EB ring neuropil or the superior protocerebrum (SP). Poxn knockdown in ppd5-derived progeny results in identity transformation of engrailed-expressing hemi-lineages from SP to EB-specific circuits. In contrast, lineage-specific knockdown of engrailed leads to reduced numbers of Poxn-expressing ring neurons. These findings establish neuroblasts ppd5-derived ring neurons as lineage-related sister cells that require engrailed and Poxn function for the proper formation of EB circuitry in the adult central complex of Drosophila (Bridi, 2019).

The Drosophila central complex is a composite of midline neuropils that include the protocerebral bridge, the fan-shaped body, the ellipsoid body (EB), the noduli and the lateral accessory lobes. These neuropils are interconnected in a modular way whereby columnar projection neurons leading to and from the central complex connect all its components that are themselves intersected by tangential layers of neural processes, which together form functional modules, each representing a segment of sensory space. Functional studies have identified specific roles for the central complex in higher motor control, courtship and orientation behaviours, visual memory and place learning, as well as sleep, attention, arousal and decision-making (Bridi, 2019).

In contrast to expanding insights into the physiological role of the central complex in regulating behaviour, its developmental origin and genetic specification has largely remained elusive. Earlier work described a primordial central complex at late larval/early pupal stages, which becomes fully formed by 48 h after puparium formation. Genetic studies have identified several alleles of as-yet unidentified genes, as well as orthodenticle, Pax6/eyeless, Pox neuro (Poxn), tay-bridge, roundabout, Pdm3 and semaphorin as genes involved in normal formation of central complex sub-structures (Bridi, 2019).

This study investigate the origin and formation of EB ring neurons R1-R4 in the developing and adult brain of Drosophila. Bilateral symmetric neuroblasts ppd5 were identified in the embryonic procephalic neuroectoderm as founder cells of neuronal progeny that constitute R1-R4 subtypes of tangential ring neurons in the adult EB. Mutant analysis and targeted genetic manipulations reveal a lineage-specific requirement of engrailed (en) and Poxn activity that determines the number and identity of ppd5-derived progeny and their EB ring-specific connectivity pattern in the adult central complex of Drosophila (Bridi, 2019).

Previous studies suggested the Drosophila EB -- as part of the central complex -- develops from precursor cells that differentiate during larval development and during pupal stages generate the EB neuropil. Lineage analysis demonstrates that at least part of its origin can be traced back to the embryonic procephalic neuroectoderm. This study identified Engrailed-expressing neuroblasts ppd5 as embryonic stem cells that give rise to Poxn-expressing progeny, which ultimately differentiate into EB ring neurons. Genetic tracing with en-Gal4 identified R1-R4 ring neurons, suggesting that embryonic neuroblasts ppd5 are the major source of Poxn-expressing progeny leading to EB ring neurons detected in this study. Based on their position, morphology, gene expression patterns and axonal fasciculation, these findings suggest that ppd5-derived larval lineages correspond to previously described larval lineages variously called 'EB-A1/P1', 'DALv2/3', 'MC1' or 'DM'. It was previously demonstrated that these larval lineages express Poxn and give rise to gamma-amino butyric acid (GABA)-ergic ring neurons in the central complex of the adult brain. It therefore is proposed to (re-) name them according to their embryonic origin (Bridi, 2019).

Subclass-specific Gal4 lines together with Poxn expression identifies these lineage-related, ppd5-derived sister cells as R1-R4 ring neurons. Moreover, brain-specific Poxn-Gal4 mediated labelling identifies ring neurons and their axonal projections covering all layers of the EB neuropil, thus suggesting neuroblasts ppd5 give rise to the majority, if not all, of ring neuron subtypes. The ontogenetic relationship between Engrailed-expressing neuroblasts ppd5 and Poxn-expressing EB ring neurons is affirmed by the fact that en-Gal4 and Poxn-Gal4-targeted RNAi-mediated knockdown of Poxn causes similar EB neuropil-specific phenotypes. Together, these data establish that ppd5-derived progeny are clonal units contributing to the EB ring neuron circuitry in the central complex in Drosophila (Bridi, 2019).

How are these units specified? In both insects and mammals, the patterning and specification of neural lineages is regulated by genetic programs from neurogenesis to neuronal differentiation. This study in Drosophila shows that the development and specification of EB-specific circuit elements is likewise dependent on the lineage-specific activity of developmental regulatory genes. Early formation and maintenance of Poxn-expressing ppd5 lineages requires engrailed function as revealed with a deficiency removing both engrailed orthologues, en and invected. Previous studies showed that, engrailed/invected are required for the specification of neuroblast identity in the developing nervous system, suggesting that engrailed is also required for the specification of ppd5. A later, lineage-specific function of engrailed was found in the specification of ring neuron numbers, onsistent with its transient expression in Poxn+ lineages in the embryonic brain but not at later developmental stages nor in adult ring neurons. engrailed codes for a homeodomain transcription factor mediating the activation and suppression of target genes, regulatory interactions that are required for neural lineage formation and specification in the procephalic neuroectoderm. In contrast, no function for Poxn in embryonic brain development has been reported, suggesting that Poxn is only during later stages of development required for lineage and/or neuronal specification in the central brain (Bridi, 2019).

Indeed, experiments identify a postembryonic requirement of Poxn in the specification of ppd5-derived progeny. Previous studies showed that zygotic mutations of Poxn perturb EB neuropil formation, in that presumptive ring neurons are unable to project their axons across the midline and as a consequence, the EB ring neuropil is not formed. In the present study, en-Gal4-targeted knockdown of Poxn reveals Engrailed-expressing cells that project across the midline and form a ring-like neuropil instead of their normal ipsilateral projections to the SP. Significantly, no ppd5-derived GFP-labelled cells were observed that project ipsilaterally towards the SP, neurons that are normally detectable with en-Gal4 targeted GFP expression in the adult brain. Furthermore, en>Poxn-IR-targeted, EB neuron-like projections do not form a torroidal ring but are rather characterised by a ventral cleft. These en>Poxn-IR cells aberrantly retain Engrailed expression even though their axonal projection and connectivity pattern clearly identify them as ring neurons that are normally devoid of Engrailed but instead express Poxn. Together these data suggest that, based on their morphology, Engrailed expression, axogenesis and ring-specific projection patterns, en>GFP cells normally projecting to the SP have been transformed into EB ring neurons in en>mCD8::GFP,Dcr2,Poxn-IR flies (Bridi, 2019).

The resulting additional ring neurons in en>mCD8::GFP,Dcr2,Poxn-IR flies are accompanied with a ventrally open EB ring neuropil. A comparable phenotype is seen in brains of Poxn(757)>Poxn-IR flies which are characterised by an increased number of Poxn(757)-Gal4-targeted ring neurons, suggesting that increasing numbers of EB ring neurons lead to an arch-like neuropil reminiscent of the arch-like EB seen in the majority of arthropods. In support of this notion, previous work has demonstrated that in vivo amplification of ppd5-derived progenitor cells can lead to fully differentiated supernumerary GABAergic ring neurons that form functional connections often characterised by a ventrally open EB ring neuropil. Together, these data identify differential roles of Poxn activity during neuroblast lineage formation, in that Poxn is required for cell identity determination of ppd5-derived progeny, as well as for the specification of cell numbers and terminal neuronal projections of EB ring neurons (Bridi, 2019).

These Poxn functions in ppd5-derived brain lineages are reminiscent of Poxn activity in the peripheral nervous system (PNS) which mediates the specification of sensory organ precursor (SOP) cell lineages giving rise to external sense organs, the tactile and gustatory bristles, respectively. In these SOP lineages, differential Poxn activity determines progeny fate between chemosensory (gustatory) or mechanosensory (tactile) neuronal identities. Furthermore, SOP lineage-specific Poxn function specifies the number of these neurons and their connectivity pattern. The apparent functional commonalities between Poxn-mediated specification of ppd5 neuroblast-derived lineages in the brain and SOP lineages in the PNS, suggest that evolutionarily-conserved mechanisms underlie the development and specification of clonal units as cellular substrates for neural circuit and sensory organ formation (Bridi, 2019).

The cytoarchitecture of both the insect and mammalian brain are characterised by neural lineages generated during development by repeated asymmetric divisions of neural stem and progenitor cells. These ontogenetic clones are thought to constitute building blocks of the insect and mammalian brain. In support of this notion, lineage-related progeny constitutes sets of circuit elements of the mushroom bodies and antennal lobes in Drosophila. Clonal relationship also characterises the lineage-dependent circuit assembly in the mammalian brain, where stem cell-like radial glia give rise to clonally-related neurons that synapse onto each other, as has been shown for cortical columns and GABAergic interneurons in the neocortex and for striatal compartments of the basal ganglia. The current study in Drosophila shows that a pair of bilateral symmetric, engrailed-expressing embryonic stem cells, neuroblasts ppd5, give rise to R1-R4 subtypes of tangential ring neurons that contribute to the layered EB neuropil. Thus, ppd5 neuroblast lineages constitute complete sets of circuit elements intrinsic to the adult central complex in Drosophila (Bridi, 2019).

It has been suggested that clonal expansion of neural lineages contributed to the evolution of complex brains and behaviours. Key to this hypothetical scenario are ancestral circuit elements in the form of genetically encoded stem cell-derived clonal units, like the ones described in the current study. In such a scenario, lineage-related ancestral circuit elements might have been multiplied and co-opted or diversified during the course of evolution. Multiplication and co-option have been suggested for the evolution of the multiple-loop architecture of the basal ganglia that allows processing of cognitive, emotional and motor information. In line with this hypothesis, quantitative control of the transcription factor Prospero is sufficient to cause clonal expansion of ring-neuron circuitry in Drosophila (Shaw, 2018), which has been implicated in cognitive and motor information processing and resembles extensive correspondences to vertebrate basal ganglia, ranging from comparable developmental genetics to behavioural manifestations and disease-related dysfunctions (Bridi, 2019).

In contrast to multiplication and co-option, the diversification of stem cell lineages can equally contribute to neural circuit evolution. The current results identify differential and tightly regulated spatio-temporal functions of engrailed and Poxn that lead to the differentiation of ppd5 progeny into hemi-lineage specific identities in the adult brain. Loss of engrailed affects the formation of precursors cells, whereas its lineage-specific knockdown affects the number of Poxn expressing ring neurons. Correspondingly, en-Gal4-driven lineage-specific knockdown of Poxn results in an identity transformation of Engrailed-expressing neurons in the adult brain in that they no longer project to the SP, but instead reveal an EB ring-neuron identity. These data indicate a binary switch of hemi-lineage identities as the result of a feed-forward mechanism between engrailed and Poxn. engrailed may activate transcription (directly or indirectly) of Poxn, which in turn represses engrailed to permit differentiation of R1-R4 neurons, thereby regulating the specification of neuronal identities in ppd5 hemi-lineages. This hypothesis is consistent with lineage tracing and MARCM experiments, as well as the transient expression of engrailed in embryonic ppd5 lineages but not in adult EB ring neurons. However, further studies are required to elucidate the nature and extend of these putative regulatory interactions between Engrailed and Poxn (Bridi, 2019).

In summary, these findings establish a causal relationship between a pair of bilateral symmetric embryonic stem cells, neuroblasts ppd5 and the lineage-related assembly of their EB ring neuron progeny as structural units of the central complex in Drosophila. Based on these observations it is proposed that amplification and diversification of ontogenetic clones together with the repurposed use or exaptation of resulting circuitries, is a likely mechanism for the evolution of complex brains and behaviours (Bridi, 2019).

Both the structure and the amount of sleep are important for brain function. Entry into deep, restorative stages of sleep is time dependent; short sleep bouts selectively eliminate these states. Fragmentation-induced cognitive dysfunction is a feature of many common human sleep pathologies. Whether sleep structure is normally regulated independent of the amount of sleep is unknown. This study shows that in Drosophila melanogaster, activation of a subset of serotonergic neurons fragments sleep without major changes in the total amount of sleep, dramatically reducing long episodes that may correspond to deep sleep states. Disruption of sleep structure results in learning deficits that can be rescued by pharmacologically or genetically consolidating sleep. Two reciprocally connected sets of ellipsoid body neurons were identified that form the heart of a serotonin-modulated circuit that controls sleep architecture. Taken together, these findings define a circuit essential for controlling the structure of sleep independent of its amount (Liu, 2019).

This study describes a circuit that regulates of sleep structure without affecting the total amount of sleep. 5HT acts to enhance the response of 5HT7-GAL4+ neurons to basally active excitatory inputs. 5HT-dependent calcium signals are blocked by TTX, while its ability to increase cAMP is not, supporting the existence of these active excitatory inputs to 5HT7-GAL4+ cells. In contrast, VT-GAL4+ cells do not have basally active excitatory inputs. 5HT modulation of the circuit likely occurs primarily via inputs to 5HT7-GAL4+ neurons since the response of VT038828-GAL4 (VT-GAL4)+ neurons is weaker and lower affinity. Whether there are other, perhaps situationally active, inputs to this circuit is currently unknown (Liu, 2019).

Within the ellipsoid body (EB) the circuit is complex. VT-GAL4+ neurons are functionally connected with the 5HT7-GAL4+ group. VT-GAL4+ neurons provide feedback inhibition to a subset of 5HT7-GAL4+ neurons, which enhances fragmentation, likely via output to non-central complex regions. How inhibition of a subset of the 5HT7-GAL4+ cells acts to modulate the behavioral output of the rest of the population is not yet clear, but it is noted that many of the 5HT7-GAL4+ cells are GABAergic. While all the details of the circuit's complex dynamics remain to be discovered, it is clear that this circuit has a profound and selective effect on sleep architecture (Liu, 2019).

The circuit described in this study is modulated by 5HT, a neurochemical known to be important for regulation of behavioral states in many species. While 5HT in mammals is important in a wide variety of contexts, it was controversial for nearly half a century whether it promoted sleep or wakefulness. In Drosophila, 5HT has only been thought to promote sleep. The current data show that upregulation of serotonergic signaling can also induce sleep fragmentation, suggesting that 5HT's role in sleep in flies exhibits a complexity similar to that of its roles in mammals. The genesis of this apparently conserved complexity may be the extensive involvement of 5HT in non-sleep processes. For an animal in the wild, sleep has inherent risks: predation and loss of opportunities for mating or feeding are just a few. Sleep/wake systems in the brain must control arousal state in collaboration with systems that assess competing needs. 5HT, because it is central to so many critical behavioral circuits, is ideally poised to be an integration point for sleep and the general state of the animal. The diverse, circuit-specific, roles in sleep that 5HT exhibits across phyla may be a result of its ubiquity (Liu, 2019).

The role this study has uncovered for 5HT as a regulator of sleep architecture aligns well with this idea. The daily neuronal activity profile reported by Tric-LUC, a calcium sensor that drives luciferase expression in response to neuronal activity, in sleep fragmentation-generating neurons maps to dawn and dusk, when crepuscular organisms such as fruit flies are most active. Fragmentation of sleep at these times would presumably be beneficial since flies would not enter into deep sleep states at times when they should be feeding and mating. Interestingly, the circuit described in this study accomplishes this feat by increasing P(doze), the probability of falling asleep from a wake state, leaving the scaling of P(wake), a parameter associated with dopamine and arousal, free to be modulated by other factors (e.g., danger from predation, appearance of potential mates). The fact that long sleep bouts can be prevented without putting the animal into a hyperaroused state is advantageous, allowing flexible responsiveness to changing conditions. The involvement of P(Doze), a parameter associated with sleep drive, is also congruent with the sleep-promoting role of 5HT in other brain regions (Liu, 2019).

While controlled sleep fragmentation appears to assist in active period behavior, there is also a need for consolidated sleep. In both mammals and Drosophila, sleep has electrophysiologically distinct substrates with progressively higher arousal thresholds that appear in an ordered fashion during a sleep episode. The deeper sleep stages in mammals, REM and slow wave sleep, are strongly associated with maintenance of cognitive function. Fragmentation of sleep, because it truncates sleep episodes before deeper stages are reached, can result in a selective deprivation of deep sleep stages even when total sleep is not changed. In this study, it was demonstrated that decreasing sleep consolidation, without changing the amount of sleep, can disrupt associative learning. These results suggest that in Drosophila, like in mammals, there are time-dependent changes in the depth of sleep that are important for its beneficial effects. This idea is also supported by modeling and analysis of the structure of fly sleep, which indicate that there are time-dependent changes in the probability of sleep-wake transitions consistent with the existence of deep sleep stages that are only accessed in long sleep episodes (Liu, 2019).

Fragmentation of sleep induced by activation of 5HT inputs to the EB also produced an increase in sleep after the activation was terminated. Excess sleep in the recovery day after a perturbation is a hallmark of a homeostatic process. Homeostatic regulation of total sleep time has been previously demonstrated in Drosophila, but the data suggest that there is also homeostatic regulation of sleep quality. In mammals, individual sleep substates have been demonstrated to be homeostatically regulated- selective deprivation of REM or slow wave sleep, in the absence of loss of total sleep time, drive rebound increases of the deprived stage and mechanical sleep fragmentation has been shown to lead to an increase in total sleep. The ability of the EB circuit in Drosophila to selectively modulate sleep structure, without changing the total amount of sleep, has allowed for the first time the selective probing of the cognitive importance of long sleep bouts and deep sleep stages in the fly. The fact that fragmentation triggers rebound sleep implies that these long sleep bouts may also be important for the general health benefits of sleep (Liu, 2019).

Wolff, T. and Rubin, G. M. (2018). J Comp Neurol. PubMed ID: 30084503 Neuroarchitecture of the Drosophila central complex: A catalog of nodulus and asymmetrical body neurons and a revision of the protocerebral bridge catalog.

The central complex, a set of neuropils in the center of the insect brain, plays a crucial role in spatial aspects of sensory integration and motor control. Stereotyped neurons interconnect these neuropils with one another and with accessory structures. Over 5000 Drosophila melanogaster GAL4 lines were screened for expression in two neuropils, the noduli (NO) of the central complex and the asymmetrical body (AB), and multicolor stochastic labelling was used to analyze the morphology, polarity and organization of individual cells in a subset of the GAL4 lines that showed expression in these neuropils. Nine NO and three AB cell types were identified and are described in this study. The morphology of the NO neurons suggests that they receive input primarily in the lateral accessory lobe and send output to each of the six paired noduli. The AB is demonstrated to be a bilateral structure which exhibits asymmetry in size between the left and right bodies. The AB neurons are shown to directly connect the AB to the central complex and accessory neuropils, that they target both the left and right ABs, and that one cell type preferentially innervates the right AB. It is proposed that the AB be considered a central complex neuropil in Drosophila. Finally, highly restricted GAL4 lines are presented for most identified protocerebral bridge, NO and AB cell types. These lines, generated using the split-GAL4 method, will facilitate anatomical studies, behavioral assays, and physiological experiments (Wolff, 2018).

Located at the center of the insect brain, the central complex is a set of highly interconnected neuropils that processes complex, multisensory information from the environment, integrates it with information about the insect's internal state and past experiences, and guides motor outputs that drive appropriate behavioral responses (Wolff, 2018).

One of the most studied roles of the insect central complex is the integration of sensory information, predominantly from visual input. The output of this sensory processing encompasses diverse motor and behavioral responses. In this capacity, the central complex regulates locomotor behaviors such as handedness, turn direction, initiation and termination of walking. The central complex is thought to play a key role in migration, navigation and orientation using input such as celestial cues and displays responses to looming stimuli suggestive of an involvement in generating escape responses in the locust and fly. The central complex has been suggested to contain a ring attractor network that maintains a representation of the fly's heading direction that may be useful for navigation and orientation in visual conditions as well as in darkness. The central complex is also involved in the formation and recall of short- and long-term visual memories, in the processing of olfactory and gustatory inputs and in maintaining information about the fly's satiety state (Wolff, 2018).

Understanding the core principles of operation of the central complex has been greatly enabled by the dissection of behavior at a single neuron level and the neuron-by-neuron assembly of circuits. A comprehensive anatomical atlas and genetic lines that enable manipulation of individual cell types are invaluable tools for this strategy. This study describes the neuronal composition of the NO and the AB, neither of which has been extensively studied in Drosophila. An understanding of the function of the noduli in behavior lags far behind that of the other central complex structures: the protocerebral bridge (PB), fan- shaped body (FB) and ellipsoid body (EB). The only documented roles for the NO in Drosophila are in the time course of walking activity and in influencing handedness during locomotion. The locust neurons that connect the PB, EB and NO and the PB, FB and NO are sensitive to polarized light. Most recently, recordings from optic-flow-sensitive neurons that connect the lateral accessory lobe (LAL) to the NO and other neurons that link the NO to the FB in the bee have demonstrated the NO are involved in path integration. Finally, the fact that this structure appears to be present only in the subclass of winged insects has led to the speculation that the noduli may regulate flight (Wolff, 2018).

Structural conservation of the central complex across insect species is strong, but not absolute. The discussion that follows focuses on the anatomy of the Drosophila neuropils. The PB, FB, EB and NO are midline structures and exhibit a stratified organization. The PB is a handlebar- shaped structure in the posterior dorsal brain. The EB is shaped like a torus and is tilted on its dorso-ventral axis such that its ventral border defines the anterior margin of the central complex. The FB lies between the PB and EB and represents the largest of the four central complex structures. The bilateral noduli, historically called ventral tubercles, are the most ventral neuropil of the central complex and are nestled beneath the FB. There are three pairs of noduli neatly stacked on top of one another from dorsal to ventral; each pair is bisected by the midline. The dorsal nodulus (NO 1) displays some hint of a transverse division whereas the medial (NO2) and ventral (NO3) noduli exhibit longitudinal segmentation. NO2 is divided into dorsal and ventral subdomains (NO2D and NO2V) and NO3 has three subdomains (Wolff, 2018).

The neuropils considered to be components of the central complex have evolved over time. Power's 1943 description of the central complex includes the FB, EB, and 'ventral tubercles', or NO. By the mid-1970s, the modern view of the central complex had emerged: Williams included the PB within the locust central complex, alongside the FB, EB and NO. The asymmetrical body (AB) is a relatively inconspicuous structure located at the midline, adjacent to the ventral FB. It was first described in the fly as a round, almost exclusively right-hemisphere structure. The AB was observed in both hemispheres in just 7.6% of 2,250 brains immunolabeled with an antibody against the Fasciclin II (Fas II) protein, which is expressed in this structure. Flies with bilateral ABs were reported to have disrupted long-term memory, leading to the suggestion that asymmetry of this structure is important for long-term, but not short-term, memory. Although thousands of GAL4 lines that drive expression in small subsets of neurons in the larval and adult fly brains have been examined, the AB is the only reported instance of an asymmetric structure in the adult fly brain (Wolff, 2018).

Elements with likely homology to the Drosophila AB have also been described in the grey flesh fly Neobellieria bullata and the blowfly Calliphora erythrocephala. In both species, these bodies occur bilaterally, and one of the two is consistently smaller and less densely innervated than the other. In addition, the smaller of the two appears fragmented. A previous study identified five GAL4 lines that show asymmetric innervation of the AB. That analysis revealed lines ranging from a strong right hemisphere bias in innervation to those with asymmetric but bilateral expression, with more conspicuous expression in the right AB. This study builds on the previous study by providing a systematic characterization of the neurons that target the AB, and leads to a proposal that the AB be added as the fifth neuropil of the Drosophila central complex (Wolff, 2018).

In this work, a characterization is presented of cell types of the NO and AB, including morphology, presumed polarity and population size. A set of split-GAL4 lines for NO and AB cell types, reagents was also generated and characterized that will greatly facilitate functional studies. In addition, since publication of a description of the neurons that arborize in the PB, this study has gained several new insights into the PB neurons. These include: 1) one new PB neuron family has been identified; 2) a neuron identified by in a previous study has since been found in the GAL4 collection and is characterized in this study using the multicolor flip-out technique (MCFO); and 3) a set of split-GAL4 lines was generated for PB cell types (Wolff, 2018).

The asymmetrical body: the fifth central complex structure The neuropils considered to be constituent components of the central complex have changed over the decades. The central body), included what became known as the ellipsoid body and the fan-shaped body. The noduli were known as the ventral tubercles. More recently, the central complex has been defined as 'a group of modular neuropils across the midline of the insect brain', '...interconnected neuropils and nuclei that populate the midline of the forebrain-midbrain boundary region', and 'a system of interconnected neuropils lying at, or about, the midline of the protocerebrum'. Although the modular architecture of the central complex structures is conspicuous (e.g. the glomeruli of the PB and the trajectory patterns of neurons that project to, from, and within the central complex structures), it is the assigned boundaries that encompass the central complex that seem to be the feature that defines these structures as members of the central complex (Wolff, 2018).

This study has illustrated that the Drosophila AB, which appears to be a structure that is distinct from the FB, meets the criteria outlined above for central complex neuropils: It is a midline neuropil; it falls within the boundaries of the central complex; and it is interconnected (to the FB and SLP) by a network of previously undocumented (with one exception) neurons (Wolff, 2018).

Since the AB meets all the criteria previously used to define neuropils as components of the central complex, it is proposed that the AB be added as a fifth neuropil of the central complex of Drosophila. The AB is not unique to Drosophila. A previous study describe the presence of likely homologous bilateral, asymmetrically sized ABs in N. bullata and C. erythrocephala. That work also identified a tangential FB neuron that bears a resemblance to the SLP-AB-FBl8 neuron described in this study (Wolff, 2018).

It remains to be determined if the AB is more widely represented in other insect orders. The right AB is significantly larger than the left. At a minimum, this difference is likely due to a combination of smaller arbors in the left AB and the lower frequency with which the left AB is targeted: only the ipsilateral-contralateral- projecting form of the SLP-AB neuron, which arborizes in both the left and right ABs, targets the left AB, whereas the ipsilateral and contralateral-projecting forms of the SLP-AB neuron target exclusively the right AB. Thus, the right AB appears to receive a disproportionately larger share of information from the SLP, although the right and left hemispheres appear to be equally represented as sources of input. Notably, this left-right bias is restricted to the AB, as a parallel preference is not shown for the SLP. The availability of genetic lines that target AB-specific cell types will enable experiments aimed at revealing the relevance of this left-right bias (Wolff, 2018).

Three unusual features distinguish the five most commonly seen NO neurons described from other central complex neurons. First, in contrast to the majority of PB neurons described to date, the projections of four of these five NO neurons are ipsilateral. Second, while anatomical features identify distinct input and output neuronal populations in other central complex neuropils, the noduli appear to be sites for receiving primarily input from other neuropils (boutons appear to be the predominant anatomical feature in the noduli in confocal micrographs of NO, PB-FB-NO and PB-EB-NO neurons). Golgi preparations and data from the likely locust equivalent of the PB-FB-NO neuron (the CPU4 neuron), however, indicate these NO arbors are mixed; perhaps the intensity of the dense populations of boutons masks the presence of spines in confocal preparations. Third, although the noduli do receive input from central complex neuropils (e.g. via the PB-FB-NO and PB-EB-NO neurons, from FB tangential neurons, etc.), the majority of direct input for this new set of neurons is provided by just one neuropil, the LAL. Such a restricted thoroughfare of communication is in stark contrast to the PB neurons, for example, which have a much broader and more diverse network of direct communication. The LAL.s-CREc.s-NO3 Pc.b cell type is distinct from the other four common NO neurons in that it delivers contralateral, rather than ipsilateral, input from the LAL and CRE to NO3P. The posterior compartment of NO3 is therefore unique in that it is the only nodulus subcompartment to communicate directly with the contralateral hemisphere. Given that NO3P (and NO3M) also receives ipsilateral terminals from the LAL via LAL.s-CREi.s-NO3P/Mi.b, this subcompartment may act as a limited integration center between the fly's left and right sensory fields (Wolff, 2018).

Physiological data from two neurons in the sweat bee offer insight into a likely role for the NO neurons described in this study. The TN1 and TN2 neurons ('noduli tangential neurons') share a high degree of anatomical homology with the LAL- NO neurons: TN1 and TN2 are ipsilateral neurons with input branches in the lateral central brain and blebbed branches in the noduli. Recordings from these two cells reveal they fire in response to simulated backward and forward flight, respectively, and that the rate of firing is dependent on the stimulus velocity, suggesting these neurons encode speed using optic- flow and can thereby track the distance traveled by the bee. Similar physiological features and path integration functions would not be unexpected for the apparent homologous Drosophila neurons. The LAL is the primary source of input for the NO neurons described in this study and its activity may provide additional insight into the roles of the LAL-NO neurons. It is a large, bilateral neuropil that is highly interconnected with neuropils of the central complex. Functionally, the LAL is considered to be a sensorimotor integration center, based on several lines of evidence in various insect species. For example, in crickets and moths, activity in LAL neurons is associated with walking. In the locust, assorted LAL neurons exhibit changes in activity in response to various aspects of flight, implicating this brain region in flight control. In Drosophila, LAL neurons involved in walking backwards have been documented (Wolff, 2018).

It has been suggested that the noduli are involved in walking and motor control in Drosophila. The neurons implicated in left-right turning bias in locomotion are the PB-FB-NO neurons, which have presumed input (fine terminals) in the PB, and presumed output (boutons) in the FB and NO. It has been speculated that the bias to turn in one direction or the other is influenced by an interplay between the nodulus subdomains that are targeted by the different PB-FB-NO cell types (Wolff, 2018).

Direct communication between the PB-FB-NO neurons and the LAL-NO neurons is not unexpected, as a previous study has shown synaptic contacts between the bumblebee equivalents of these two cells, the CPU4 and TN cells, respectively. Considering the sensorimotor contribution made by the LAL in various types of movement, the LAL-NO neurons described in this study are strong candidates to contribute to the circuits involved in turning (Wolff, 2018).

The catalog of NO neurons described in this study is incomplete. Analyses of other GAL4 lines has identified several large-field FB neurons that also arborize in the noduli, as well as other brain regions that are currently being characterized; some of these neurons are illustrated in the Golgi stains of a previous study. Other cell types with arbors in the noduli that have so far eluded identification. Finally, it seems likely that there would be output neurons from the NO, although such neurons have not been identified. Electron microscopic-level analysis should provide a path to identifying these neurons (Wolff, 2018).

The debate continues to swirl over what constitutes a distinct cell type. Morphology and function have long been accepted as reliable criteria to distinguish cell types. While morphology is a straightforward and easy means of classifying cell types, it can be misleading in that cells that appear identical may have functional differences. For example, a previous study describe clearly distinct physiological roles for two PB neurons that appear to have indistinguishable morphology at the light level (Wolff, 2018).

Morphological features evident with light microscope-level resolution will therefore likely be insufficient to distinguish all cell types, so knowledge of some combination of synaptic connectivity, functional properties and the genetic programs used to specify these attributes will be necessary to fully define cell types. Similar limitations confound the assignment of neuropil boundaries and sub-compartments (Wolff, 2018).

Synaptic density varies considerably across brain regions and this variation has provided landmarks used to define the neuropils of the fly brain. While the boundaries of some structures are unambiguous (e.g., the PB and EB), neuropil margins are not universally so clear-cut, with many neuropils appearing to meld seamlessly with adjacent neuropils. The opportunity to map the domains of arbors within neuropils identifies distinct regions that are not revealed by differences in synaptic density (e.g. wedge and tile domains in the EB). For example, the mushroom body lobes can be divided into a series of non-overlapping compartments with distinct functions by the extent of the arbors of dopaminergic input neurons and mushroom body output neurons. The LAL provides an example of one neuropil that may have functionally distinct subregions. It is a large neuropil with no obvious boundaries revealed by anti-Brp staining, yet the arbors of many neurons that target this neuropil exhibit strong regional preferences. Mapping the domains of these arbors may identify regions that are functionally distinct (Wolff, 2018).

Three major efforts aimed at cataloging all the neurons in the Drosophila brain are in progress. One, typified by this and others' work, characterizes one structure at a time using light microscopy in combination with the generation and analysis of highly specific GAL4 driver lines. The second method is a modern implementation of the Golgi approach of randomly labeling small numbers of neurons in order to describe their morphology. And the third, which is now becoming practical at the required scale, involves reconstruction of neuronal morphology and circuits through analysis of image volumes collected using electron microscopy. It is believed that such light and electron microscopic-level analyses will be highly synergistic. Light microscopy, with genetically marked cells, provides the ability to observe the morphology of hundreds of individual cells of the same cell type in many different individuals, providing insights on stereotypy. However, its dependence on GAL4 drivers means that completeness of coverage cannot be assured. Conversely, electron microscopic analysis, while usually limited to a single sample, not only ensures completeness but also enables visualization and quantification of synaptic connectivity. Moreover, since EM samples do not carry transgenes expressing ectopic membrane proteins that can interfere with development, wiring errors may be less likely. While only electron microscopy is likely to provide the complete wiring diagram of a circuit, light level analysis of genetic driver lines will be needed to provide the critical bridge between circuit maps and the tools required to precisely manipulate the activity of their individual components (Wolff, 2018).

The central complex (CX) is a midline-situated collection of neuropil compartments in the arthropod central brain, implicated in higher-order processes such as goal-directed navigation. This study provides a systematic genetic-neuroanatomical analysis of the ellipsoid body (EB), a compartment which represents a major afferent portal of the Drosophila CX. The neuropil volume of the EB, along with its prominent input compartment, called the bulb, is subdivided into precisely tessellated domains, distinguishable based on intensity of the global marker DN-cadherin. EB tangential elements (so-called ring neurons), most of which are derived from the DALv2 neuroblast lineage, predominantly interconnect the bulb and EB domains in a topographically organized fashion. Using the DN-cadherin domains as a framework, this connectivity was first characterized by Gal4 driver lines expressed in different DALv2 ring neuron (R-neuron) subclasses. 11 subclasses were identified, 6 of which correspond to previously described projection patterns, and 5 novel patterns. These subclasses both spatially (based on EB innervation pattern) and numerically (cell counts) summate to the total EB volume and R-neuron cell number, suggesting that this compilation of R-neuron subclasses approaches completion. EB columnar elements, as well as non-DALv2 derived extrinsic ring neurons (ExR-neurons), were also incorporated into this anatomical framework. Finally, the connectivity between R-neurons and their targets was addressed, using the anterograde trans-synaptic labeling method, trans-Tango (Omoto, 2018).

The central complex (CX) is an evolutionarily conserved, higher-order neuropil in the arthropod brain thought to integrate sensory and motor information to coordinate and maintain locomotor behavior, thus enabling appropriate navigation. Drosophila mutations that produce structural abnormalities in CX neuropils result in flies with deficiencies in walking and flight. More targeted manipulations, such as silencing of specific CX neuron subclasses, compromise vision-based memories associated with spatial orientation and location. Similar themes emerge from anatomical, electrophysiological, and behavioral studies investigating the CX in other insects. In the cockroach CX, for example, single unit activity correlated with changes in locomotor intensity, turning behavior, or heading direction have been identified. In addition, electrical stimulation of CX neurons in the freely walking cockroach has yielded direct evidence linking CX activity to downstream locomotor output. In other insects, such as locust, cricket, monarch butterfly, and dung beetle, neurons in the CX are tuned to celestial visual cues such as the sun or pattern of polarized skylight. These cues provide the stable environmental signals required to accurately derive relative heading information for short or long range navigations (Omoto, 2018).

The CX consists of four neuropil compartments: the upper (CBU) and lower (CBL) halves of the central body (CB), protocerebral bridge (PB), and paired noduli (NO). In Drosophila, the upper and lower halves of the CB are designated as the fan-shaped body (FB) and ellipsoid body (EB), respectively (see General overview of the ellipsoid body (EB): neuronal interactions and compartmentalization). Recently, the asymmetrical body, a paired neuropil located ventral of the FB and adjacent to the NO, has been proposed as a fifth neuropil compartment of the CX. These neuropil compartments are largely formed by two orthogonally arranged neuronal populations: (1) columnar (small-field) neurons which interconnect the CX compartments along the antero-posterior axis; (2) tangential (large-field) neurons which provide input from lateral brain neuropils to the CX. Terminal arborizations of these neurons define distinct vertical columns and horizontal layers that can be visualized by markers for synaptic or cell adhesion proteins that globally label, but exhibit variable density in, the neuropil. Based on Bruchpilot immunostaining, seven layers were identified in the Drosophila CBU (=FB). The CBL (=EB) also exhibits a layered organization. In Drosophila, this compartment undergoes a morphogenetic transformation during pupal development, whereby the lateral ends of the originally bar-shaped EB primordium bend ventrally to adopt a toroidal arrangement. As a result, tangential neurons of the EB display a circular shape, and hence were called 'ring neurons'. Likewise, layers within the EB are annuli, rather than horizontal slabs. Based on labeling with DN-cadherin, this study has defined five distinct annular domains, termed anterior (EBa), inner and outer central (EBic and EBoc), and inner and outer posterior (EBip and EBop) domains (Omoto, 2018).

Clonal studies in Drosophila show that the neuronal architecture of the CX is organized into lineage-based modules, a ground plan that is likely conserved across insects. A lineage refers to the set of sibling neurons derived from an individual neural progenitor called a neuroblast, and the entire central brain is generated from a fixed number of approximately 100 of such neuroblasts. Four lineages (DM1-4) give rise to the large number of columnar neurons of the CX. The great diversity observed among these neurons is achieved via temporal patterning of molecular determinants in dividing progenitors. Lineages giving rise to the tangential neurons of the CX have been characterized morphologically, but have not yet received much attention experimentally. The most notable exception is lineage DALv2/EBa1 (henceforth called DALv2), that generates ring neurons of the EB. Ring neurons project their axons to distinct annular domains of the EB, and typically possess short globular dendrites ('microglomeruli') in the bulb (BU), a neuropil compartment located laterally adjacent to the EB. The BU encompasses three main partitions [anterior (BUa), superior (BUs), and inferior (BUi) bulb] that are associated with different annular domains of the EB. Furthermore, the BUs and BUi appear to be divisible into anterior (aBUs/aBUi) and posterior (pBUs/pBUi) regions. Input to the BU is provided by neurons of two additional lineages, DALcl1 and DALcl2 (also called AOTUv3 and AOTUv4, respectively). As part of the anterior visual pathway, DALcl1/2 form so-called tubercular-bulbar (TuBu) neurons which project from the anterior optic tubercle to the BU, relaying visual information to ring neurons and thereby the CX as a whole. TuBu neurons form two lineally segregated parallel channels, with DALcl1 establishing connections with ring neurons located in the peripheral domain of the EB via the BUs, and DALcl2 with central ring neurons via the BUi (Omoto, 2018).

Detailed functional studies are beginning to shed light on the circuitry involving ring neurons and their TuBu afferents and columnar efferents. Two-photon calcium imaging has revealed a discrete focus of neural activity, or 'bump,' within a population of columnar neurons ('E-PGs') that interconnect the EB, PB, and gall (GA) of the LAL. E-PG neurons encode an internal compass representation via the activity bump, which dynamically tracks the fly's heading. Additional columnar neuron populations that interconnect the PB, EB, and NO, called P-EN neurons, compute the animals' heading by controlling the movement of the bump in the clockwise or counter-clockwise direction. These findings suggest that the EB may operate as a critical hub in the CX, acting as an interface between neurons that transmit and distribute sensory information (TuBu and ring neurons), and circuits that encode and update a representation of heading direction (E-PG and P-EN neurons). In addition, internal state information is likely integrated into the EB network by additional ring neurons subclasses that signal physiological needs such as sleep and hunger drive (Omoto, 2018).

To make further inroads in understanding how the EB circuitry operates, a comprehensive knowledge of ring neurons and their upstream and downstream connectivity is required. Ultimately, a comprehensive analysis of single cells and their synaptic contacts on the light and electron microscopy level will yield complete coverage of the EB wiring diagram, and certainly inform understanding of how EB-related computations are implemented. However, a current description of subclass-specific projection patterns using genetic driver lines provides a framework to posit inter-class neural interactions that can then be tested physiologically and/or behaviorally, and will assist future efforts for such high-resolution anatomical maps. To this end, this study sought to expand on previous works using this genetic-anatomical approach to more thoroughly describe the EB neuropil. Gal4 driver lines that label ring neuron subclasses were screened and subsequently distinguished from each other based on defined criteria. Many drivers label populations corresponding to previously identified ring neuron subclasses, in addition to several, yet uncharacterized populations. The novel subclasses were given new names per the historical nomenclature system. Columnar elements were also incorporated into this anatomical framework. Based on the domain innervation pattern of each line, putative interactions between elements within the EB network are proposed. Finally, ring neuron drivers were subjected to the anterograde trans-synaptic labeling method, trans-Tango. Ring neurons occupying central domains of the EB commonly display homotypic interactions, such that neurons of a given subclass predominantly form synaptic interactions with other neurons in the same subclass. On the other hand, ring neurons occupying the peripheral domains typically display a larger degree of output into the columnar network. This highlights a fundamental difference in the connectivity, and potentially the functions, of ring neurons in different domains (Omoto, 2018).

This work serves to build upon previous anatomical studies by further clarifying the neuronal architecture of the Drosophila EB. Five definitive DN-cadherin domains constituting the EB neuropil provide fiducial landmarks with which neuron classes can be placed into spatial context. Based on this framework, this study reports several novel ring neuron subclasses and proposes potential interactions between ring, columnar, and neuromodulatory neurons in the EB. Lastly, putative postsynaptic partners of R-neurons were experimentally mapped using trans-Tango, revealing insight into how information may be distributed throughout the EB and the rest of the CX. In addition to the neuroanatomical description of different populations, the identification of driver lines enables genetic access to label or manipulate these populations. This provides an entry point for future studies to probe the functional properties of each class and test the interactions proposed herein. The following summarizes the primary findings, speculates on the functional significance of CX wiring principles, and places this study into a developmental-neuroanatomical context with previous works in Drosophila and homologous structures in other insects (Omoto, 2018).

The CX is viewed as a critical hub for goal-directed navigational behavior in insects. Streams of sensory information from different modalities must converge onto this center of sensorimotor integration to guide navigational decisions based on current trajectory, learned information, and motivational state. Central to this notion was the identification of a stable compass representation that tracks the flies heading in the E-PG neuron population. The robustness of this neural correlate of angular orientation, manifested as a single calcium activity 'bump' that moves around the EB, depends on both visual and proprioceptive cues (Seelig, 2015). Heavily relying upon studies in other insect species as a basis for comparison, recent progress has been made toward identifying the neural pathways that transmit sensory information to the Drosophila CX, with visual input being the most well characterized. The fly CX receives visual information via the anterior visual pathway (AVP), a circuit defined by three successive layers. Information is transmitted from the optic lobe medulla to the anterior optic tubercle, from the tubercle to the bulb (BU), and from there to the EB, via medullo-tubercular (MeTu), tuberculo-bulbar (TuBu), and DALv2 ring neurons (R-neurons), respectively. Parallel ensembles of TuBu neurons terminate in a topographically organized fashion onto the microglomerular dendrites of distinct R-neuron subclasses within the BU. Specific computations are implemented across successive layers in this pathway, such as the integration of recent visual history and self-motion, which may inform downstream behavior. Ring neurons transmit processed visual information concerning features and landmarks to the EB, likely as a stable allothetic reference to guide bump dynamics in E-PG neurons. The interaction between tangential elements of the EB and columnar neurons such as E-PG neurons has been suggested in other insects, and confirmed by GFP reconstitution across synaptic partners (GRASP) in Drosophila. Indeed, this study provides further evidence via trans-Tango that R2 neurons, which are tuned to visual features, provide direct presynaptic input to E-PG neurons. The calcium activity bump in E-PG neurons also shift in total darkness, demonstrating the existence of a proprioceptive input channel that can update the heading representation in the EB in the absence of visual input. It is posited that transmission of idiothetic cues to the CX is mediated in part by R1 and/or ExR4 neurons, as their neurite distribution and polarity suggests feedback from the LAL, a proposed motor signaling center (Omoto, 2018).

Conceivably, the information received by different R-neuron subclasses is transmitted to their ring-shaped neurites, and is processed via connections within the same subclass (homotypic interactions) and/or between subclasses (heterotypic interactions), the extent of which depends on the R-neuron subclass in question. As such, the R-neuron system likely displays recurrent connectivity to enable persistent activity required for memory processes, as has been shown for mushroom body circuits that support courtship memory. Indeed, inner ring neurons (likely R3d and R3p), which comprise a critical nucleus of visual working memory, display prominent homotypic interactions. Future work to define the mechanisms underlying intra-subclass interactions and experiments to perturb them, are required to assess the functional significance of these homotypic interactions (Omoto, 2018).

R-neurons, particularly subclasses of which occupy peripheral EB domains, provide input to several different columnar neuron populations. This study provides novel insight into the nature of subclass-specific, input-output communication between the ring and columnar networks. An important avenue of future work will be to elucidate the tuning properties of each R-neuron subclass and determine the contribution of each input to compass representation. Presumably, R-neuron subclasses that provide prominent, direct input to E-PG neurons, such as R2 or R4m, would exhibit the most influence over compass representation (Omoto, 2018).

Circuit flexibility is likely facilitated by neuromodulatory input on a moment-by-moment basis, which may reconfigure information flow through the network and thus the output of the system. Neuromodulation would likely occur at multiple processing stages, as evidenced by the wide-spread neurites of dopaminergic neurons. For example, a single PPM3 neuron, innervates the GA/LAL, BU, and EBoc/op. It is envisaged that neurite-specific signaling and plasticity may regulate distinct processing nodes, akin to what has been demonstrated for dopaminergic neurons that encode protein hunger. Similarly, 5-HT may also influence R-neuron activity as projections from the serotonergic neurons, ExR3 [corresponding to the posterior medial protocerebrum, dorsal cluster (PMPD)], most prominently innervate EBic. The effect of serotonin may be receptor and circuit specific; distinct 5-HT receptor isoforms are differentially expressed in specific R-neuron subclasses (Omoto, 2018).

For clarity, the five EB domains defined by the global marker DN-cadherin should be reconciled with previously used anatomical terminology of the EB. Frontal sections of the EB at different anteroposterior depths shows that DN-cadherin domains are distinct, annular entities. These domains correspond to 'layers' in other insects, and have sometimes been also referred to as layers in Drosophila as well. Therefore, N-cadherin EB domains are synonymous with layers. Each domain is best represented using a 'dorsal standard view': a horizontal section through the EB containing a lengthwise perspective of the EB canal. From this standard view, the N-cadherin domains are also clearly organized along the anteroposterior axis. Three anteroposterior subdivisions of the EB have been referred to as 'shells,' in line with terminology used for the FB. It is proposef that the anterior most shell encapsulates the anterior domain of the EB (EBa), and therefore consists of only one layer. The intermediate shell encapsulates the inner central (EBic) and outer central (EBoc) domains, and consists of two layers. Finally, the posterior shell encapsulates the inner posterior (EBip) and outer posterior (EBop) domains, and consists of two layers. For example, P-EN neurons occupy the EBop domain, which resides in the posterior EB shell (Omoto, 2018).

Previously, four substructures denoted as 'rings' [EBA (Anterior), EBO (Outer), EBC (Center), EBP (Posterior)], were based on anti-disks large (DLG) immunostaining and roughly correspond to the DN-cadherin domains. Like the DN-cadherin domains, each 'ring' was proposed to contain specific R-neuron subclasses. Based on the ring neuron subclasses to comprise each 'ring', it is inferred that EBA corresponds to EBa and EBic in the current classification system. Furthermore, EBO is EBoc, EBC is EBip, and EBP is EBop (Omoto, 2018).

How does the annular domain structure of the Drosophila EB compare to the lower division of the central body (CBL) described for other insects? Similar to the EB, the CBL represents a multilayered neuropil compartment formed by the neurite contributions of tangential and columnar elements. In insects such as locust (Schistocerca gregaria), which will be used as the primary basis for comparison in the following, the kidney bean or sausage-shaped CBL corresponds to the torus-shaped EB in Drosophila. In locusts, the CBL is effectively located ventrally of the upper division of the central body (CBU), whereas the homologous structures in Drosophila (EB and FB, respectively) are arranged in an antero-posterior fashion. This difference is reflective of a 60° anterior tilt of the locust neuraxis, as evidenced by the peduncle, which extends horizontally in flies but is oriented almost vertically in the locust. In the dung beetle (Scarabaeus lamarcki) and monarch butterfly (Danaus plexippus), the CBL are also sausage-shaped, but the neuraxis orientation is like that of Drosophila. Differences in neuraxis orientation influence the comparison between the internal architecture of the locust CBL and fly EB. The locust CBL is subdivided along the dorso-ventral axis into six horizontal layers (although not stacked seamlessly on top of one another). Based on the expression of global markers, the Drosophila EB is divided into toroidal domains (EBa/ic/oc/ip/op). Considering the tilt in neuraxis, it is posited that dorsal strata (layers 1-2) of the locust CBL roughly correspond to more posterior domains (EBip/op) of the fly EB, whereas ventral strata (layers 3-6) correspond to more anterior EB domains (EBa/ic/oc). Corroborating this notion is the fact that fly P-EN neurons innervate EBop, and the locust homologs (called CL2 neurons) innervate dorsal layers of the CBL (Omoto, 2018).

The EB and its domains, as well as other structures of the CX, are established by the neurite contributions of distinct neuronal populations. How is the neuronal diversity and connectivity of the CX developmentally established? The CX, and brain in general, is organized into structural-genetic modules called lineages; a lineage comprises the set of sibling neurons derived from an individual neural progenitor (neuroblasts). Each neuroblast forms a spatially discrete cluster of neurons with shared wiring properties; sibling neurons extend a limited number of fasciculated axon tract(s) and innervate specific brain compartments. Most brain lineages are 'type I' neuroblast lineages, whose neuroblasts undergo a series of asymmetric divisions each of which renews the neuroblast and produces a ganglion mother cell. Columnar neurons of the CX are generated from four type II lineages which are larger and more complex than type I, with neuroblasts first producing a set of intermediate progenitors which in turn, give rise to ganglion mother cells (Omoto, 2018).

While the columnar neurons contributing to the EB are derived from type II lineages, the tangential elements (R-neurons) are largely derived from a single paired type I neuroblast, forming the lineage DALv2 (also called EBa1). Neurons of the DALv2 lineage have been studied in developmental contexts in a number of previous works. Production of secondary neurons by DALv2 begin around 24 h after hatching. According to Kumar (2009), one of the DALv2 hemilineages undergoes apoptotic cell death, implying that the DALv2 R-neurons forming the adult EB represent a single hemilineage. Cursory heat-shock inducible single-cell clonal analysis carried out in the present study suggests that distinct R-neuron subclasses are born during specific time windows and therefore represent sublineages of DALv2 (Figure 4). Thus, clonal induction shortly after the onset of secondary neuroblast proliferation (20-48 h after hatching) yielded exclusively outer R-neurons of the R4m subclass. At increasingly later time points, these types of clones become rare, and disappeared entirely at induction times after 96 h. The converse is the case for inner ring neurons (R3d/m), which could be induced in increasing numbers with later time points of induction. Given that only a fraction of the overall number of R-neuron subclasses was represented among clones analyzed in this study, additional studies are required to settle the exact birth order of different R-neuron subclasses (Omoto, 2018).

The following provides a brief historical account of ring neuron definitions, attempt to resolve discrepancies in the literature when possible, and provide rationale for naming conventions used in this work (Omoto, 2018).

The R-neuron type corresponds to ring neurons of the DALv2 lineage, with four R-neuron subclasses described in an initial study (R1-4). Two other ring neuron types were designated as 'extrinsic ring neurons' (ExR-neurons), based on large projections outside of the EB; in this study, with this feature were pooled into a single type, the ExR-neurons. The first described type of extrinsic R-neuron (the ExR1 subclass) likely corresponds to helicon cells. The second type (the ExR2 subclass), due to its innervation of the caudal EB, ExR2 may correspond to the EBop-innervating PPM3 dopaminergic neuron. The serotonergic neurons that innervate the EB, corresponding to the PMPD neurons, designate in this study as ExR3. Therefore, ExR1-3 are posteriorly localized ExR-neurons, likely deriving from the DM3-6 lineages. Due to its wide arborization and non-DALv2 based origin, ring neurons of lineage BAmv1, with perikarya in the anterior cortex, were designated in this study a fourth type of ExR-neuron (ExR4); the possibility cannot be excluded that ExR2 from a previous study may correspond to ExR4-neurons, as they too innervate the caudal EB. Furthermore, the 'P'-neurons, described a previous study as having ventrally localized cell bodies and also innervate the caudal EB, likely correspond to what this study designates as ExR4-neurons (Omoto, 2018).

Driver line c105 was found in an earlier study to label R1 neurons, due to their centrifugal arborization pattern, inner ring localization, and extension into the posterior layers of the EB. However, c105-positive R1 neurons exhibit ventrally projecting neurites into the LAL and lack BU microglomeruli, in contrast to what was defined as R1 in a previous study. Due to R1 being the predominant designation this R-neuron subclass thereafter, this classification as R1 is retained in the current study (Omoto, 2018).

In more recent studies, the driver 38H02-Gal4 has been described as labeling R4 (or an R4-subset), in several studies. 38H02-Gal4 does in fact label R4m (based on BUa microglomeruli and centripetal EBoc innervation pattern), but also strongly labels R5. Two other drivers, 15B07-Gal4 and 28D01-Gal4, were used to target EB neurons required for visual-thermal associations in place learning, and were described as labeling 'R1 and R4,' or 'R1 alone,' respectively. Anatomical re-assessment of these drivers reveals that 15B07-Gal4 labels R3d, R3p, and R4d, whereas 28D01-Gal4 labels a neuron subclass indicative of R3m (Omoto, 2018).

In summary, the dorsal view of the EB in conjunction with DN-cadherin immunostaining provide criteria to more definitively identify ring neuron subclasses for future studies. The model organism Drosophila offers unique advantages to examine the circuit motifs that support the broadly relevant computations underlying the processes attributed to the CX; (1) the neurons comprising the CX are spatially and numerically confined, (2) genetic access to label, assess connectivity between, or functionally manipulate, specific neuron types within it, and (3) amenability to electro- or optophysiological recordings, oftentimes in the behaving animal. To fully leverage these advantages, this study provides a systematic description of the ring neuron subclasses comprising the EB, genetic tools to access them, and provide insight into their interactions with other neurons of the CX (Omoto, 2018).

Many animals maintain an internal representation of their heading as they move through their surroundings. Such a compass representation was recently discovered in a neural population in the Drosophila melanogaster central complex (see Anatomy suggests a potential circuit mechanism to update a compass representation), a brain region implicated in spatial navigation. This study used two-photon calcium imaging and electrophysiology in head-fixed walking flies to identify a different neural population that conjunctively encodes heading and angular velocity, and is excited selectively by turns in either the clockwise or counterclockwise direction. These mirror-symmetric turn responses combine with the neurons' connectivity to the compass neurons to create an elegant mechanism for updating the fly's heading representation when the animal turns in darkness. This mechanism, which employs recurrent loops with an angular shift, bears a resemblance to those proposed in theoretical models for rodent head direction cells. These results provide a striking example of structure matching function for a broadly relevant computation (Turner-Evans, 2017).

A stable internal representation of heading is fundamental to successful navigation. Neurons that maintain such a representation in darkness have been reported across various species. Several computational models have been proposed to explain how a population representation of heading might be updated using angular velocity signals from different neural populations, but identifying connections between neurons that carry and integrate these disparate signals has been challenging in mammals. This study took advantage of the small size, strong topography and well-described anatomy and cell types of the fly central complex to identify a candidate neuron population, P-ENs, which carry angular velocity signals. Cell-type-specific genetic tools were used to perform electrophysiological recordings from single P-EN neurons and two-photon calcium imaging from entire populations of both P-ENs and the previously described 'compass neurons' (E-PGs) in head-fixed walking flies to demonstrate how these neurons together create an elegant circuit mechanism to update a heading representation when the fly turns in darkness. The circuit motif underlying this mechanism shares some characteristics with past conceptual models of head-direction cell function (Turner-Evans, 2017).

The rate model that was implemented in this study was able to capture the essence of the observed network activity, reproducing physiological activity in response to an input that is specific to one side of the protocerebral bridge, but uniform otherwise. This suggests a level of control over moving the activity bump that is quite simple to implement in neural circuitry. In addition, the model is agnostic to the type of input that is needed to rotate the bump. It does, however, require inputs that are activated when the fly turns, with a strength proportional to the strength of the turn, and that such inputs preferentially innervate one hemisphere to create a mirror-symmetry in the system. This description anatomically matches at least one known cell type: PB_G1/2-9.b-SPSi.s (Wolff, 2015). The model also requires inhibition to maintain a stationary bump and linear velocity integration. The widely arborizing and glutamatergic PB18.s-GxΔ7Gy.b neurons may provide such large-scale inhibition onto the P-EN neurons (Turner-Evans, 2017).

Some discrepancies remain between the proposed model and the experimental evidence presented in this study. The model assumes only one P-EN neuron per protocerebral bridge glomerulus (Wolff, 2015), which puts a strong constraint on the angular velocity integration properties of the circuit. In particular, although the circuit displays linear velocity integration within the typical range of angular velocities, the activity bump gets 'stuck' at individual P-EN neurons for small turns. That is, when the fly turns slowly, the corresponding small inputs to the circuit do not trigger bump movements. No such bump dynamics were observed in the imaging experiments, indicating that other, unexplored factors may help smooth bump movement in the actual circuit. Noise in the circuit, potential gap junctions and dendro-dendritic connections within and between E-PG and P-EN neurons, as well as the activity of other cell types in the circuit, such as the PB_G1-8.s-EBt.b-D/Vgall.b neurons (Wolff, 2015), may all play a role in smoothing bump movement. These factors may also contribute to differences in bump shape and width between the model and experimental data. Further, the model suggests that E-PG activity is directly passed to the P-EN neurons in the protocerebral bridge, possibly with some anatomical offset and modulation through inhibition. Indeed, almost coincident bumps of activity were observed in the bridge for the two cell types. However, while functional connectivity showed a clear connection from the P-EN neurons to the E-PG neurons, the connectivity, electrophysiology, and imaging results suggested that the E-PG to P-EN connection might be more indirect and also recruit inhibition. In the functional connectivity experiments, very strong activation of the E-PG population reliably excited the P-EN neurons, but weaker excitation evoked a variety of responses. Electrophysiological recordings also revealed an unanticipated complexity in the tuning of the P-ENs' membrane potential. Membrane potential tuning curves generally showed a peak at the same heading as the spike rate tuning curves, but also a pronounced trough about 150° distant from that peak. That trough, likely a result of inhibition in the circuit, was not always evident in the spike rate tuning. Finally, in two-color imaging, offsets were observed of up to one glomerulus between the E-PG and P-EN activity on the ipsilateral side of the bridge, and unexpected P-EN activity on the contralateral side, also offset from the E-PG activity. These results were consistent for both color indicator pairings, as well as in experiments involving a second driver line, suggesting that the effects are not merely an artifact of indicator kinetics or co-expression in another population of neurons. It was noted that during slow rotations, when P-EN activity is low and the E-PG bump is weak, these offsets decreased and, depending on the driver line used, also differed between the ipsi- and contralateral side during a turn. These ae taken as indications that the connectivity between the E-PG and P-EN neurons in the protocerebral bridge may be partly indirect. Future studies will address how excitatory and inhibitory connectivity between these populations and others shape the circuit's compass function (Turner-Evans, 2017).

Still uncertain is whether an activity bump can be independently sustained in the P-EN and E-PG populations, or in the left vs. right P-EN subpopulations. The connections from P-ENs to E-PGs may be the substrate that sustains the maintenance of E-PG bump position in the ellipsoid body in the absence of both visual and self-motion cues (Seelig, 2015), as in the current model. The significant reduction seen in E-PG bump amplitude and PVA (population vector average) strength when synaptic transmission from P-ENs was blocked is supportive of such an idea. Whether E-PG input is similarly essential to the maintenance of P-EN bump strength is less clear, but P-EN heading tuning hints at a dependence on E-PG input. On the other hand, appropriate local connections between nearby neurons either in the ellipsoid body or in the protocerebral bridge may allow bumps of activity to be independently sustained in the E-PG neurons. Signs of such internal connections come, for example, from evidence of presynaptic specializations of E-PGs in the ellipsoid body. Bump persistence could also be achieved through long time-scale cellular biophysics. Future experiments and electron microscopy-based circuit reconstruction efforts should provide stronger constraints on the space of possible models, and clarify the functional and behavioral relevance of the actual circuit structure (Turner-Evans, 2017).

For a circuit mechanism in which phase relationships and conjunctive coding are important, calcium imaging may seem an unreliable arbiter of truth. Somatic single cell recordings, on the other hand, can be hard to interpret given the intricate projection patterns of fly neurons and the compartmentalization of information processing that this can produce. However, the results from calcium imaging and electrophysiology experiments in P-EN neurons were found to be in broad agreement. The electrical signature of P-EN responses to angular and forward velocity mirrored seem with calcium imaging in the noduli. The measured width of a single P-EN neuron's receptive field (~60°) was lower than that observed with calcium imaging (~110°), but this may arise from the slow decay kinetics of calcium indicators. One inconsistency between results, however, related to the imaging of neural activity in the protocerebral bridge. Based on imaging in the noduli and electrophysiology, it was expected that turns in one direction would evoke a steady decrease in activity on the other (contralateral) side of the bridge with increasing rotational velocity. Instead, imaging in the bridge showed a mild increase in activity at higher velocities, albeit while preserving the expected asymmetry between the ipsi- and contralateral side. It is hypothesized that this calcium signal might represent synaptic inputs to the P-ENs more than their spiking activity (Turner-Evans, 2017).

Blocking P-EN output using shi^TS had two effects on the E-PG bump: Its amplitude was reduced and its position sometimes changed dramatically during small turns (visible as an increase in variability and low R² for the correlation of changes in heading versus PVA. The bump amplitude decrease in shi^TS flies at high temperature can be readily explained by the reduction in synaptic input to the E-PGs -- indeed, in the firing rate model P-EN input is essential to the maintenance of the E-PG bump. Several factors may explain why the E-PG bump did not completely disappear during this manipulation. First, cell-intrinsic properties of the E-PG neurons may contribute to the persistence of activity in those neurons even in the absence of external input. Second, the shi^TS block may have been incomplete, meaning that there was sufficient P-EN drive even at high temperatures to keep the E-PG bump alive. Third, it the possibility of gap junctions between P-EN and E-PG neurons, which the experiments would not block, cannot be ruled out. Finally, other neuron types may also provide synaptic input to the E-PG population in the ellipsoid body. Some of these possibilities have been suggested in a recent study that used the anatomy of protocerebral bridge neurons to create a spiking model that generates ring attractor dynamics (Kakaria, 2017; Turner-Evans, 2017 and references therein).

Further, the conceptual and firing rate models would imply that if the P-EN to E-PG connections were entirely removed, E-PG activity would be unable to follow the fly's turns. However, the E-PG activity does still track the fly's turns at high temperature, when the P-EN synaptic output should be blocked. This may, once again, be the result of an incomplete block. It is speculated that one reason that the E-PG bump makes large movements across the ellipsoid body even during small turns is that a reduction in bump amplitude destabilizes the compass representation. Thus, fluctuations in the activity of the E-PGs elsewhere in the ellipsoid body may exert a greater influence on the movements of the bump than under normal conditions, when activity in distant E-PGs is likely to be suppressed. Yet another possibility that could explain the bump's movements is raised by a parallel study, which provides further evidence for P-ENs serving a role in angular integration and describes a second subtype of P-EN neurons that likely also influences the position of the E-PG bump (Green, 2017; Turner-Evans, 2017 and references therein).

The coordinated activity of the E-PG population and its control by the P-EN population when the fly turns are strongly evocative of a compass. The animal could, in principle, use such a neural compass to tether its actions to local landmarks or other sensory cues during navigation, and maintain its bearings in the temporary absence of such cues . Consistent with this idea, the PVA computed with E-PG population activity tracks the fly's heading quite accurately even in darkness. However, it is not yet known how downstream circuits read out E-PG population activity. Thus, although the PVA metric is a useful representation of E-PG compass-like activity, whether downstream circuits perform similar computations to extract the fly's heading is unclear. Further, the PVA was derived by combining the strength and angular position of activity in the E-PG population. Although both these features of E-PG activity likely influence downstream neurons, their specific influence on such neurons will depend on the precise connectivity of the circuit, something that a combination of functional connectivity studies and electron microscopy may reveal in time. Although there is considerable evidence across insects suggesting that CX neurons influence action initiation and turning movements , the connection of E-PG and P-EN neurons to the largely unidentified class of CX neurons that drive behavioral decisions is as yet unclear (Turner-Evans, 2017).

This study has focused on the effects of self-motion cues on bump movement, to which end most of the experiments were conducted with flies walking in the dark. However, E-PG activity is strongly influenced by visual cues, as evidenced by the fact that cue jumps can reset the bump position (Seelig, 2015; Kim, 2017). The angular velocity representation of P-EN neurons, by contrast, seemed unaffected by the presence of closed loop visual feedback. Thus, while a circuit mechanism was suggested for updating heading representation in the dark using self-motion signals, it is anticipated that strong sensory inputs, including those from visual cues, control updating in other circumstances. For example, it was previously observed that the ring neurons retinotopically respond to visual cues (Seelig, 2013). As the putative ring neuron axons arborize in the ellipsoid body along with the E-PG dendrites, it may be possible for them to convey visual information to the E-PG neurons, influencing the movement of the bump of activity. Further, it was suggested above that the E-PG to P-EN connection in the protocerebral bridge may be indirect and recruit sources of inhibition. There exist a few classes of bridge interneurons which may serve as intermediaries in E-PG to P-EN connections. Future studies should help clarify their role in the compass network (Turner-Evans, 2017).

Fly E-PG neurons share several characteristics with mammalian head direction cells. Both head direction cells and E-PG neurons maintain one stable bump of activity and both track the animal's heading in darkness, a feature that is well described by appropriately wired ring attractor models. Rodents that are deprived of proprioceptive and motor efference signals, as in passive transport experiments, show impaired heading representation. To update their heading in darkness, head direction cells in rodent thalamic nuclei and post-subiculum are thought to depend on angular velocity input from the vestibular system, mediated by the dorsal tegmental nucleus. Although 75% of neurons in this region were found to encode angular head velocity, only about a third of those did so in the mirror-symmetric, turn-direction-selective fashion of Drosophila P-EN neurons that were describe in this study (Turner-Evans, 2017).

Individual P-EN neurons were deterministic in their left-right mirror-symmetric rotation tuning, but diverse in the range of rotational velocities that their tuning curves spanned. Indeed, the measured bandwidth of individual P-ENs ranged anywhere between 30° and 270°/s. This diversity may reflect the diversity of tuning of the three to four P-EN neurons that innervate each protocerebral bridge glomerulus (estimated from cell body counts. Such a range of sensitivities and bandwidths would permit a more precise tracking of the flies' turns across a wide range of rotational velocities (Turner-Evans, 2017).

The origins of angular velocity responses in P-ENs are as yet unclear, but these responses show a latency relative to the fly's turning movements that are estimated to be ~150 ms, suggesting that they arise from proprioception rather than motor efference. Anatomically, both the two halves of the protocerebral bridge as well as the two noduli are mirror-symmetric structures innervated by a number of neuron types in a lateralized manner, making them likely candidates for receiving such rotation-tuned input. In the cockroach, neurons encoding angular as well as forward velocity have been recorded in the fan-shaped body, a substructure of the central complex that is evolutionarily conserved in flies. Of note, only one of the forty turn responsive neurons in the latter study showed bidirectional modulation, with excitation for turns in the preferred direction and inhibition for turns the other way, a hallmark of the P-EN neurons. These studies, which relied on extracellular recordings and did not identify cell types, found that changes in spike rate regularly preceded locomotor changes instead of tracking them as was found for the fly P-ENs. If it is assumed that neurons of the type recorded in the cockroach also exist in the fly, it is not yet clear whether the P-EN/E-PG compass network that is described in this article exploit advance information about expected changes in angular velocity (Turner-Evans, 2017).

A striking aspect of the fly compass system is its structural symmetry. Mirror symmetry is a prominent feature of the anatomical layout of the protocerebral bridge. The developmental origins of the anatomical positions of central complex neurons have been the focus of numerous studies. However, although the two sides of the protocerebral bridge and the noduli are tuned to rotations in opposite directions, maintaining symmetry at the large scale, the activity of the E-PGs and P-ENs at the scale of bridge glomeruli breaks this symmetry. During a turn, bumps of activity propagate through the left and right sides of the bridge in parallel, in a manner reminiscent of windshield wipers, rather than obeying mirror symmetry. This pattern of activity, together with the connectivity of protocerebral bridge glomeruli and ellipsoid body sectors ensures that the E-PG bump moves smoothly around the ellipsoid body when the fly turns (Turner-Evans, 2017).

More broadly, topographical organization is a striking feature of many sensory circuits, but structure often follows computational function in neural circuits in the central brain as well. The feedforward pathways to and from the Mauthner cell make clear these neurons role in rapid escape behavior, and the parallel delay loops of the barn owl auditory system and the electric fish point to their comparative roles in localizing prey. The anatomical shift of the P-EN neurons with respect to the E-PG neurons provided an immediate clue to a potential structure/function relationship, that of a mechanism for shifting the bump of E-PG activity to update their internal representation of heading. The fact that topography often matches topology in the small fly brain makes the system ideal for the identification of circuit mechanisms underlying complex computations. Only time -- and perhaps large scale circuit reconstruction efforts -- will tell whether such network motifs are also present, but perhaps better hidden, in the more distributed circuits of much larger brains (Turner-Evans, 2017).

The conserved nature of fan-shaped body neuroarchitecture in insects such as Drosophila and the grasshopper makes it likely that there is also a high degree of correspondence among their commissural fascicles. The current study now enables the development of the fan-shaped body to be understood at the level of individually identified commissural fascicles, and so provides the basis for interspecific comparisons of central complex development involving Drosophila, Tenebrio and the sightless dipluran Campodea where similar patterns of axon decussation are found. Equally, mutant analyses in Drosophila may allow the pattern of decussation present in the grasshopper to be understood with greater precision. For example, the topographic decussation of axons at stereotypic locations in both species suggests the presence of choice points across the midbrain similar to that reported for the ventral nerve cord. Although the mechanism has yet to be identified in the brain, a dysregulation of cell surface adhesion/recognition molecules in a graded manner across the midbrain represents one possibility. In the peripheral nervous system of Drosophila mutant for the cell surface molecule fasciclin III, for example, axons switch fascicles to an incorrect branch of the segmental nerve and so project to inappropriate body wall muscles, while in the visual system, relative expression levels of adhesion molecules have been found to regulate the wiring of neurite fascicles. Glia have also been shown to direct neuronal axogenesis in the CNS and midline glia are present in the both the grasshopper and Drosophila brain during commissure formation. In the karussell mutant (mutation affecting β-spectrin), for example, dysregulation of midline glia belonging to the pointed group results in commissural axons of the ventral nerve cord decussating between anterior and posterior fascicles, a neuroarchitecture not found in the wild type (Boyan, 2017).

The central complex comprises an elaborate system of modular neuropils which mediate spatial orientation and sensory-motor integration. The neuroarchitecture of the largest of these modules, the fan-shaped body, is characterized by its stereotypic set of decussating fiber bundles. These are generated during development by axons from four homologous protocerebral lineages which enter the commissural system and subsequently decussate at stereotypic locations across the brain midline. It is not clear how the decussating bundles relate to individual lineages, or if the projection pattern is conserved across species. This study traced the axonal projections from the homologous central complex lineages into the commissural system of the embryonic and larval brains of both the grasshopper and Drosophila. Projections into the primordial commissures of both species are found to be lineage-specific and allow putatively equivalent fascicles to be identified. Comparison of the projection pattern before and after the commencement of axon decussation in both species reveals that equivalent commissural fascicles are involved in generating the columnar neuroarchitecture of the fan-shaped body. Further, the tract-specific columns in both the grasshopper and Drosophila can be shown to contain axons from identical combinations of central complex lineages, suggesting that this columnar neuroarchitecture is also conserved (Boyan, 2017).

In both the grasshopper S. gregaria and Drosophila, the fan-shaped body with its prominent columnar neuroarchitecture (see Wiring of the central complex subserves information processing) comprises the largest module of the adult central complex. In the grasshopper, this columnar neuroarchitecture develops from an initially orthogonal primary axon scaffold during the second half of embryogenesis and is functional at the time of hatching. The neuroarchitecture is generated when subsets of axons from four lineages (termed W, X, Y, Z) in each protocerebral hemisphere innervate the existing commissural system but then decussate from anterior to more posterior lying fascicles at stereotypic locations across the central brain in a process known as 'fascicle switching'. In Drosophila, decussation of axons from four putatively equivalent lineages to those of the grasshopper also occurs , but during the larval to pupal transition, so that the resulting neuroarchitecture is essentially an adult feature. Species comparisons reveal that fascicle switching is present at some stage of development in the central brain of all arthropods and so may be considered a conserved mode of axogenesis (Boyan, 2017).

A major drawback in understanding of axon decussation in the insect brain has been the lack of a systematic identification of the embryonic commissural fascicles involved. In the grasshopper, for example, although a map of all commissures for the adult brain has been available for some time, the embryonic commissures have to date only been superficially allocated into anterior (ac) and posterior (pc) subsets, in keeping with the nomenclature for the early ventral nerve cord. Further, while the pioneers of the w, x, y, and z tracts from each protocerebral hemisphere have been shown to project into the primary commissural fascicle of the brain just after its formation early in embryogenesis and then to fasciculate with its pioneers, the early axons from the W, X, Y, and Z lineages remain within the commissural fascicles they originally pioneer. Later, in growing axons from these same lineages, however, subsequently decussate but the commissural fascicles involved have remained undescribed. This study reconstructed the axon projections from representative lineages of the central complex into the commissural system of the brain at various developmental stages in both the grasshopper and Drosophila. At the same time, the commissural organization itself was analyzed at these stages using the nomenclature applicable to the grasshopperand Drosophila. This analysis leads to the conclusion that in setting up the columnar neuroarchitecture of the fan-shaped body, comparable choices are being made by subsets of axons from equivalent lineages in both the grasshopper and Drosophila, consistent with a conserved wiring plan for this brain region (Boyan, 2017).

In both Drosophila and the grasshopper, the axon scaffold of the embryonic brain comprises an orthogonal system of axonal projections around the stomodeum. Anterior to the stomodeum, this scaffold in Drosophila had earlier been resolved to the level of grouped anterior or posterior commissures, but not individual fascicles. Recent studies, using specific Gal-4 lines, on the other hand, have documented a large number of single projections from larval/pupal neurons of the protocerebrum to the protocerebral bridge and then to the fan-shaped body, ellipsoid body and noduli, many of a commissural nature. Such commissural elements may now be integrated into a plan equivalent to that developed for the grasshopper (Boyan, 2017).

The Drosophila central brain consists of developmental-structural units of macrocircuitry formed by the sibling neurons of single neuroblasts. Lineage guides the connectivity and function of neurons, providing input to the central complex, a collection of neuropil compartments important for visually guided behaviors. The ellipsoid body (EB) is formed largely by the axons of ring (R) neurons, all of which are generated by a single lineage, DALv2. Two further lineages, DALcl1 and DALcl2, produce neurons that connect the anterior optic tubercle, a central brain visual center, with R neurons. Finally, DALcl1/2 receive input from visual projection neurons of the optic lobe medulla, completing a three-legged circuit that is called the anterior visual pathway (AVP). The AVP bears a fundamental resemblance to the sky-compass pathway, a visual navigation circuit described in other insects. DALcl1 and DALcl2 form two parallel channels, establishing connections with R neurons located in the peripheral and central domains of the EB, respectively. Although neurons of both lineages preferentially respond to bright objects, DALcl1 neurons have small ipsilateral, retinotopically ordered receptive fields, whereas DALcl2 neurons share a large excitatory receptive field in the contralateral hemifield. DALcl2 neurons become inhibited when the object enters the ipsilateral hemifield and display an additional excitation after the object leaves the field of view. Thus, the spatial position of a bright feature, such as a celestial body, may be encoded within this pathway (Omoto, 2017).

The anterior visual pathway (AVP) conducts visual information from the medulla of the optic lobe via the anterior optic tubercle (AOTU) and bulb (BU) to the ellipsoid body (EB) of the central complex. This paper analyzes the formation of the AVP from early larval to adult stages. The immature fiber tracts of the AVP, formed by secondary neurons of lineages DALcl1/2 and DALv2, assemble into structurally distinct primordia of the AOTU, BU, and EB within the late larval brain. During the early pupal period (P6-P48) these primordia grow in size and differentiate into the definitive subcompartments of the AOTU, BU, and EB. The primordium of the EB has a complex composition. DALv2 neurons form the anterior EB primordium, which starts out as a bilateral structure, then crosses the midline between P6 and P12, and subsequently bends to adopt the ring shape of the mature EB. Columnar neurons of the central complex, generated by the type II lineages DM1-4, form the posterior EB primordium. Starting out as an integral part of the fan-shaped body (FB) primordium, the posterior EB primordium moves forward and merges with the anterior EB primordium. This paper documents the extension of neuropil glia around the nascent EB and BU and analyzes the relationship of primary and secondary neurons of the AVP lineages (Lovick, 2017).

Ring attractors are a class of recurrent networks hypothesized to underlie the representation of heading direction. Such network structures, schematized as a ring of neurons whose connectivity depends on their heading preferences, can sustain a bump-like activity pattern whose location can be updated by continuous shifts along either turn direction. A population of fly neurons in the ellipsoid body has been shown to represent the animal's heading via bump-like activity dynamics (see Bump attractors and spontaneous pattern formation). This study combined two-photon calcium imaging in head-fixed flying flies with optogenetics to overwrite the existing population representation with an artificial one, which was then maintained by the circuit with naturalistic dynamics. A network with local excitation and global inhibition enforces this unique and persistent heading representation. Ring attractor networks have long been invoked in theoretical work; this study provides physiological evidence of their existence and functional architecture (Kim 2017).

Studies of neural circuits near the sensory periphery have produced deep mechanistic insights into circuit functions. However, it has been more challenging to understand circuit functions in central brain regions dominated by recurrent networks, which often produce complex neural activity patterns. These dynamics play a major role in shaping cognitive functions, such as the maintenance of heading information during navigation. A heading representation must be unique (because an animal can face only one direction at a given time) and persistent (to allow an animal to keep its bearings in darkness), yet must allow updating that matches the magnitude and speed of heading changes expected from the animal's movements. Theoretically, this can be accomplished by ring attractor networks (see, for example, Balanced neural architecture and the idling brain), wherein the position of a localized subset of active neurons in a topological ring represents the animal's heading direction. However, whether the brain uses these hypothesized networks is still unknown. A recent study reported that a population of neurons, called E-PG neurons [to signify their predominantly spiny (and, thus, putatively post-synaptic) projections within the ellipsoid body ('E-') and their predominantly bouton-like projections within the protocerebral bridge ('-P') and the gall ('G')], in the Drosophila melanogaster ellipsoid body (EB) appears to use bump-like neural activity dynamics to represent the animal's heading in visual environments and in darkness. This study establishes essential properties of the network that enables this representation (Kim 2017).

Whether the E-PG population activity bump tracks the fly's heading direction relative to its visual surroundings during tethered flight was determined first. Two-photon imaging with the genetically encoded calcium indicator GCaMP6f was performed to record dendritic calcium activity of the entire E-PG population in the EB while the fly was flying in a virtual-reality LED arena. The azimuthal velocity of the visual scene was proportional to the fly's yaw velocity. As with walking flies, E-PG population activity during flight was organized into a single bump, whether the visual scene contained a single bar or a more complex pattern. The activity bump closely tracked the fly's heading in flight and persisted in darkness. However, unlike in walking, the activity bump seldom tracked the fly's motor actions in darkness, potentially because tethering deprives the fly of normal sensory feedback about its rotational movements from its halteres. Although the location of the activity bump eventually drifted in some flies, the bump's movement was, on average, uncorrelated to the animal's turning movements in darkness. These findings suggest that the representation of heading in the E-PG population has intact, visually driven dynamics as well as persistence, but is largely uncoupled from updating by self-motion cues during tethered flight (Kim 2017).

To test whether the fly's compass network enforces a unique bump within the EB, advantage was taken of the relative persistence of the visually evoked activity bump in darkness, and asked whether this bump could coexist with an 'artificial' bump of activity. Localized optogenetic stimulation was used to create artificial activity bumps in different locations within the E-PG population. Using a transgenic fly line in which E-PG neurons coexpressed CsChrimsonand GCaMP6f, alternating two-photon laser scan lines of excitation (higher laser intensity) and imaging (normal laser intensity) were used to monitor changes in E-PG population dynamics in response to an optogenetically created spot of local activity. By varying the intensity of stimulation light delivered to the target location, bumps were created of increased calcium activity. As the new bump formed, activity at the previous location began to decline and eventually disappeared without significantly perturbing the fly's behavior. When the optogenetic excitation was terminated, the amplitude of the artificially created bump settled at levels typically evoked by sensory stimuli and did not disappear; it either stayed in the induced location for several seconds or slowly drifted away (Kim 2017).

The bump's uniqueness may arise through either recurrent mutual suppression or an indirect mechanism whereby strong bump activity in the EB functionally inhibits feedforward sensory inputs to other E-PG neurons. To discriminate between these alternatives, two locations on the EB ring were simultaneously excited. A reference location was excited at a fixed laser power, and a second, spatially offset location was excited at increasing levels of laser power. The reference bump could always be suppress by increasing laser power at the second location above a certain threshold, consistent with mutual suppression (Kim 2017).

Recurrent suppression can ensure a unique activity bump through a simple winner-take-all (WTA) circuit. However, an animal's representation of its angular orientation should favor more continuous updates based on turning actions. Such gradual, ordered drift to nearby locations would be more consistent with continuous, or ring, attractor models. This study therefore examined changes in the location of an artificially created bump after the stabilization of its peak activity at the 'natural' level. The experiments were performed in darkness to untether the bump from any potentially lingering visual input. If EB dynamics were driven by a WTA network, bumps would be expected to disappear at times and to jump to random distant locations. In contrast, the bump drifted gradually around the EB; this finding suggests that the fly's heading representation is updated through functionally excitatory interactions between neighboring E-PG neurons, consistent with a ring attractor model. These observations together rule out the possibility that network dynamics in darkness result purely from cell-intrinsic mechanisms or slowly decaying visual input. Most important, direct manipulation of E-PG neuron activity changed the network state, which implies that E-PG neurons do not merely mirror dynamics occurring in a different circuit, but are themselves an important component of the ring attractor (Kim 2017).

The next area of focus was the effective connectivity pattern underlying ring attractor dynamics in the E-PG population. A wide range of network structures can, in principle, implement ring attractors. This study focused efforts to a model space between two extreme network architectures that are analytically solvable: (1) a 'global model' based on global cosine-shaped interactions and (ii) a 'local model' based on relatively local excitatory interactions. Under constraints of a fixed bump width of 90° to match physiological observations and an assumption of effectively excitatory visual input without any negative bias, both models could explain the basic properties of bump dynamics, including its uniqueness and its persistence in darkness. The network's response to more artificial conditions, such as abrupt visual stimulus shifts, was therefore probed (Kim 2017).

How the E-PG population responded to unnatural, abrupt visual shifts was examined experimentally first. Depending on the distance of the shift, the E-PG bump either 'flowed' continuously (shorter shift distances) or 'jumped' to the new location (longer shift distances). In simulations, both models predicted a mixture of jump and flow responses, depending on the strength and width of the abruptly shifting visual input. For example, weak wide input induced flows and strong narrow input evoked jumps. However, the jump-flow balance predicted by the two models differed and was more consistent with the local model in several aspects. First, the visual input strength inferred from normal conditions was much weaker than required by the global model for bump jumps. Second, the global model required a much-wider-than-normal range of visual input strengths to explain jumps at multiple distances. Third, using parameters consistent with the rest of the findings, it was possible to reproduce the jump-flow ratio that was observed with the local model but not with the global model (Kim 2017).

To obtain more concrete evidence, model predictions were compared to experimentally observed bump dynamics, under conditions in which input strength, polarity, and shift distance were controlled through optogenetic stimulation. To simulate moderate and large input shift distances, two small regions were sequentially stimulated in the EB-each with an angular width of 22.5°-separated by either 90° or 180°. The stimulation laser power was varied to detect the threshold required for the bump to jump. The laser power required to elicit a jump was not significantly different between the two different shift distances, favoring the local model. The strength of input to the network was then inferred by comparing the amplitude of the optogenetically evoked bump to natural bump amplitudes in darkness. The optogenetic input strength required to induce jumps was smaller than the global model's prediction but matched that of the local model and the range of the inferred visual input strength under normal conditions. Finally, intermediate models that lie between the extremes of the local and global models were then test; any model that exhibited the observed jumps in response to a weak 22.5°-wide input had narrow connectivity profiles was then found. All these observations were once again consistent with the local model (Kim 2017).

In mammals, heading representations are thought to be distributed across multiple neural populations and multiple brain areas. In Drosophila as well, the compass system likely involves multiple cell types, including neurons in the protocerebral bridge (PB). Further, occasional changes observed in the dynamics suggest network modulation by other factors not yet known. For example, sometimes sudden changes were observed in E-PG dynamics, as when the amplitude of the sensory-evoked activity bump changed depending on whether or not the tethered fly was flying and, occasionally, during flight. Nonetheless, the E-PG population provides a powerful physiological handle on the internal representation of heading: a single activity bump moving through topographically arranged neurons. The experimental approach this enabled provides one avenue for investigating which of multiple populations are key circuit components of a computation and which simply read out the results of that computation. It was found that the artificial bump created by directly manipulating E-PG population activity displays natural dynamics, which indicates that these neurons are a key component of the heading circuit (Kim 2017).

The finding that the uniqueness of the E-PG activity bump is ensured via global competition strengthens the conclusion that this population encodes an abstract internal representation of the fly's heading direction. Such abstract representations permit an animal to untether its actions from the grasp of its immediate sensory environment and thereby confer flexibility in both time and behavioral use. Combining an analysis of artificially induced bump dynamics with theoretical modeling allowed interrogation of this recurrent circuit architecture. It was found that the effective network connectivity profile was consistent with ring attractor models characterized by narrow local excitation and flat long-range inhibition. This neural circuit motif of local excitation and long-range inhibition is ubiquitous across many brain areas and across animal taxa. Such observations support the idea that common circuit motifs might be evolutionarily adapted to serve as crucial building blocks of cognitive function (Kim 2017).

Visual projection neurons (VPNs) provide an anatomical connection between early visual processing and higher brain regions. This study characterized lobula columnar (LC) cells, a class of Drosophila VPNs that project to distinct central brain structures called optic glomeruli. This study anatomically describes 22 different LC types and show that, for several types, optogenetic activation in freely moving flies evokes specific behaviors. The activation phenotypes of two LC types closely resemble natural avoidance behaviors triggered by a visual loom. In vivo two-photon calcium imaging reveals that these LC types respond to looming stimuli, while another type does not, but instead responds to the motion of a small object. Activation of LC neurons on only one side of the brain can result in attractive or aversive turning behaviors depending on the cell type. These results indicate that LC neurons convey information on the presence and location of visual features relevant for specific behaviors (Wu, 2016).

This report presents anatomical and functional studies of lobula columnar (LC) cells, prominent visual projection neurons from the lobula to target regions in the central brain called optic glomeruli. Comprehensive anatomical analyses of the dendritic arbors and central brain projections of LC neurons support the notion that these cells encode diverse visual stimuli, distinct for each LC cell type, and convey this information to cell-type specific downstream circuits. Precise genetic tools that target individual LC cell types allowed exploration of the behavioral consequences of optogenetic activation of these cell types. Activating cells of single LC neuron types was often sufficient to evoke a range of coordinated behaviors in freely behaving flies. Using two-photon calcium imaging from head-fixed flies, two LC cell types with activation phenotypes similar to avoidance responses, were shown to selectively encode visual looming, a stimulus that also evokes similar avoidance behaviors, while a third cell type responded strongly to a small moving object. These results suggest that LC cell types encode visual features that are relevant for specific behaviors. Activation of LC cells in only one brain hemisphere can result in either an attractive or repulsive directional turning response, depending on cell type. Thus which LC neuron channel is activated determines the valence of the behavior, whereas comparison across the brain by two such channels of the same type provides information about the location of relevant visual features (Wu, 2016).

Anatomical properties of LC neurons have been previously described both in Drosophila and other Diptera. This work extends these studies by providing a comprehensive description of LC neurons in Drosophila, including the identification of several previously unreported cell types. Further, these anatomical analyses with were combined the generation of highly specific genetic markers (split-GAL4 lines) for each cell type. Each of the 22 LC types described has morphologically distinct dendritic arbors in the lobula with stereotyped arbor stratification, size and shape. As observed in the medulla, where synapse-level connectomics data are available for many cell types, different layer patterns and arbor shapes are likely to reflect differences in synaptic connectivity and neuronal computation. Arbors of LC neurons are found in all lobula strata, though with large differences between layers. Only LC4 (and perhaps LPLC1 and LPLC2) cells are potentially postsynaptic to neurons in the most distal lobula layer, Lo1, while other strata such as Lo4 and Lo5B include processes of more than half of the LC types. The presence of at least some LC dendrites in each lobula layer implies that all of the about 50 different interneuron types that convey visual information from the medulla, and to a lesser extent from the lobula plate, to the lobula, are potentially presynaptic to some LC cells, although a far smaller number is likely presynaptic to any single LC cell type. The predicted differences in the synaptic inputs to different LC cell types also suggest that they will differ in their responses to visual stimuli. Thus, individual LC neuron types are expected to encode specific visual stimuli, while the population of all LC cell types together should signal a wide range of behaviorally relevant visual features (Wu, 2016).

The visual responses of several LC cell types measured using two-photon calcium imaging support the expectation that different types selectively respond to different visual features. The three LC neuron types examined preferentially responded to distinct stimuli, with either a dark looming stimulus (LC6 and LC16) or a small moving object (LC11) evoking the strongest measured responses. LC6 and LC16 showed stronger responses to a dark expanding disc than to related stimuli such as an expanding bright disk or a darkening stimulus that lacks the expanding motion. The reduction in the LC6 and LC16 responses when the edge motion is removed from the stimulus is precisely what is expected of loom-sensitive neurons and is reminiscent of behavioral studies in houseflies showing that darkening contrast combined with edge motion is the most effective stimulus for triggering takeoffs. Consistent with their similar responses in the imaging experiments, LC6 and LC16 have very similar lobula layer patterns while LC11 has a different arbor stratification indicating that LC11 receives inputs from a different set of medulla cell types than LC6 and LC16 (Wu, 2016).

It is likely that the selectivity for visual stimuli observed in LC neuron responses is both a property of the stimulus selectivity of their inputs-some selectivity was seen while imaging in the dendrites of a few LC cell types-and specific computations implemented by individual LC neuron types. In addition, cells post-synaptic to the LC cells may integrate the responses of several individual LC neurons of the same type to provide more robust detection of specific visual features. For example, while LC6 and LC16 cells as populations are strongly excited by dark looming stimuli, it is currently unknown whether individual LC6 and LC16 neurons, which have dendritic extents well below the maximum size of the looming stimuli, and also well below the size known to elicit maximal behavioral responses, show the same response properties. The anatomical data and genetic reagents provide a starting point for the additional functional and ultra-structural studies that will be required to elucidate the circuit mechanisms that produce the response properties of these and other LC cell types (Wu, 2016).

The suggestion that LC cells are feature-responsive neurons has been partly based on the apparent dramatic reduction in retinotopy between LC neuron dendrites, which have a retinotopic arrangement in the lobula, and their axons, which appear to discard this spatial information as they converge onto optic glomeruli, the cell-type specific target regions in the central brain. This study extended previous analyses of LC neuron arbor convergence by directly visualizing multiple single LC cells in a glomerulus in the same fly. These experiments revealed no detectable retinotopy of LC cell processes in most glomeruli even at this cellular level of resolution. It is possible that the responses of individual LC cells carry information about retinotopic position; given the comparatively small size of LC dendrites (the lateral spread of even the largest LC cells covers less than 20% of visual columns) and the retinotopic distribution of these dendrites in the lobula it would be surprising if they did not. Such retinotopic responses could for example be relevant for those LC cell types that appear to have presynaptic sites in the lobula and are thus likely to provide input to retinotopically organized circuits. However, with the caveat that synapse-level connectivity was not examined, for most LCs the available anatomical information appears to support the view that much retinotopic information is discarded at the glomerulus level. Consistent with this anatomical observation, the calcium imaging experiments from single LC cell types revealed visual responses to localized stimuli that could be measured throughout a cross-section of the glomerulus without clear retinotopic arrangement of the responding axons. Because of the columnar nature and apparently restricted visual field of the dendrites of LC neurons, the features computed by individual LC neurons are likely to be well defined in subregions of the eye, with perhaps downstream circuits required to integrate these locally-extracted features, as discussed above for looming. There is currently little insight into how these computations are initiated in the optic glomeruli and this remains an exciting area for future investigation (Wu, 2016).

Unlike the other LC neurons, it was found that LC10, and to a lesser extent LC9, cells retain some retinotopic information in the arrangement of their axon terminals indicating that the loss of retinotopy is not a necessary consequence of axonal convergence onto a glomerular target region. More specifically, it was observed that the order of LC10 axonal terminals in the anterior optic tubercle (AOTu) along the DV axis matches the sequence of AP positions of the corresponding dendrites in the lobula. This organization could facilitate synaptic interactions of LC10 cells corresponding to different azimuthal positions in the visual field with distinct target cells. Consistent with a possible general role of the AOTu in the processing or the relaying of retinotopic information, retinotopic responses have recently been observed in the dendrites of central complex neurons that, mainly based on work in other insects, are thought to be synaptic targets of output neurons of the lateral zone of the AOTu (Wu, 2016).

It was found that, independent of the presence or absence of retinotopy at the glomerulus level, positional information can be extracted from the differential activity of LC cells between the two optic lobes. It was directly demonstrated this capability by genetically restricting optogenetic LC neuron activation to only one optic lobe. This unilateral activation evoked directional turning responses relative to the activated brain side. Thus, LC neuron signaling appears to convey information on both different visual features and their location. This may further extend the similarities to the antennal lobes where differences in odorant receptor neuron activity between the left and right antennal lobes may contribute to odorant tracking (Wu, 2016).

Activation of different types of LC neurons can induce distinct behaviors including jumping, reaching, wing extension, forward walking, backward walking and turning. While specific activation phenotypes have been reported for a variety of cell types and behaviors, many of these studies have focused on command-like neurons thought to orchestrate specific motor programs. By contrast, the activation phenotypes reported in this study result from the optogenetic stimulation of different types of related visual projection neurons. A plausible interpretation of these results is that activation of LC neurons can mimic the presence of the visual features that these neurons normally respond to and thus elicits behavioral responses associated with these fictive stimuli. This possibility is supported by several lines of evidence from studies of LC6 and LC16. First, optogenetic depolarization of each of these cell types evokes a specific behavioral response-backward walking for LC16 and jumping for LC6-that resembles a similar natural avoidance or escape behavior. Second, backward walking and jumping can both also be elicited by presentation of a predator-mimicking visual loom and, third, in calcium imaging experiments both LC16 and LC6 showed a preferential response to a similar looming stimulus compared to a number of related stimuli. Although this study did not explore LC10 response properties, it is noted that LC10-activation phenotypes also show similarities to natural behaviors: movements resembling the directed foreleg extension displayed during activation-evoked reaching occur, for example, during gap-climbing behavior and in aggressive fly-fly interactions (Wu, 2016).

Overall, the LC neuron activation phenotypes that were observed suggest that the encoding of visual information at the level of LC neurons is sufficiently specialized to contribute to distinct behavioral responses in a cell-type dependent fashion. However, patterns of LC neuron activation that produce more refined fictive stimuli than were employed in the current work will be required to fully explore the LC neuron behavioral repertoire. Likewise, more comprehensive physiological studies of the response properties of the LC cell types will be needed (Wu, 2016).

How does LC cell activation evoke specific behavioral responses? In the simplest scenario, LC neuron depolarization could directly activate a single postsynaptic premotor descending interneuron that would then in turn trigger the observed behavior. This appears plausible in some cases: for example, activation of LC4 neurons (called ColA cells in larger flies) might evoke a jumping response via activation of the Giant Fiber (GF) cells, a pair of large descending neurons known to be postsynaptic to ColA and LC4 and which have a known role in escape behavior. For other LC cell types, there is currently no evidence suggesting a direct connection to descending neurons. For example, candidate descending neurons for the LC16 backward walking response, the moon-walker descending interneurons, do not have dendrites in or near the LC16 glomerulus. Responses to diverse visual stimuli, some of which may derive from LC neuron activity, have also been observed in higher order brain centers without direct connections to LC neurons such as the central complex (Wu, 2016).

The activation experiments also provide several indications that the signaling downstream of LC neurons is likely to be more complex; for example, activation of a single LC cell type can give rise to multiple behaviors such as reaching, wing extension and turning for LC10, or backward walking and turning for LC16. Changes of the spatial pattern of LC neuron activation, as in the stochastic labeling experiments of this study, can further modify activation phenotypes. For example, unilateral LC16 activation primarily evokes turning away from the location of LC16 activation, not backward walking, suggesting that the relative differences in LC16 activity between the two eyes can guide the direction of motor output through downstream signaling. Furthermore, several different LC neuron types may contribute to the same or similar behaviors, as suggested by the jumping phenotypes of LC4, LC6, LC15, LPLC1 and LPLC2. Presumably, visual signals and other information downstream of LC neurons are integrated to select appropriate behavioral actions. Such additional processing is also suggested by the cases of neurons with overlapping response properties but distinct activation phenotypes such as LC6 and LC16. It is also noted that some responses to LC neuron activation appear to be context dependent; for example, reduced forward walking was observed for several LC cell types on the platform of the single-fly assay that is much smaller than the arena used in the arena assay (Wu, 2016).

In addition, this study examined onoy the behavior of standing or walking flies and LC neuron signaling might have different consequences depending on the behavioral state. For example, looming stimuli can also elicit avoidance responses in flying flies, but these responses differ from the takeoff or retreat behaviors of walking animals. Therefore, while LC cell activity appears to convey visual information that is specialized for sets of related behavioral responses, LC neurons do not appear to instruct a single behavioral output (Wu, 2016).

The most common activation phenotypes observed in the screen were apparent avoidance responses. Furthermore, in addition to the LC cells examined in this study, other VPNs may also contribute to avoidance behaviors. This predominance of avoidance phenotypes is perhaps not unexpected. Since escape responses have to be fast and reliably executed under many different conditions, neurons that signal features that can evoke escape may be particularly likely to show phenotypes in an activation screen. Given the importance of predator avoidance for fly survival, it appears plausible that a considerable fraction of visual output neurons might be utilized for the detection of visual threats ranging from looming to small objects. Furthermore, it is likely that CsChrimson-mediated depolarization of an entire population of LC neurons is more similar to the pattern of neuronal activity induced by an imminent collision, and thus responses of many individual loom-sensitive neurons, so it is not surprising that the activation screen revealed at least two looming-sensitive neuron types (Wu, 2016).

The escape-inducing neurons that were identifed in this study could provide inputs to different escape response pathways, such as long- and short-mode escape, or act as multiple inputs to the same downstream circuits. Interestingly, neurons with avoidance-like activation phenotypes project to two separate groups of adjacent glomeruli, one in the dorsal Posterior Ventrolateral Protocerebrum (PVLP; LC6, LC16 and also LC15) and one more ventral and medial (LC4, LPLC1 and LPLC2), the latter two with dendrites in the lobula and lobula plate. This spatial organization may facilitate synaptic interactions of functionally related LC neuron types with common downstream pathways for a specific behavior. The second group is close to dendritic branches of the GF, large descending neurons required for short-mode responses in Drosophila and a postsynaptic partner of LC4/ColA and possibly also the two LPLC cell types. LC6 terminals do not overlap with GF dendrites and LC6 cells may play a role in the GF-independent escape pathways that have been proposed in both Drosophila and housefly. Parallel neuronal pathways involved in escape behaviors have been identified or postulated in both vertebrates and invertebrates, but a contribution of several identified visual projection neurons to such pathways, as suggested by the activation screen, has not been previously reported. Different visual output neurons with distinct tuning of their response properties to looming parameters such as speed, size, luminance change or edge detection might have evolved to ensure robust responses to avoid predators or collisions. It is, however, currently not known whether LPLC1, LPLC2, LC4 and LC15 are indeed sensitive to looming stimuli and if so, whether their response details differ from LC16, LC6 and each other. Nevertheless, the identification of these neurons opens the possibility to examine the potential contribution of several visual pathways to avoidance behaviors (Wu, 2016).

LC neurons are a subset of the about a hundred VPN cell types that relay the output of optic lobe circuits to targets in the central brain. These data strongly support existing proposals for LC cell types as feature-detecting neurons, which have been mainly based on the distinct anatomical properties of LC cells. While these anatomical features distinguish LC neurons from many other VPNs, an association of VPN pathways with specific behaviors is not unique to LC cell types. The notion that individual neuronal pathways are tuned for specific behavioral requirements is a prominent theme in invertebrate neuroethology, with these neurons described as 'matched filters' for behaviorally relevant features of the external world. A number of previously studied VPN pathways, outside of the LC subgroup, have been described as encoding specific behaviorally related visual stimuli. In particular, very similar to the current results for LC6 and LC16, a group of tangential cells of the lobula and lobula plate (Foma-1 neurons) were found to respond to looming visual stimuli and, upon optogenetic activation, trigger escape responses. And perhaps most famously, the long-studied lobula plate tangential cells (LPTCs), such as the HS and VS cells, integrate local motion signals so as to preferentially respond to global optic flow patterns that are remarkably similar to visual motion encountered during specific behavioral movements. These findings are consistent with the idea that, at the outputs of the fly visual system, VPN pathways are found whose encoding properties are already well matched to particular fly behaviors or groups of behaviors. Matching the response properties of these deep sensory circuits to behavioral needs may be a general evolutionary solution to the challenge of dealing with the complexity of the visual world with limited resources (Wu, 2016).

LC neurons have long been recognized as a potential entry point for the circuit-level study of visual responses outside of the canonical motion detection pathways. This study has provided a comprehensive anatomical description of LC cell types and genetic reagents to facilitate such further investigations. It was also shown that activation of several LC cell types results in avoidance behaviors and that some of these same LC types respond to stimuli that can elicit such behaviors. Other LC neurons appear to mediate attractive behavioral responses. This work provides a starting point for exploring the circuit mechanisms both upstream and downstream of LC neurons (Wu, 2016).

Prolonged wakefulness leads to an increased pressure for sleep, but how this homeostatic drive is generated and subsequently persists is unclear. From a neural circuit screen in Drosophila, this study identified a subset of ellipsoid body (EB) neurons whose activation generates sleep drive. Patch-clamp analysis indicates these EB neurons are highly sensitive to sleep loss, switching from spiking to burst-firing modes. Functional imaging and translational profiling experiments reveal that elevated sleep need triggers reversible increases in cytosolic Ca(2+) levels, NMDA receptor expression, and structural markers of synaptic strength, suggesting these EB neurons undergo 'sleep-need'-dependent plasticity. Strikingly, the synaptic plasticity of these EB neurons is both necessary and sufficient for generating sleep drive, indicating that sleep pressure is encoded by plastic changes within this circuit. These studies define an integrator circuit for sleep homeostasis and provide a mechanism explaining the generation and persistence of sleep drive (Liu, 2016).

Many animals navigate using a combination of visual landmarks and path integration. In mammalian brains, head direction cells integrate these two streams of information by representing an animal's heading relative to landmarks, yet maintaining their directional tuning in darkness based on self-motion cues. This study used two-photon calcium imaging in head-fixed Drosophila melanogaster walking on a ball in a virtual reality arena to demonstrate that landmark-based orientation and angular path integration are combined in the population responses of neurons whose dendrites tile the ellipsoid body, a toroidal structure in the centre of the fly brain. The neural population encodes the fly's azimuth relative to its environment, tracking visual landmarks when available and relying on self-motion cues in darkness. When both visual and self-motion cues are absent, a representation of the animal's orientation is maintained in this network through persistent activity, a potential substrate for short-term memory. Several features of the population dynamics of these neurons and their circular anatomical arrangement are suggestive of ring attractors, network structures that have been proposed to support the function of navigational brain circuits (Seelig, 2015).

Visual landmarks can provide animals with a reliable indicator of their whereabouts. In the absence of such cues, many animals track their position relative to a reference point by continuously monitoring their own motion, a process called path integration. Estimates of position based purely on self-motion cues, however, can accumulate error over time. Successful navigation then, requires animals to flexibly combine these distinct sources of information. In mammalian brains this process of integration is evident in head direction cells, which are neurons sensitive to an animal's heading relative to visual cues in its surroundings that maintain their representation of heading in total darkness using self-motion cues. With their smaller brains and identifiable neurons, insects offer tractable systems to examine the integrative neural computations underlying navigation. Indeed, many insects (for example, desert ants and honeybees) are known to navigate using landmarks and path integration. Experiments in a variety of insects indicate the involvement of the central complex (CX)-a brain region conserved across insects-in such behaviour. In the fruitfly, behavioural genetics experiments have suggested that the CX is required for several components of navigation, including memory for visual landmarks, patterns and places, and directional motor control. Electrophysiological recordings in immobilized locusts and butterflies have revealed a map-like representation for the orientation of electric field vectors of polarized light, which may enable sun-compass navigation. Extracellular recordings from CX neurons in tethered walking cockroaches have shown encodings of turning direction and of wide-field optic flow, a potential cue for self-motion. However, previous studies of visual responses in the CX were conducted under conditions in which insects passively viewed visual stimuli. This study sought to uncover integrative neural processes relevant to navigation in the CX by allowing a tethered fly to control and respond to visual stimuli while simultaneously recording its neural activity and behaviour (Seelig, 2015).

Two-photon imaging was used with the genetically encoded calcium indicator GCaMP6f to monitor neural responses in the CX while a head-fixed fly walked on an air-supported ball within a light-emitting-diode (LED) arena. In previous experiments, a subset of neurons was identified with projections to the CX, and specifically to rings of the ellipsoid body (EB), that show strong tuning to localized visual features including vertical stripes, a class of stimuli that also induce innate fixational responses in flies. To probe how such visual information might be used within the CX this study focused on a class of columnar neurons of the CX, each of which sends dendrites to a specific wedge of the EB. These neurons are termed EBw.s neurons. The dendritic responses of the entire EBw.s population were monitored in the EB during walking, both under closed-loop virtual reality conditions in which the rotation of visual patterns was driven by the fly's turning movement on the ball and in darkness (Seelig, 2015).

This network was found to use information from both landmark-based and angular path integration systems to create a compass-like representation of the animal's orientation in the environment. Previous studies have described static visual maps in the CX. Such maps may allow navigating insects to maintain a sun-compass-based heading direction. This study found that EBw.s neurons track the fly's orientation relative to visual landmarks in a variety of different visual environments, suggesting that the CX dynamically adapts to estimate the fly's orientation within its visual surroundings. Subsets of ring neurons are likely to bring information about spatially localized visual features to specific rings of the ellipsoid body. It is not yet clear how this information is converted into an abstract and flexible representation of the animal's orientation relative to landmarks, but EBw.s responses in a symmetric environment with two indistinguishable cues hint at an underlying winner-take-all process for landmark selection. Combining landmark orientation with information about the animal's movement effectively creates an internal reference frame for the animal in its surroundings. Many of the proposed functions of the CX in directed locomotion, visual place learning, and action-selection, may rely on this internal reference. Although the EBw.s population tracks the fly's rotational movements in darkness, it is not yet known where and how translational motion, an important component of a complete navigational system, is incorporated. Additionally, although the calcium sensor that was chosen for imaging experiments has the temporal resolution necessary to capture EBw.s representations of the fly's angular rotation, it lacks the precision necessary for determination of whether EBw.s activity represents the fly's predicted future orientation or its estimate of current orientation (Seelig, 2015).

The observation that EBw.s activity was maintained in the absence of self-motion suggests that internal dynamics play a significant role in shaping neural activity in the fly brain, much as they do in the brains of larger animals. Persistent activity in the CX can maintain compass information when the fly is standing in darkness for 30 s - two orders of magnitude longer than might be explained by calcium sensor decay kinetics. Persistent activity has been shown to support maintenance of eye position in the goldfish and has been proposed to underlie working memory in mammals. In the CX, this activity may allow the fly to retain a short-term orientation memory even when landmarks are temporarily out of sight. Consistent with this notion, the EBw.s activity bump largely remained tethered to the position of one landmark even in the presence of another identical landmark in front of the fly. The bump also did not always shift instantaneously following an abrupt displacement of visual landmarks, as if temporarily retaining the original orientation reference before locking on to its new position (Seelig, 2015).

Several models have been proposed to explain how visual landmark and self-motion cues are integrated at the level of head direction cell activity in mammals. Most rely on circuits organized as ring attractors: neurons are schematized as being arranged in a circle based on their preferred directions, with connection strengths that depend on their angular separation. With initial sensory input and an appropriate balance of recurrent excitation and inhibition, such a circuit can generate and sustain a localized activity bump. The bump's position on the circle corresponds to the animal's heading which is then updated by directional drive from self-motion signalling neurons. Direct experimental evidence in support of these models has been difficult to obtain in mammals owing to the distributed nature of the underlying circuits. Although the functional connectivity between EBw.s neurons is not yet known, several of the expected features of ring attractor models were observed in the dynamics of this population of CX neurons: organization of activity into a localized bump, movement of the bump to neighbouring wedges based on self-motion, drift in bump location in darkness, persistent activity, and both abrupt jumps and gradual transitions of the activity bump when triggered by strong visual input. Cell-intrinsic mechanisms could also underlie some of these features, including, for example, persistent activity. The genetic tools available in Drosophila to target and manipulate the activity of identified cell types should allow different models for visually guided orientation and angular path integration to be discriminated at the level of synaptic, cellular and network mechanism (Seelig, 2015).

Although anatomy is often the first step in assigning functions to neural structures, it is not always clear whether architecturally distinct regions of the brain correspond to operational units. Whereas neuroarchitecture remains relatively static, functional connectivity may change almost instantaneously according to behavioral context. This study imaged panneuronal responses to visual stimuli in a highly conserved central brain region in the fruit fly, Drosophila, during flight. In one substructure, the fan-shaped body, automated analysis reveals three layers that are unresponsive in quiescent flies but become responsive to visual stimuli when the animal is flying. The responses of these regions to a broad suite of visual stimuli suggest that they are involved in the regulation of flight heading. To identify the cell types that underlie these responses, activity was imaged in sets of genetically defined neurons with arborizations in the targeted layers. The responses of this collection during flight also segregate into three sets, confirming the existence of three layers, and they collectively account for the panneuronal activity. These results provide an atlas of flight-gated visual responses in a central brain circuit (Weir, 2015).

Insects exhibit an elaborate repertoire of behaviors in response to environmental stimuli. The central complex plays a key role in combining various modalities of sensory information with an insect's internal state and past experience to select appropriate responses. Progress has been made in understanding the broad spectrum of outputs from the central complex neuropils and circuits involved in numerous behaviors. Many resident neurons have also been identified. However, the specific roles of these intricate structures and the functional connections between them remain largely obscure. Significant gains rely on obtaining a comprehensive catalogue of the neurons and associated GAL4 lines that arborize within these brain regions, and on mapping neuronal pathways connecting these structures. Toward this end, small populations of neurons in the Drosophila melanogaster central complex were stochastically labeled using the multicolor flip-out technique and a catalogue was created of the neurons, their morphologies, trajectories, relative arrangements and corresponding GAL4 lines. This report focuses on one structure of the central complex, the protocerebral bridge, and identifies just 17 morphologically distinct cell types that arborize in this structure. This work also provides new insights into the anatomical structure of the four components of the central complex and its accessory neuropils that are arborized by PB neurons include the crepine (CRE), rubus (RUB), gall (GA), and lateral accessory lobe (LAL). Most strikingly, the protocerebral bridge was found to contain 18 glomeruli, not 16, as previously believed. Revised wiring diagrams that take into account this updated architectural design are presented. This updated map of the Drosophila central complex will facilitate a deeper behavioral and physiological dissection of this sophisticated set of structures (Wolff, 2014).

The work presented in this study builds on published studies by both defining previously unidentified anatomical features of each of the four components of the central complex as well as updating wiring diagrams to accommodate these new anatomical insights. This paper also reports new cells and new features of previously identified cells and the genetic reporter lines that reveal them, with the prospect that these will form an essential stepping stone both to synaptic studies at the electron microscope level and to functional studies. The most significant new insights from this work are summarized below. As noted earlier, the statements below are drawn from neurons that arborize in the PB (Wolff, 2014).

The most surprising finding of this study is that the Drosophila protocerebral bridge comprises 18 glomeruli. This finding has an important impact on the wiring relationships between the glomeruli and their respective vertical units in the FB, the columns, and in the EB, the wedges and a new volume described in this study, the tiles (Wolff, 2014).

The longstanding belief regarding the correspondence between the PB and FB and PB and EB wedges was that there is a 1:1 relationship between the vertical subdomains of these structures. The finding that there are 18 glomeruli raised the possibility that the FB and EB also exhibit an octodecimal organization. However, compelling evidence is provided that there are just 16 wedges in the EB. and it was further show, that some cells arborize in just half a wedge, indicating the further division of wedges into 32 demi-wedges. The observation that a simple 1:1 correspondence between the PB and EB wedges is lacking and, furthermore, that there are also demi-wedges, has implications for how the system is wired to accommodate this numerical discrepancy (Wolff, 2014).

Another unexpected finding from this work is the existence of a second EB volume that also partitions the EB around its circumference: the tile domain. Tiles are distinct from wedges in that there are half as many tiles as wedges (eight tiles), they are functionally distinct from the wedge (output versus input, respectively), and these two volumes survey different volumes of the EB since they extend to different depths of the EB. Only two PB cell types target the tile domain (Wolff, 2014).

Although a columnar morphology is apparent in layers 1–8 of the FB in nc82-labeled samples, the organization of the cells that populate these layers is not universally columnar. There is a minimum of nine layers in the FB, yet a columnar organization (i.e., vertical stratification) of cell arbors is restricted to layers 1–5 for single column widths (where the columnar organization for layers 4 and 5 is revealed by the PB_G1–8.s-FBℓ3,4,5.s.b-rub.b cell); wider, more loosely organized arbors occur in more dorsal layers. With the exception of arbors in layer 1, the column borders are not rigid, as neighboring arbors overlap one another, sometimes extensively. The unique tooth-like structure of layer 1 of the FB definitively shows that there are nine columns in this layer. Due to the overlap of arbors, it is more difficult to count columns in the other layers, but layers 2 and 3, which exhibit a tighter columnar organization than more dorsal layers, likely have 16 columns based on mapping data. This would be consistent with parallel divisions of the EB (wedges) (Wolff, 2014).

Other new anatomical features and subdomains are described in this study. First, it was shown that each of the noduli has subcompartments. The dorsal noduli, NO₁, have medial and lateral subcompartments. The medial noduli, NO₂, consist of two distinct subcompartments, NO₂D and NO₂V; no PB cell type arborizes in either of these subcompartments. The ventral noduli, NO₃, consist of three distinct subcompartments, NO₃A, NO₃M, and NO₃P. The nubbin is a partial shell on the dorsal, anterior face of the EB, and the "gall tip" is a region at the dorsal tip of the gall. Finally, two undefined regions to which some cells project and that are not clearly demarcated are the dorsal and ventral gall surround (Dga-s and Vga-s) (Wolff, 2014).

Even though the subdomains of the central complex structures can be distinguished from one another, they apparently do not function as isolated subunits. Rather, there is shared communication between most of these subunits. At least for the neurons described in this study (i.e., those that arborize in the PB), both pre- and postsynaptic arbors in the glomeruli, EB tiles, wedges and shells, FB columns, and NO₁ (medial and lateral domains) can extend into neighboring domains. This sharing of information is not obvious between NO₂D and NO₂V, nor between NO₃A and NO₃M. The boundary between NO₃P and NO₃M is too obscure to evaluate if arbors in these two domains are completely restricted or shared, although in the examples they appear to be restricted. The frequency and degree to which arbors overlap in the various subunits is cell type-dependent. While some arbors exhibit no or minimal intrusion into adjacent volumes, overlap between neighboring units could serve important circuit functions (Wolff, 2014).

This work identifies 17 unique cell types that arborize in the protocerebral bridge. These fall into four classes: cells that 1) are intrinsic to the PB (n = 2), 2) are intrinsic to the central complex (an additional 6), 3) arborize in the FB, EB, or NO in addition to extra-central complex regions (e.g., the gall; n = 6), and 4) arborize exclusively in the PB and regions outside the central complex (n = 3). Cells that arborize in the PB receive their input from the EB, LAL, PS, IB, and also from within the PB. One cell previously identified in another study (Lin, 2013) was not targeted by any of the ∼35 lines analyzed in this work. To the extent that it is possible to construct wiring diagrams from the images shown in these two studies, it appears that the circuits for these cells are also identical between Drosophila and Schistocerca. In addition, one cell type was identified in this study that was not characterized in (Wolff, 2014).

The combined total from this work anda previous study brings the current number of identified cell types that arborize in the protocerebral bridge of Drosophila to 18. A potential 19th cell type was seen just twice, and in neither case could the entire cell be traced. Its PB arbor is spiny and very sparse, and while it clearly arborizes in the central brain, it is not clear if it arborizes elsewhere within the central complex. It is also possible that this "cell type" only constitutes a variant. There will likely be additional cells identified that arborize in the PB, although this number is predicted to be small. A complete inventory of all cells in the Drosophila brain awaits a full reconstruction at an electron microscope level (Wolff, 2014).

The wiring diagrams described in this study differ from published reports, in part due to the fact that previous authors were unaware of the existence of 18 glomeruli in the PB and therefore based their models on the historic interpretation that there are 16 glomeruli. This numerical revision and new insights into the anatomical substructure of the central complex components are the primary basis for revisions of existing circuit diagrams (Wolff, 2014).

Although there are 18 glomeruli, no small-field neuron arborizes in all 18 glomeruli. Instead, most cell types arborize in either G1–G8 or G2–G9. Each of these categories adheres to the following basic wiring principle: Cells that arborize in the lateral four glomeruli of each side of the PB stay ipsilateral in the second neuropil (either the FB or EB, depending on the neuron) and cross to the contralateral side at the third neuropil, whereas cells that arborize in the medial four glomeruli cross to the contralateral side in the second neuropil. Consequently, because there are two subsets of PB neurons, the glomerulus that targets a given column, wedge, or tile is shifted by one glomerulus, depending on the subset of cell type. Furthermore, the observation that no small-field neurons arborize in all 18 glomeruli suggests the number of columns in the FB and wedges in the EB would not need to exceed 16 in order to maintain a 1:1 correspondence between the PB and FB/EB (Wolff, 2014).

Arbors from cells that target the PB alternate with one another in the second neuropil such that arbors from the left glomeruli alternate with those from the right glomeruli. The PB wiring diagrams presented in this study differ somewhat from a recent account (Lin, 2013), as follows. The most lateral FB column (or EB wedge) is occupied not by the ipsilateral G9 (or G8 for the G1–G8 cells), as previously described, but instead by the contralateral G2 (or G1 for the G1–G8 cells). This circuit therefore reverses the pattern in the second neuropil (FB or EB) from one in which the most lateral (L) glomeruli project to the most lateral columns (or wedge or tile) on the ipsilateral side to one in which the medial (M) glomeruli from the contralateral half of the PB project to the most lateral columns. In other words, previously published diagrams indicate a pattern of LMLMLM from lateral to medial in the second neuropil, whereas this report shows that pattern to be MLMLML (Wolff, 2014).

The projection map shown in this study for cells that connect the PB to layer 1 of the FB illustrates conclusively the projection pattern between these domains. Obtaining an accurate map between the PB and layers 2 and 3 is difficult given the greater overlap between arbors of cells in these two layers, but the projection patterns observed between the PB and layers 2 and 3 of the FB are consistent with the PB:FBℓ1 map (Wolff, 2014).

As noted above, distribution of information is not always restricted to the subdomains of each central complex structure. When information is shared between neighboring domains (or alternating domains, in the case of the cells that arborize in the dorsal or ventral gall), generally only a small portion of the arbor is shared. The functional significance of these zones of overlap remains to be determined (Wolff, 2014).

Connections between central complex structures are remarkably restricted. For example, of those neurons that arborize in the PB and FB, only FB layers 1, 2, and 3 connect to the noduli, and only to NO₂ and NO₃. The only link between the PB and NO₁ is via the ellipsoid body, so whereas NO₂ and NO₃ can be considered to work in conjunction with the FB to elicit a behavior based on output from the PB, NO₁ cooperates with the EB to elicit a behavior based on output from the PB. In fact, communication is even more specific: layer 3 of the FB communicates directly only with NO₂, layer 2 directly only with the anterior subcompartment of NO₃, and layer 1 directly only with NO₃M and NO₃P. Furthermore, the cells in G1 do not communicate directly with the noduli at all—neither via the FB nor via the EB. The absence of direct connections between the PB and upper layers of the FB is also noteworthy. This streamlined and highly segregated network of connections within and between central complex structures suggests a high degree of regional specialization in function for the components of the central complex (Wolff, 2014).

The roles of central complex structures, their subdomains, and related neuropils are poorly understood. While functions remain largely unknown, many circuits described in this study are informative in various other ways. For example, some identify commonalities in function between neuropil subregions, such as FBℓ2 and NO₃A, which are both arborized by a common neuron. Other circuits reveal spatial segregation between neuropils. The most intriguing instance is the exclusive relay of information between the ventral gall and even-numbered glomeruli and the dorsal gall and odd-numbered glomeruli, which demonstrates that both information and information flow can be spatially segregated from the glomeruli to the gall. It will be interesting to learn the functional role of the gall and why it segregates a portion of the information it receives and sends, as well as what sort of behavioral response requires this rigidly alternating spatial distribution in the PB. Finally, the absence of connections between neuropils may prove informative in functional studies. For example, G1 is distinct from G2–G8 in that it lacks direct connections with the noduli, and NO₁ is distinct from NO₂ and NO₃ in that it communicates with the PB via the EB rather than the FB, raising the questions of what behaviors G1 does and does not contribute to, and what the differences are in behavioral outputs from NO₁ and NO₂/NO₃ (Wolff, 2014).

The observation that there are 18 glomeruli in the Drosophila PB has significant implications for both the architecture and evolution of the Drosophila brain with respect to the brains of other neopterans. Although the suite of genetic tools available in locusts, bees, beetles, and other insects does not yet include MCFO, improvements in imaging and histology may prove sufficient to reevaluate the number of glomeruli in these species, given that glomeruli can be accurately counted in brains that are labeled only with nc82. Either some or all of these species also have 18 glomeruli, or an extra pair of glomeruli arose in Drosophila. The latter may be unlikely but would raise some intriguing possibilities about how the anatomical correspondence and circuitry between glomeruli in the PB and equivalent vertical partitions in the FB and EB PB circuits may differ between flies and other insects thought to have the same basic cellular composition and organization within the central complex, and how the geometric coordinates would then have had to shift along the axis of the PB to effect accurate behavioral responses (Wolff, 2014).

In Drosophila, aversive olfactory memory is believed to be stored in a prominent brain structure, the mushroom body (MB), and two pairs of MB intrinsic neurons, the dorsal paired medial (DPM) and the anterior paired lateral (APL) neurons, are found to regulate the consolidation of middle-term memory (MTM). This study reports that another prominent brain structure, the ellipsoid body (EB), is also involved in the modulation of olfactory MTM. Activating EB R2/R4m neurons does not affect the learning index, but specifically eliminates anesthesia-sensitive memory (ASM), the labile component of olfactory MTM. It was further demonstrated that approximately two-thirds of these EB neurons are GABAergic and are responsible for the suppression of ASM. Using GRASP (GFP reconstitution across synaptic partners), potential synaptic connections were revealed between the EB and MB in regions covering both the presynaptic and postsynaptic sites of EB neurons, suggesting the presence of bidirectional connections between these two important brain structures. These findings suggest the existence of direct connections between the MB and EB, and provide new insights into the neural circuit basis for olfactory labile memory in Drosophila (Zhang, 2013).

Previous studies have shown that the EB plays an essential role in visual pattern memory, orientation memory and place learning (Pan, 2009; Ofstad, 2011; Kuntz, 2012), and thus it is usually considered to be a center of visual learning and mem- ory. Interestingly, one study on NMDA receptors reported that the EB is required for olfactory long-term memory (LTM) consolidation; however, the underlying neural circuits remain uninvestigated. The current results reveal that a group of EB neurons, the c819-labeled R2/R4m neurons, plays an inhibitory role in the modulation of MTM but not the immediate memory. This points to a new function of the EB in olfactory cognition and further demonstrates that the EB could be involved in the process of olfactory aversive learning and memory from an earlier stage than previously thought (Zhang, 2013).

The presence of dense GABA-like immunoreactivity has been demonstrated in the EB ring and RF tract suggesting that the bulk of EB neurons are GABAergic. Although this finding has been confirmed by several other studies, the function of these GABAergic neurons in cognition is still unclear. The current results further reveal that approximately two-thirds of the c819-EB neurons are GABAergic, and they play an inhibitory role in ASM modulation. The GABAA receptor, resistant to dieldrin (RDL), has been shown to be highly expressed in the MB lobes and the EB. It is thus possible that these EB GABAergic neurons function through RDL receptors (Zhang, 2013).

As a component of MTM, ASM has been suggested to be stored in the MB and to be modulated by MB intrinsic APL and DPM neurons, of which the neural terminals are restricted to the MB. This study reports the EB, a brain structure separate from the MB, is also involved in the modulation of ASM. WEB neurons may be both presynaptic and postsynaptic to MB neurons, suggesting that they may suppress 3 h ASM via putative direct connections between the EB and MB. That the MB and EB are two discrete but possibly interconnected and interacting brain regions, suggests that it is important to study the process of learning and memory over a more widely distributed neural network. However, the interaction between neurons from different structures may endow the network with greater capacity for more complex activities (Zhang, 2013).

It is also interesting to discover that activating c819-EB GABAergic neurons during training impaired 3 h ASM instead of immediate learning performance. Recently it has been reported that blocking two pairs of dopaminergic neurons during intertrial intervals in spaced training suppresses the formation of 24 h LTM by interfering with the slow oscillations in these dopaminergic neurons. Since this study has shown that bidirectional connections may exist between the EB and MB, it is proposed that there may be a small feedback circuit between the EB and MB, which may have prolonged oscillations and therefore affect 3 h ASM consolidation. Further functional imaging studies may provide more clues on how this neural circuit functions (Zhang, 2013).

How the brain perceives sensory information and generates meaningful behavior depends critically on its underlying circuitry. The protocerebral bridge (PB) is a major part of the insect central complex (CX), a premotor center that may be analogous to the human basal ganglia. By deconstructing hundreds of PB single neurons and reconstructing them into a common three-dimensional framework, this study has constructed a comprehensive map of PB circuits with labeled polarity and predicted directions of information flow. The analysis reveals a highly ordered information processing system that involves directed information flow among CX subunits through 194 distinct PB neuron types. Circuitry properties such as mirroring, convergence, divergence, tiling, reverberation, and parallel signal propagation were observed; their functional and evolutional significance is discussed. This layout of PB neuronal circuitry may provide guidelines for further investigations on transformation of sensory (e.g., visual) input into locomotor commands in fly brains (Lin, 2014: PubMed).

Studying the insect visual system provides important data on the basic neural mechanisms underlying visual processing. As in vertebrates, the first step in visual processing in insects is through a series of retinotopic neurons. Recent studies on flies have found that these converge onto assemblies of columnar neurons in the lobula, the axons of which segregate to project to discrete optic glomeruli in the lateral protocerebrum. This arrangement is much like the fly's olfactory system, in which afferents target uniquely identifiable olfactory glomeruli. Whole-cell patch recordings show that even though visual primitives are unreliably encoded by single lobula output neurons because of high synaptic noise, they are reliably encoded by the ensemble of outputs. At a glomerulus, local interneurons reliably code visual primitives, as do projection neurons conveying information centrally from the glomerulus. These observations demonstrate that in Drosophila, as in other dipterans, optic glomeruli are involved in further reconstructing the fly's visual world. Optic glomeruli and antennal lobe glomeruli share the same ancestral anatomical and functional ground pattern, enabling reliable responses to be extracted from converging sensory inputs (Mu, 2012).

Retinotopic output neurons from the lobula of Drosophila, which have axon diameters of 0.5 μm or less, do not transmit action potentials. This is typical of the many small interneurons in the insect visual system that are arranged as repeat ensembles. However, mushroom body intrinsic neurons (Kenyon cells) may be a special exception to this because very few of them, at any time, are required to accurately encode odorant identity through the mechanism of 'sparsening' (Mu, 2012).

In the dipteran Phormia, relays connecting the medulla to the lobula and lobula plate have axon diameters of between 0.5 μm and 3 μm. These neurons generally respond with graded potentials as do the larger axon diameter (2–4 μm) lamina monopolar cells, which extend from the lamina to the medulla. Even the 15 μm diameter axons of 'giant' motion sensitive neurons in the lobula plate can conduct by graded potentials in addition to spiking responses. However, there are some exceptions. In the hoverfly Eristalis tenax, object-detecting neurons relaying from lobula show clear spiking responses, as do neurons in Phormia that respond to moving bars (Mu, 2012).

Signal reliability is also critical for neurons that occur as single pairs of uniquely identifiable neurons, or as very small populations in the brain, or small subsets of Kenyon cells, each subset encoding an odorant. In Drosophila, such neurons conduct by spikes or by mixed codes: membrane potential fluctuations and action-potentials. Examples are the wide-field directional selective tangential cells of the lobula plate which occur either as a uniquely identifiable set of 3 HS neurons, or as 11 uniquely distinct VS neurons which collaborate to mediate responses to changes in optic flow. Neurons with long axons, such as the unique pairs of interneurons linking the central body with many areas of the lateral protocerebrum and deutocerebrum, or those which carry data from the brain to thoracic ganglia also invariably conduct by spikes (Mu, 2012).

In contrast to uniquely identifiable pairs or small clones of neurons belonging to the midbrain, many neurons in the optic lobes occur as ensembles of identical, clonally related neurons. In the medulla of Drosophila and other fly species, there are about 50 types of retinotopic neurons, spaced one to each retinotopic column. In the lobula, there are about 15 different clones of output neurons, each of which comprises an ensemble of about 40 identical neurons. Each neuron of an ensemble subtends 6–9 visual sampling units of the retina and has dendrites, and thus receptive fields (Okumura, 2007), that overlap with a surround of at least 8–12 neurons of the same clonal identity. This anatomical arrangement ensures that 8–12 neurons of the same clone view the same part of the visual field (Mu, 2012).

In Drosophila, such outputs from the lobula have extremely thin axons and these cells conduct by graded potentials. Each LCN exhibits significant membrane-voltage fluctuations, which likely reflect the many postsynaptic sites from medulla afferents. An important finding of this study was that recordings from many single L1CNs show that none reliably encodes a visual primitive, whereas the summed responses of L1CNs show clear responses to defined visual stimuli. Thus, since any ensemble of LCNs converges at its unique glomerulus, it is expected that a subset of LCNs will respond to a given visual stimulus and that the summed responses of this subset would drive postsynaptic neurons of their target glomerulus. In larger dipterans, it has also been shown that different optic glomeruli respond to different visual primitives (Mu, 2012).

It is conceivable that the weak responses recorded in individual LCNs is due to the long electrotonic distance between soma and axon. However there are two major reasons to reject the idea that these non-spiking characteristics are artifactual. Firstly, using an identical recording methodology, small spiking neurons in the mid brain were also shown to have long thin neurites between their cell body and their main integrative region. Secondly, recordings of the smallest retinotopic neurons in the medulla of a larger fly species, Phaenicia sericata, consistently showed that they encode data in a non-spiking fashion, irrespective of the location of the electrode in the neuron (Mu, 2012).

If an individual output neuron from the lobula complex can have subtle and variable responses to specific visual stimuli, but the summed responses of a subset of LCNs belonging to the same clone show a clearer response, might local interneurons post-synaptic to their terminals in their relevant optic glomerulus integrate input signals and unambiguously respond to the same visual stimuli? Recordings of a LIN in the giant fiber optic glomerulus complex suggest this is case: the LIN responds unambiguously to a looming stimulus, whereas the response of the single type 2 lobula plate-lobula columnar neuron (LPL2CN) to the same stimulus can only be resolved from a power spectrum analysis. However, when responses of many of the same type of lobula output neurons are summed, their collective response is unambiguous (Mu, 2012).

The giant fiber (GF) glomerulus receives LPL2CN inputs and contains LIN processes as well as one major dendritic process of the GF. The GF glomerulus glomerular local interneuron (LIN) responds to looming stimuli, and responds to intensity decrements. Looming stimuli activate the GF glomerulus's LPL2CN inputs. Responses by the LIN are also the same as those that drive the GF. That the LIN rapidly adapts to looming stimuli whereas GF does not suggests that several LINs are associated with the glomerulus and that these may recruit signals from successive groups of activated LPL2CN afferents. Though it remains to be demonstrated that the LPL2CN clone is presynaptic to the LIN and GF, there is strong evidence in larger dipteran species that the Col A afferents, which also converge on the GF glomerulus, are directly presynaptic to the GF. For example, electromicroscopy studies have shown that in Musca domestica, Col A cells establish electrical synapses onto the GF, and cobalt introduced into the GF passes, rather spectacularly, into the entire array of Col A afferents. Col A cells in the fly Phoenicia serricata respond with graded potentials to decrements in illumination and to movement of edges. The convergence of Col A neurons and neurons of the LPL2CN clone at the GF glomerulus does suggest that there is a more complex control system eliciting GF responses than has been hitherto envisaged (Mu, 2012).

The convergence of axons from a clone of optic lobe outputs to an optic glomerulus suggests a mechanism that establishes reliable downstream responses: one or more local interneurons of the glomerulus complex integrate and average inputs from members of an isomorphic population of retinotopic relay neurons from the lobula complex. Recordings from the GF glomerulus show that its LIN responds reliably to the same looming stimulus that drives the LPL2CN afferent supply to that glomerulus. The demonstration that the LIN response is relatively noise-free, suggests that one function of LINs is to disambiguate information carried by afferents to a glomerulus, from synaptic noise generated at the dendritic trees within the lobula. Noise free information could then be relayed by the LIN to the glomerulus' projection neurons. These are of two types: premotor descending neurons, such as the GF, which project to the thoracic ganglion; and relay neurons which project to higher centers in the brain, such as the dorsal protocerebral lobes, and their connections to the central complex (Mu, 2012).

This convergence of lobula outputs to uniquely identifiable optic glomeruli in the brain's first segment, the protocerebrum, is comparable to the convergence of olfactory sensory neurons (OSN) to antennal lobe glomeruli in the brain's second segment, the deutocerebrum, where each unique glomerulus in the fly's antennal lobe is targeted by the axons of a specific set of olfactory sensory neurons (OSNs) on the antenna, which expresses a particular olfactory receptor protein. In the antennal lobe, noisy signals from OSNs are refined by local interneurons and then relayed to higher centers by projection neurons. The present results provide electrophysiological evidences that noisy signals in an isomorphic population of lobula outputs are similarly refined by local interneurons of the optic glomerular complex. Therefore reliable responses in an optic glomerulus are established through convergence and signal averaging processes (Mu, 2012).

The present studies further support the proposition that the optic glomerular complex and the antennal lobes are serially homologous neural systems having the same principle anatomical and functional organization, and with the common function of refining and integrating incoming signals (Strausfeld, 2007). Glomerular organization in the protocerebrum and deutocerebrum reflect a ground pattern that can be identified in every ganglion of the central nervous system. Throughout, each type of receptor, representing one or another modality, sends its axon to a specific domain in the relevant ganglion. These domains, in some ganglia represented by glomerular volumes, in others by allantoid or ovoid ones, are connected by spiking and non-spiking local interneurons which integrate the sensory input and relay behaviorally meaningful information to central neuropils and to motor circuits. Such arrangements evolutionarily derive from an ancestral ground pattern seen in archaic arthropods, each segment of which was composed of identical elements. As demonstrated by the protocerebrum and deutocerebrum, present day insects reflect this ancestral ground pattern even in the brain, despite each segment having evolved its unique sensory configuration (Mu, 2012).

Drosophila vision is mediated by inputs from three types of photoreceptor neurons; R1-R6 mediate achromatic motion detection, while R7 and R8 constitute two chromatic channels. Neural circuits for processing chromatic information are not known. This study identified the first-order interneurons downstream of the chromatic channels. Serial EM revealed that small-field projection neurons Tm5 and Tm9 receive direct synaptic input from R7 and R8, respectively, and indirect input from R1-R6, qualifying them to function as color-opponent neurons. Wide-field Dm8 amacrine neurons receive input from 13-16 UV-sensing R7s and provide output to projection neurons. Using a combinatorial expression system to manipulate activity in different neuron subtypes, it was determined that Dm8 neurons are necessary and sufficient for flies to exhibit phototaxis toward ultraviolet instead of green light. It is proposed that Dm8 sacrifices spatial resolution for sensitivity by relaying signals from multiple R7s to projection neurons, which then provide output to higher visual centers (Gao, 2009).

Many animals respond differentially to light of different wavelengths: for example, most flying insects exhibit positive phototactic responses but prefer ultraviolet (UV) to visible light, whereas zebra fish are strongly phototactic to ultraviolet/blue and red light but weakly to green. Unlike true color vision, which distinguishes lights of different spectral compositions (hues) independently of their intensities, spectral preferences are strongly intensity-dependent and innate, probably reflecting each species' ecophysiological needs. Thus, water fleas (Daphnia magna) avoid harmful UV but are attracted to green light, which characterizes abundant food sources. Daylight is rich in UV, so flying insects' preference for UV over visible light is probably related to the so-called open-space response, the attraction towards open, bright gaps and away from dim, closed sites. The receptor mechanisms for spectral preference has been well studied in flying insects, especially in Drosophila. Two or more photoreceptor types with distinct spectral responses are required to detect different wavelengths of light, and mutant flies lacking UV-sensing photoreceptors exhibit aberrant preference for green light. However, the post-receptoral mechanisms of spectral preference are entirely unknown. Furthermore, it is not clear how spectral preference is related to true color vision. Color-mixing experiments suggest that color vision spectral preference are independent in honeybees. In Drosophila, however, spectral preference experiments have revealed that the phototactic response towards UV is significantly enhanced by the presence of visible light, suggesting a 'color' contrast effect in spectral preference behavior. Identifying and characterizing the neural circuits that process chromatic information is the first step to understanding the post-receptoral mechanisms of spectral preference and thus color vision (Gao, 2009).

With recent advances in genetic techniques that manipulate neuronal function, Drosophila has re-emerged as a model system for studying neural circuits and functions. In particular, the Gal4/UAS expression system combined with the temperature-sensitive allele of shibire makes it possible to examine the behavioral consequences of reversibly inactivating specific subsets of neurons. Such interventions allow direct comparisons between the connections of a neuron and its function, thereby establishing causality (Gao, 2009).

The Drosophila visual system comprises the compound eye and four successive optic neuropils (lamina, medulla, lobula and lobula plate. The compound eye itself has some 750 ommatidia, populated by two types of photoreceptors. The outer photoreceptors R1-R6, which are in many ways equivalent to vertebrate rod cells, express Rh1 opsin and respond to a broad spectrum of light, and are thus presumed to be achromatic. The inner photoreceptor neurons R7 and R8 have complex opsin expression patterns: R7s express one of two ultraviolet (UV)-sensitive opsins, Rh3 and Rh4, while beneath R7 the R8s coordinately express blue-sensitive Rh5 or green-sensitive Rh6 opsins. The achromatic R1-R6 channel mediates motion detection. R1-R6 innervate the lamina, where the achromatic channel input diverges to three or more pathways mediated by three types of lamina neurons, L1-L3. Their synaptic connections have been analyzed exhaustively at the electron microscopic (EM) level. Genetic dissection indicates that these three pathways serve different functions in motion detection and orientation. Much like vertebrate cones, R7 and R8 photoreceptors are thought to constitute chromatic channels that are functionally required for spectral preference behaviors. The axons of R7 and R8 penetrate the lamina and directly innervate the distal medulla, where until now their synaptic connections have been completely unknown (Gao, 2009).

The medulla, the largest and most heavily populated optic neuropil, is organized into strata (M1-M10) and columns, in a manner reminiscent of the mammalian cortex. All visual information converges upon the distal strata of the medulla: the axons of R7 and R8 directly innervate strata M6 and M3, respectively, while L1-L3 transmit information from the R1-R6 channel to multiple medulla strata (M1/5, M2, and M3, respectively). The R7, R8, and L1-L3, which view a single point in visual space innervate a single medulla column and there establish a retinotopic pixel. Previous Golgi studies have revealed about 60 morphologically distinct types of medulla neurons. Each arborizes in a stereotypic pattern within specific strata of the medulla, and projects an axon to a distinct stratum of the medulla, lobula or lobula plate. The distinct morphological forms of different types of medulla neurons reflect, at least in part, their diverse patterns of gene expression. Although it is widely presumed that the medulla incorporates key neural substrates for processing color and motion information, little is known about its synaptic circuits and their functions. EM analyses of synaptic circuits have not been possible because of the complexity of this neuropil, while electrophysiological investigations are technically challenging because of the small size of neurons (Gao, 2009).

This study investigated the chromatic visual circuits in the medulla. Using a combination of transgenic and histological approaches, the first-order interneurons in the medulla that receive direct synaptic inputs from the chromatic channels, R7 and R8 were identified. These neurons were subdivided based on their use of neurotransmitters and gene expression patterns. By systematically inactivating and restoring the activity of specific neuron subtypes, the neurons that are necessary and sufficient to drive a fly's phototactic preference to UV were identified (Gao, 2009).

Previous electrophysiological and histological studies have demonstrated that Drosophila photoreceptor neurons are histaminergic and that R7 and R8 photoreceptors provide the predominant histamine-immunoreactive input to the medulla. Two ionotropic histamine-gated channels, Ort (ora transientless; HisCl2) and HisCl1 have been identified. Mutants for ort exhibit defects in motion detection and their electroretinograms (ERGs), indicating that Ort is required to transmit R1-R6 input to the first-order interneurons (Gengs, 2002). To test whether Ort is required for visually guided behavior, flies' phototaxis towards either UV or green light in preference to dark was examined. This phototactic response is mediated primarily by the more sensitive, broad-spectrum photoreceptors, R1-R6, although R7 cells also contribute to UV, but not green, phototaxis under the light-adapted condition. Wild-type flies exhibit stronger phototaxis towards UV than towards green light by approximately one order of magnitude, and light-adaptation, when compared with dark-adaptation, reduces the sensitivity to UV and green light by approximately two orders of magnitude. In contrast, strong transallelic combination ort¹/ort^US2515 mutant flies exhibit much weaker phototaxis towards either UV or green light (by about three and two orders of magnitude, respectively) as compared with wild-type. In negative geotaxis assays, ort mutants exhibit no apparent motor defects, suggesting that their reduced phototaxis was not a motor system defect but rather a visual deficit. In addition, the ort mutation affects UV phototaxis more severely than green phototaxis. It is speculated that Ort plays a role in relaying signals from UV-sensing R7s to their first-order interneurons, and that HisCl1 may participate in phototaxis, especially towards green light (Gao, 2009).

To assess whether Ort is required to transmit chromatic input mediated by R7 and R8, a quantitative spectral-preference assay was used. This spectral-preference assay tests the phototaxis towards UV in preference to green and depends on R7, but not significantly on R1-R6, function. This behavior depends on the circuit comparing UV and green light and likely reflects salience of UV and green lights rather than a simple linear summation of their phototactic responses. It was found that wild-type flies prefer short-wavelength UV to longer-wavelength green light in an intensity-dependent fashion. In contrast, homozygous null ort¹ mutants and strong transallelic combination ort¹/ort^US2515 mutants (as well as other allelic combinations, ort^P306/ort^US2515 and ort¹/ort^P306) all exhibited reduced UV preference. Over five orders of magnitude in the ratio of UV/green intensities, the proportion of ort mutant flies that chose UV was significantly lower than that for wild-type flies. To quantify the UV preference, the isoluminance point, the UV/green intensity ratio at which flies found light of either wavelength equally 'attractive', was determined, and the negative logarithm of the intensity ratio was used as a measure of UV attractiveness. The UV attractiveness for ort mutants was significantly lower than that for wild-type flies (Attr_UV/G=2.52±0.23) but higher than that for sevenless mutants (Gao, 2009).

Given that ort null mutants still exhibits phototaxis, whether the other histamine receptor, HisCl1, might contribute to UV preference, was examined. HisCl1¹³⁴ null mutants were found to exhibit UV preference indistinguishable from the wild-type. In contrast, strong allelic combination HisCl1¹³⁴ ort¹/HisCl1¹³⁴ ort^P306 double-mutants shows weak phototaxis towards green light, while double-null HisCl1¹³⁴ ort¹ mutants, like the phototransduction mutant NorpA, fails to discriminate between wavelengths in the UV and green. It is concluded that Ort is essential for optimal UV preference while HisCl1 plays at most a minor and partially redundant role. It is noted that double-null HisCl1¹³⁴ ort¹ mutants are not entirely blind and still exhibit very weak fast phototaxis, suggesting that there might be residual synaptic transmission between photoreceptors and the first-order interneurons despite of the absence of these two known histamine receptors (Gao, 2009).

It was reasoned that the first-order interneurons must express the histamine receptor Ort in order to respond to their inputs from histaminergic R7 and R8 terminals (see The histamine chloride channel Ort is expressed in subsets of lamina and medulla neurons). To identify these first-order interneurons, the ort promoter region was determined using comparative genomic sequence analysis (Odenwald, 2005; Yavatkar, 2008). In the ort locus, four blocks of non-coding sequence were found that are highly conserved among 12 species of Drosophila. The first three sequence blocks (designated C1-C3) are localized to the intergenic region and the first intron and are therefore likely to contain critical cis-elements. ort-promoter constructs driving Gal4 or LexA::VP16 were generated, designating these ort^C1-3-Gal4 and ort^C1-3-LexA::VP16. Both driver systems drove expression patterns in identical subsets of neurons in the lamina, medulla cortices and in the deep C and T neurons of the lobula complex, except that ort^C1-3-Gal4 drove somewhat patchy expression with lower intensity. The fourth block of conserved sequences, located at 3'UTR, contains putative microRNA binding sites and, as examined in ort^C1-4-Gal4, did not drive expression in additional cells, suggesting that it does not contain critical cis-elements. Overall, the expression patterns of these ort promoter constructs resembled previously published ort expression patterns from in situ hybridization (Gao, 2009).

Comparative genomic sequence analysis was also performed for the HisCl1 locus and two blocks of highly conserved sequence (C1 and C2), located in the first introns of the HisCl1 gene and its neighboring gene (CG17360) were identified. A HisCl1-Gal4 construct was generated that included these conserved sequences. It was found that HisCl1-Gal4 drove strong expression in the lamina epithelial glia cells (as recently also reported by Pantazis (2008) and medulla cells that are not well characterized. This result is consistent with previous EM data that lamina epithelial glia enwrap each cartridge and are postsynaptic to R1-R6. Insofar as both the behavioral requirement and expression pattern indicate that Ort but not HisCl1 plays a critical role in the visual system, focus was placed on Ort in the following analyses (Gao, 2009).

Whether using the ort-promoter Gal4 drivers to express Ort is sufficient to rescue defects in the visual behavior ort mutants was examined. It was found that ort^C1-4-Gal4-driven Ort expression restored a preference for UV (Attr_UV/G=2.25±0.34) in ort mutants to the wild-type level. Since Ort, but not HisCl1, is also required in lamina neurons for normal ERG and motion detection responses (Gengs, 2002; Rister, 2007), the rescued flies were examined for these functions too. It was found that expressing Ort in ort mutants using ort^C1-3-Gal4 restored, at least qualitatively, the 'on'- and 'off'-transients of the ERG, which report transmission in the lamina, as well as the optomotor behavior. These findings are consistent with the observation that ort-Gal4 drives reporter expression in lamina neurons L1-L3. In contrast, expressing Ort in lamina neurons L1 and L2 using an L1/L2-specific driver (L1L2-A-Gal4) rescued both the ERG, at least qualitatively, and optomotor defects, but not the UV preference of ort mutants. Thus, the actions of the ort-Gal4 drivers recapitulated the endogenous Ort expression pattern in the first-order interneurons of R1-R6 and R7 (Gao, 2009).

Next, whether the Ort-expressing neurons are required for UV reference and motion detection was examined. ort^C1-4-Gal4 or ort^C1-3-LexA::VP16 driving a temperature-sensitive allele of shibire, shi^ts1, so as to block synaptic transmission in specific neurons was found to significantly reduce the UV attractiveness at non-permissive, but not permissive, temperatures. This reduction was smaller than that caused by inactivating the R7s alone. These results suggest that Ort-expressing neurons might mediate both UV and green phototaxis, presumably by relaying R7 and R8 channel signals, although the existence of an ort-independent UV-sensing pathway cannot be ruled out. Similarly, inactivating Ort-expressing neurons abolished the flies' ability to detect motion. Thus, it is concluded that Ort-expressing neurons are required for both spectral preference and motion detection (Gao, 2009).

To identify the Ort-expressing neurons that could be synaptic targets of the R7 and R8 channels, a single-cell mosaic technique based on the flip-out genetic method was employed. In this system, the ort^C1-3-Gal4 flies that also carried the transgenes UAS>CD2,y+>CD8-GFP and hs-Flp were used. The flipase activity induced by brief heat-shock at the second- or third-instar larval stages excised the FLP-out cassette in small random populations of cells, thereby allowing Gal4 to drive the expression of the CD8-GFP marker. From over 1000 brain samples, 459 clones of transmedulla neurons, the projection neurons that arborize in the medulla and project axons to the lobula, were examined (see Ort is expressed in subsets of transmedulla neurons). To identify the exact medulla and lobula strata in which these processes extend, expression patterns of a series of known cell-adhesion molecules were screened, and three useful stratum-specific markers, FasIII, Connectin, and Capricious were found. In particular, anti-FasIII immunolabeled medulla and lobula strata of interest and, with MAb24B10 immunolabeling, was used primarily to identify the medulla and lobula strata. Based on the morphologies and stratum-specific locations of the arborization and axon terminals, four types of Ort-expressing projection neurons were assigned to types previously described from Golgi impregnation. These were Tm2, Tm5, Tm9, and Tm20. In addition, the ort-promoter driver labeled, albeit at a lower frequency, centrifugal neurons C2 and T2, and three types of medulla neurons with processes solely in the medulla, Dm8, other amacrine-like and also glia-like cells. All of these cells were identified multiple times in at least two independent ort-Gal4 lines, but given the sampling nature of the single-cell mosaic technique, the possibility cannot be excluded that some very rare Ort-expressing neurons were not detected. The amacrine-like and glia-like cells had not been previously described from Golgi impregnation, suggesting that there are even more classes of medulla cell types than those previously reported (Gao, 2009).

The Ort-expressing Tm neurons exhibit type-specific patterns of arborization and axon projection (see Ort-expressing Tm neurons receive multi-channel inputs in the medulla and are presynaptic at both the medulla and lobula). Tm5 neurons extend dendrite-like processes in medulla strata M3 and M6, where R8/L3 and R7 axons terminate, respectively, and they project axons to terminate in stratum Lo5 in the lobula. This pattern suggests that they relay information from the R7 and R8 or L3 channels to the lobula. The Tm5 neurons could be readily divided into three subtypes, Tm5a, b, and c, based on their unique dendritic patterns, the spread of their medulla arborization, and their gene expression patterns. Tm5a and Tm5b have medulla arborizations of different sizes and shapes; whereas Tm5c has dendritic processes in M1, in addition to strata M3 and M5, and the axon projects to both the Lo4 and Lo5 strata. The distinct morphology of Tm5c correlates with its unique expression of the vesicular glutamate transporter. Tm9 and Tm20 extend type-specific dendrite-like processes in strata M1-M3 and projected axons to distinct lobula strata. Tm20, like Tm5, projects to Lo5 while Tm9 projects to Lo1, suggesting that Tm9 and Tm20 relay information from R8 and (via lamina neurons) R1-R6, to different strata of the lobula. In medulla strata M1-M3, Tm2 extends dendrite-like processes which did not appear to make significant contacts with R7 or R8 terminals (Gao, 2009).

To determine if the Ort-expressing Tm neurons indeed receive synaptic input from photoreceptors, serial EM reconstructions were undertaken of Tm9, Tm2, and parts of a single Tm5 cell that resemble Tm5a, as well as the afferent input terminals that innervate the medulla, including R7, R8 and L1-L5 (see also Takemura, 2008). The fine dendritic arbor of Tm20 has so far eluded reconstruction. This study found that Tm9 received direct synaptic contacts from both R8 and L3 and the Tm5 received direct synaptic contacts from R7 and L3. Thus, Tm9 and Tm5 cells were postsynaptic to both the chromatic channels and an achromatic channel. Tm2, by contrast, received synaptic contacts from L2 and L4 but not, despite its Ort expression, R7 or R8. However, the possibility cannot be excluded that Tm2 responds to paracrine release of histamine from the R8 terminal, or an unidentified histamine input in the lobula (Gao, 2009).

It was reasoned that Ort-expressing neurons might be divided into several groups based on their differential release of other neurotransmitters. To test this possibility, a series of promoter-Gal4 and enhancer trap lines driving the CD8 marker was used to label neurons with glutamatergic, cholinergic, GABAergic, serotonergic and dopaminergic phenotypes in the medulla. To determine whether these neurons also express Ort, and are thus likely to receive histaminergic input, the rCD2::GFP marker was expressed in the same animals using the ort^C1-3-LexA::VP16 driver. By overlaying two expression patterns, it was found that many Ort-expressing neurons also expressed cholinergic or glutamatergic markers, while few did so for a GABAergic and none appeared to do so for serotonergic or dopaminergic phenotypes. In particular, it was found that a group of neurons labeled by both the vesicular glutamate transporter (vGlut^OK371) and ort-Gal4 drivers extended processes in the M6 stratum where R7 axons terminate, suggesting that R7's target neurons might be glutamatergic (Gao, 2009).

To identify candidate R7 target neurons, a combinatorial gene expression system, the Split-Gal4 system, was employed to restrict Gal4 activity to glutamatergic Ort-expressing neurons. In this system, ort and vGlut promoters drive expression of the Gal4DBD (Gal4 DNA binding domain-leucine zipper) and dVP16AD (a codon-optimized VP16 trans-activation domain-leucine zipper), respectively. Thus, Gal4 activity was reconstituted only in the neurons that expressed both Ort and vGlut. A dVP16AD enhancer trap vector was used, and it was substituted for the Gal4 enhancer trap in the vGlut locus. The resulting hemidriver, vGlut^OK371-dVP16AD, in combination with a general neuronal hemidriver, elav-Gal4DBD, drove expression in a pattern essentially identical to that driven by vGlut^OK371-Gal4, indicating that the vGlut^OK371-dVP16AD enhancer trap recapitulated the expression pattern of the vGlut^OK371-Gal4 driver. The combination of the vGlut^OK371-dVP16AD and ort^C1-3-Gal4DBD hemidrivers (designated ort^C1-3∩vGlut) gave rise to expression in a subset of Ort-expressing neurons in the optic lobe, namely those that express a glutamate phenotype and are thus likely to be glutamatergic. Single-cell mosaic analysis (using hs-Flp and UAS>CD2>mCD8GFP) revealed that the combinatorial ort^C1-3∩vGlut driver was expressed in Dm8, Tm5c, and L1 neurons, as well as in the medulla glia-like cells. In contrast, cha∩ort^C1-3, the combination of cha-Gal4DBD choline acetyltransferase-Gal4DBD) and ort^C1-3-Gal4AD hemidrivers, drove expression in the Ort-expressing neurons that expressed a cholinergic phenotype, including L2, Tm2, Tm9, and Tm20. Notable among these findings, L1 and L2, paired lamina neurons that receive closely matched R1-R6 input in the lamina, express different neurotransmitter phenotypes (L1: glutamate; L2: acetylcholine) (Gao, 2009).

To determine whether glutamatergic Ort-expressing neurons confer UV preference in flies, whether expressing Ort in these neurons is sufficient to restore normal UV preference in ort mutants was examined. It was found that expressing Ort using the combinatorial ort^C1-3∩vGlut driver restored normal UV preference in ort mutants. In contrast, expressing Ort in cholinergic Ort-expressing neurons using the cha∩ort^C1-3 driver further reduced UV preference, suggesting that the cholinergic Ort-expressing neurons reduce UV attraction or, more likely, enhance green attraction. Although the cha∩ort^C1-3 and ort^C1-3∩vGlut drivers were expressed in specific subsets of Ort-expressing neurons in the optic lobe, they showed additional expression outside the visual system, and expressing shi^ts1 with either driver caused non-specific motor defects at the non-permissive temperature. Although it was not possible to test whether the glutamatergic Ort-expressing neurons were required for UV preference, the rescue results indicated that the candidate glutamatergic Ort-expressing neurons, which included Dm8 and Tm5c, were involved in UV preference (Gao, 2009).

To distinguish whether Dm8 or Tm5c is required for UV preference, the ort promoter was dissected, and three promoter-Gal4 lines were generated, each of which contained one of the three highly conserved regions (C1-C3) of the ort promoter. It was found that the second and the third conserved regions (C2 and C3) gave rise to the expression in two different subsets of Ort-expressing neurons while C1 alone gave no detectable expression. Using single-cell analysis, it was found that ort^C2-Gal4 drove expression in Dm8 and L1-L3 but not in any Tm neurons, while ort^C3-Gal4 was expressed in L2, Tm2, Tm9, C2, and Mi1 neurons. All these neurons except Mi1 expressed Ort, suggesting that the C2 and C3 fragments of the ort promoter drives expression in distinct subsets of the Ort-expressing neurons, but that the combination of all conserved regions was required to suppress Ort expression in Mi1 (Gao, 2009).

Next, whether the ort^C2 or ort^C3 neuron subsets were sufficient and/or required for UV preference was examined. It was found that expressing Ort using the ort^C2-Gal4 driver in ort mutants was sufficient to restore UV preference at least up to the wild-type level. Because the lamina neurons L1 and L2 are neither necessary nor sufficient for UV preference, this finding suggested that the Dm8 neurons alone are sufficient to drive a fly's normal preference for UV. Conversely, whether these neurons were required for UV preference was tested using shi^ts1. It was found that flies carrying ort^C2->shi^ts1 exhibited strongly attenuated UV preference at the non-permissive, but not permissive, temperatures, indicating that the ort^C2 subset is required for normal UV preference. In contrast, restoring the ort^C3 subset activity further reduced UV preference, suggesting that the ort^C3 subset inhibits UV sensing, or enhances green-sensing pathways. Moreover, blocking the activity of the ort^C3 subset using shi^ts1 did not confer a stronger UV preference, suggesting that the ort^C3 subset is sufficient but likely not required for phototactic preference to green light (Gao, 2009).

The preceding evidence indicated that the two lines, ort^C2 and ort∩vGlut, together identified the Dm8 neurons both functionally and anatomically as a substrate for UV preference. To test this possibility directly, an ort^C2-Gal4^DBD hemidriver was generated and combined with the vGlut-dVP16AD hemidriver. It was found that the combinatorial driver ort^C2∩vGlut was expressed in most Dm8 neurons as well as in a small number of L1 neurons and glia-like cells. Restoring the expression of Ort in Dm8 in ort or HisCl1 ort double-null mutants completely restored normal UV preference. Conversely, flies carrying ortC2∩vGlut->shi^ts1 exhibited reduced UV preference at the non-permissive, but not permissive, temperature. Thus, the Dm8 are necessary and sufficient for a fly's normal preference for UV (Gao, 2009).

Finally, using the single-cell mosaic method the morphology of the Dm8 neurons was examined (see Amacrine Dm8 neurons receive direct synaptic input from multiple R7 neurons). In stratum M6 the Dm8 neurons were found to extend web-like processes, which extensively overlap 13-16 R7 terminals. To determine whether Dm8 receives direct synaptic input from R7, an EM marker, HRP-CD2, was examined in the Dm8 neurons using the ort^C2-Gal4 driver, and their synaptic structure was examined at the EM level. It was found that most R7 synapses are triads and that Dm8 contributes to at least one of the three postsynaptic elements in essentially all R7 synapses. Cumulatively, Dm8 contributes to ~38% (18 out of 47 identified) of the elements postsynaptic to R7s, suggesting that Dm8 is a major synaptic target for these photoreceptors. In addition, processes of three Dm8 neurons were reconstructed spanning seven medulla columns. It was found that Dm8 processes tiled the M6 stratum with partial overlapping so that each R7 terminal was presynaptic to one or two Dm8 neurons. Examining the presynaptic structures of the Dm8 neurons at EM and light microscopic levels, revealed that the Dm8 neurons were also presynaptic to small-field medulla neurons in stratum M6, including Tm5 and at a few contacts to a cell that resembles Tm9. In summary, the wide-field Dm8 neuron serves as a major target neuron for R7 input and provides output locally in stratum M6 to small-field projection neurons (Gao, 2009).

Before this study little was known about the synaptic target neurons of the R7 and R8 photoreceptors and the chromatic pathways their connection patterns subserve. This deficit reflected the inability until recently to penetrate the medulla's complexity. This study made use of prior knowledge of neurotransmitters and their receptors in the visual system to design corresponding promoter constructs that identify the first-order interneurons. These neurons were then labeled with genetically encoded markers and their morphology and synaptic connections were examined at the light and electron microscopic levels. Finally, promoter dissection and the Split-Gal4 system were combined with neurotransmitter hemidrivers to target particular neuron subtypes. It is envisioned that the same combinatorial approach can be applied to dissect other complex neural circuits (Gao, 2009).

This study identified four types of transmedulla neurons, Tm5a/b/c, Tm9, Tm20 and Tm2, that express Ort and are therefore qualified to receive direct input from R7 or R8. These Tm neurons arborize in the medulla and project axons to the lobula, suggesting that they relay spectral information from the medulla to the lobula. Supporting this interpretation, it was found that HA-syt, a presynaptic marker, is indeed localized to their terminals in the lobula. These data support previous suggestions that the lobula plays a key role in processing chromatic information for color vision. Lobula stratum 5 appears most critical for color vision because it receives all three subtypes of Tm5 neurons as well as Tm20. Moreover, it was observed that HA-syt also localized to the dendrite-like processes of all Tm neurons in the proximal medulla, suggesting the presence of presynaptic sites at this level, too. Especially, Tm5a, Tm5b, and Tm20 all extend processes with this presynaptic marker in medulla stratum M8, supporting a previous notion that this stratum might receive chromatic information (Gao, 2009).

All three subtypes of Tm5 neurons extend processes in medulla strata M6 and M3, suggesting that there they might be postsynaptic to R7 and to R8 or L3. Using serial EM, a Tm5 subtype was partially reconstructed that receives direct synaptic input from both the chromatic UV channel of R7 and the achromatic channel of L3. Serial EM also revealed that Tm9 receives inputs from the chromatic green/blue channel of R8 as well as the achromatic L3 channel. It is tempting to speculate that the Tm9 and Tm5 neurons function as color-opponent neurons by subtracting the L3-mediated luminance signal from the R7/R8 chromatic signal (see Medulla circuits in chromatic information processing). While the detailed neural mechanism must await electrophysiological studies, these anatomical data provide direct evidence that the achromatic and chromatic channels are not segregated. Instead they converge on the first/second-order interneurons, early in the visual pathway (Gao, 2009).

Using a quantitative spectral preference test, it was determined that in flies the Dm8 neurons are both necessary and sufficient to confer the animals' UV preference. Each Dm8 receives direct synaptic input from ~14 UV-sensing R7s. By pooling multiple R7 inputs, the Dm8 neurons may achieve high UV sensitivity at the cost of spatial resolution. Consistent with this notion, Dm8 is a main postsynaptic partner for R7 terminals: essentially all of R7's presynaptic sites contain at least one Dm8 postsynaptic element. The processes of Dm8 and their synapses with R7s are largely restricted to the medulla stratum M6. The stratum-specific arborization of Dm8 readily explains why R7 photoreceptors that fail to project axons to the M6 stratum are incapable of conferring UV preference (Gao, 2009).

Dm8 itself has no direct output to higher visual centers in the lobula; instead it is presynaptic to small-field projection neurons, such as Tm5 and possibly Tm9, in the medulla. Thus, Dm8 provides lateral connections linking projection neurons. The morphologies and connections of Dm8 are thus reminiscent of those made by horizontal and amacrine cells in the vertebrate retina. The vertebrate horizontal cells form reciprocal synapses with multiple cones, and in the case where the cones are of different spectral types, the horizontal cells can establish color opponency, as demonstrated in the goldfish retina. Dm8 in Drosophila receives inputs from both Rh3- and Rh4-expressing R7s, but does not provide feedback to photoreceptor terminals, suggesting that Dm8 is unlikely to contribute to color opponency, at least not in a way analogous to vertebrate horizontal cells. Vertebrate amacrine cells have diverse subtypes, which carry out very different functions, including correlating firing among ganglion cells, modulating center-surround balance of the ganglion cells and direction selectivity. The amacrine cells in vertebrate retina receive inputs from bipolar cells and provide the main synaptic input to ganglion cells. It is thus interesting to note that while direct synaptic connections from R7s to Tm5 projection neurons exists, the indirect information flow from R7, to Dm8, and then to Tm5, is both necessary and sufficient to confer UV preference, as suggested by inactivating and restoring experiments. It is hypothesized that the direct and indirect pathways function at different UV intensity levels: Dm8 pools multiple R7 inputs to detect low intensity UV in the presence of high-intensity visible light, while under high intensity UV, Tm5 receives direct input from R7 and mediates true color vision. Further studies using electrophysiology or functional imaging would be required to determine the neural mechanisms of Dm8 (Gao, 2009).

The spectral preference assay used in this study and others measure relative 'attractiveness' of UV and green light and therefore depends on the visual subsystems sensing UV and green light as well as the interactions between these subsystems. While in simple phototaxis assays, the broad-spectrum and most sensitive photoreceptors, R1-R6, dominate simple phototactic response to both UV and green light, they, as well as their first-order interneurons L1 and L2, appear to play an insignificant or redundant role in spectral preference. Thus, R8 alone, or together with R1-R6, provides the sensory input to promote green phototaxis and/or to antagonize UV attraction. The first-order interneurons that relay R8 input in this context have yet to be identified. While anatomical analysis revealed that Tm9 receives direct synaptic input from R8, the behavioral studies provided only weak and circumstantial evidence for its role in spectral preference. Expressing Ort using the cha∩ort^C1-3 or ort^C3-Gal4 driver significantly reduced UV preference in ort mutants, and Tm9 is covered by both drivers. Furthermore, inactivating Tm9 using the ort^C3 driver and shi^ts1 did not affect UV preference, suggesting that other neurons, such as Tm20, might function redundantly. Verification of these suggestions must await the isolation of Tm9-and Tm20-specific drivers, and the corresponding behavioral studies to assay the effects of perturbing activity in these neurons. It is worth noting that Ort-expressing neurons do not include any Dm8-like wide-field neurons for R8s, and restoring activity in the ort^C3 neuron subset is sufficient to confer stronger green preference in ort mutants. It is thus tempting to speculate that Dm8 circuits evolved uniquely to meet the ecological need to detect dim UV against a background of ample visible light (Gao, 2009).

Brains contain networks of interconnected neurons and so knowing the network architecture is essential for understanding brain function. This study therefore mapped the synaptic-resolution connectome of an entire insect brain (Drosophila larva) with rich behavior, including learning, value computation, and action selection, comprising 3016 neurons and 548,000 synapses.This study characterized neuron types, hubs, feedforward and feedback pathways, as well as cross-hemisphere and brain-nerve cord interactions.Pervasive multisensory and interhemispheric integration, highly recurrent architecture, abundant feedback from descending neurons, and multiple novel circuit motifs were found. The brain's most recurrent circuits comprised the input and output neurons of the learning center. Some structural features, including multilayer shortcuts and nested recurrent loops, resembled state-of-the-art deep learning architectures. The identified brain architecture provides a basis for future experimental and theoretical studies of neural circuits (Winding, 2023).

The representation and integration of internal and external cues is crucial for any organism to execute appropriate behaviors. In insects, a highly conserved region of the brain, the central complex (CX), functions in the representation of spatial information and behavioral states, as well as the transformation of this information into desired navigational commands. How does this relatively invariant structure enable the incorporation of information from the diversity of anatomical, behavioral, and ecological niches occupied by insects? This study examined the input channels to the CX in the context of their development and evolution. Insect brains develop from ~100 neuroblasts per hemisphere that divide systematically to form "lineages" of sister neurons, that project to their target neuropils along anatomically characteristic tracts. Overlaying this developmental tract information onto the recently generated Drosophila "hemibrain" connectome and integrating this information with the anatomical and physiological recording of neurons in other species, this study observe neuropil and lineage-specific innervation, connectivity, and activity profiles in CX input channels. It is posited that the proliferative potential of neuroblasts and the lineage-based architecture of information channels enable the modification, over the evolutionary time scale, of neural networks across existing, novel, and deprecated modalities in a species-specific manner, thus forming the substrate for the evolution and diversification of insect navigational circuits (Kandimilla, 2023).

Comparing connectomes can help explain how neural connectivity is related to genetics, disease, development, learning, and behavior. However, making statistical inferences about the significance and nature of differences between two networks is an open problem, and such analysis has not been extensively applied to nanoscale connectomes. This study investigated this problem via a case study on the bilateral symmetry of a larval Drosophila brain connectome. Notions of 'bilateral symmetry' were translated to generative models of the network structure of the left and right hemispheres, allowing testing and refining of understanding of symmetry. Significant differences were found in connection probabilities both across the entire left and right networks and between specific cell types. By rescaling connection probabilities or removing certain edges based on weight, adjusted definitions are presented of bilateral symmetry exhibited by this connectome. This work shows how statistical inferences from networks can inform the study of connectomes, facilitating future comparisons of neural structures (Pedigo, 2023).

This study presents a functionally relevant, quantitative characterization of the neural circuitry of Drosophila melanogaster at the mesoscopic level of neuron types as classified exclusively based on potential network connectivity. Starting from a large neuron-to-neuron brain-wide connectome of the fruit fly, stochastic block modeling and spectral graph clustering were used to group neurons together into a common 'cell class' if they connect to neurons of other classes according to the same probability distributions. Then the connectivity-based cell classes were characterized with standard neuronal biomarkers, including neurotransmitters, developmental birthtimes, morphological features, spatial embedding, and functional anatomy. Mutual information indicates that connectivity-based classification reveals aspects of neurons that are not adequately captured by traditional classification schemes. Next, using graph theoretic and random walk analyses to identify neuron classes as hubs, sources, or destinations, pathways and patterns of directional connectivity were detected that potentially underpin specific functional interactions in the Drosophila brain. A core of highly interconnected dopaminergic cell classes functioning as the backbone communication pathway for multisensory integration. Additional predicted pathways pertain to the facilitation of circadian rhythmic activity, spatial orientation, fight-or-flight response, and olfactory learning. This analysis provides experimentally testable hypotheses critically deconstructing complex brain function from organized connectomic architecture (Mehta, 2023).

This study applied community detection methods to analyze the synapse-level reconstruction of an adult female Drosophila brain containing over 20 thousand neurons and 10 million synapses. Using a machine-learning algorithm, the most densely connected communities of neurons were found by maximizing a generalized modularity density measure. The community structure was resolved at a range of scales, from large (on the order of thousands of neurons) to small (on the order of tens of neurons). The network was found to be is organized hierarchically and larger-scale communities are composed of smaller-scale structures. These methods identify well-known features of the fly brain, including its sensory pathways. Moreover, focusing on specific brain regions, it was possible to identify subnetworks with distinct connectivity types, including the fan-shaped body and the superior neuropil, with distinct clusters of upstream and downstream brain regions dividing the neuropil into several pathways. These methods show that the fine-scale, local network reconstruction made possible by modern experimental methods are sufficiently detailed to identify the organization of the brain across scales, and enable novel predictions about the structure and function of its parts (Kunin, 2023).

The fruit fly combines surprisingly sophisticated behaviour with a highly tractable nervous system. As presente in a FlyWire companion paper, available resources include the first full brain connectome of an adult animal. This study reports the systematic and hierarchical annotation of this ~130,000-neuron connectome including neuronal classes, cell types and developmental units (hemilineages). Crucially, this resource includes 4,552 cell types. 3,094 are rigorous consensus validations of cell types previously proposed in the hemibrain connectome. In addition, 1,458 new cell types are proposed, arising mostly from the fact that the FlyWire connectome spans the whole brain, whereas the hemibrain derives from a subvolume. Comparison of FlyWire and the hemibrain showed that cell type counts and strong connections were largely stable, but connection weights were surprisingly variable within and across animals. Further analysis defined simple heuristics for connectome interpretation: connections stronger than 10 unitary synapses or providing >1% of the input to a target cell are highly conserved. Some cell types showed increased variability across connectomes: the most common cell type in the mushroom body, required for learning and memory, is almost twice as numerous in FlyWire as the hemibrain. Evidence was found for functional homeostasis through adjustments of the absolute amount of excitatory input while maintaining the excitation-inhibition ratio. Finally, and surprisingly, about one third of the cell types proposed in the hemibrain connectome could not yet be reliably identified in the FlyWire connectome. It is therefore suggested that cell types should be defined to be robust to inter-individual variation, namely as groups of cells that are quantitatively more similar to cells in a different brain than to any other cell in the same brain. Joint analysis of the FlyWire and hemibrain connectomes demonstrates the viability and utility of this new definition. This work defines a consensus cell type atlas for the fly brain and provides both an intellectual framework and open source toolchain for brain-scale comparative connectomics (Schlegel, 2023).

Schlegel, P., Yin, Y., Bates, A. S., ..., Hartenstein, V., Bock, D. D. and Jefferis, G. (2023). Whole-brain annotation and multi-connectome cell typing quantifies circuit stereotypy in Drosophila. bioRxiv. PubMed ID: 37425808

The fruit fly combines surprisingly sophisticated behaviour with a highly tractable nervous system. As presente in a FlyWire companion paper, available resources include the first full brain connectome of an adult animal. This study reports the systematic and hierarchical annotation of this ~130,000-neuron connectome including neuronal classes, cell types and developmental units (hemilineages). Crucially, this resource includes 4,552 cell types. 3,094 are rigorous consensus validations of cell types previously proposed in the hemibrain connectome. In addition, 1,458 new cell types are proposed, arising mostly from the fact that the FlyWire connectome spans the whole brain, whereas the hemibrain derives from a subvolume. Comparison of FlyWire and the hemibrain showed that cell type counts and strong connections were largely stable, but connection weights were surprisingly variable within and across animals. Further analysis defined simple heuristics for connectome interpretation: connections stronger than 10 unitary synapses or providing >1% of the input to a target cell are highly conserved. Some cell types showed increased variability across connectomes: the most common cell type in the mushroom body, required for learning and memory, is almost twice as numerous in FlyWire as the hemibrain. Evidence was found for functional homeostasis through adjustments of the absolute amount of excitatory input while maintaining the excitation-inhibition ratio. Finally, and surprisingly, about one third of the cell types proposed in the hemibrain connectome could not yet be reliably identified in the FlyWire connectome. It is therefore suggested that cell types should be defined to be robust to inter-individual variation, namely as groups of cells that are quantitatively more similar to cells in a different brain than to any other cell in the same brain. Joint analysis of the FlyWire and hemibrain connectomes demonstrates the viability and utility of this new definition. This work defines a consensus cell type atlas for the fly brain and provides both an intellectual framework and open source toolchain for brain-scale comparative connectomics (Schlegel, 2023).

Knowing one's own behavioral state has long been theorized as critical for contextualizing dynamic sensory cues and identifying appropriate future behaviors. Ascending neurons (ANs) in the motor system that project to the brain are well positioned to provide such behavioral state signals. However, what ANs encode and where they convey these signals remains largely unknown. Through large-scale functional imaging in behaving animals and morphological quantification, this study reports the behavioral encoding and brain targeting of hundreds of genetically identifiable ANs in the adult fly, Drosophila melanogaster. It is revealed that ANs encode behavioral states, specifically conveying self-motion to the anterior ventrolateral protocerebrum, an integrative sensory hub, as well as discrete actions to the gnathal ganglia, a locus for action selection. Additionally, AN projection patterns within the motor system are predictive of their encoding. Thus, ascending populations are well poised to inform distinct brain hubs of self-motion and ongoing behaviors and may provide an important substrate for computations that are required for adaptive behavior (Chen, 2023).

Whereas progress has been made in the identification of neural signals related to rapid, cued decisions, less is known about how brains guide and terminate more ethologically relevant decisions in which an animal's own behaviour governs the options experienced over minutes. Drosophila search for many seconds to minutes for egg-laying sites with high relative value and have neurons, called oviDNs, whose activity fulfills necessity and sufficiency criteria for initiating the egg-deposition motor programme This study shows that oviDNs express a calcium signal that (1) dips when an egg is internally prepared (ovulated), (2) drifts up and down over seconds to minutes-in a manner influenced by the relative value of substrates-as a fly determines whether to lay an egg and (3) reaches a consistent peak level just before the abdomen bend for egg deposition. This signal is apparent in the cell bodies of oviDNs in the brain and it probably reflects a behaviourally relevant rise-to-threshold process in the ventral nerve cord, where the synaptic terminals of oviDNs are located and where their output can influence behaviour. Perturbational evidence is provided that the egg-deposition motor programme is initiated once this process hits a threshold and that subthreshold variation in this process regulates the time spent considering options and, ultimately, the choice taken. Finally, a small recurrent circuit was found that feeds into oviDNs, and that activity in each of its constituent cell types was shown to be required for laying an egg. These results argue that a rise-to-threshold process regulates a relative-value, self-paced decision and provide initial insight into the underlying circuit mechanism for building this process (Vijayan, 2023).

Although males and females largely share the same genome and nervous system, they differ profoundly in reproductive investments and require distinct behavioral, morphological, and physiological adaptations. How can the nervous system, while bound by both developmental and biophysical constraints, produce these sex differences in behavior? This study uncovered a novel dimorphism in Drosophila melanogaster that allows deployment of completely different behavioral repertoires in males and females with minimum changes to circuit architecture. Sexual differentiation of only a small number of higher order neurons in the brain leads to a change in connectivity related to the primary reproductive needs of both sexes-courtship pursuit in males and communal oviposition in females. This study explains how an apparently similar brain generates distinct behavioral repertoires in the two sexes and presents a fundamental principle of neural circuit organization that may be extended to other species (Nojima, 2021).

Sexually reproducing species exhibit sex differences in social interactions to boost reproductive success and survival of progeny. Comparing and contrasting the anatomy, activity, and function of sexually dimorphic neurons in the brain of males and females across taxa are starting to reveal the fundamental principles of neural circuit organization underlying these sex differences in behavior. A variety of alternative neuronal circuit configurations have been proposed to generate sexually dimorphic behaviors. Many studies have identified sex differences in sensory inputs in various species; however, such differences in higher order brain circuits that organize species- and sex-specific instinctive behaviors in response to sensory cues are still poorly characterized (Nojima, 2021).

Sex is determined early in an animal's development and initiates many irreversible sexual differentiation events that influence how the genome and the environment interact to give rise to sex-specific morphology and behavior. In Drosophila, selective expression of two sex determination transcription factors (TFs), Doublesex (Dsx) and Fruitless (Fru), define cell-type-specific developmental programs that govern functional connectivity and lay the foundations through which innate sexual behaviors are genetically predetermined. Because both fru- and dsx-expressing neurons are essential for male and female reproductive behaviors, studies in the adult have focused on neurons that express these TFs to identify anatomical or molecular sex differences in neuronal populations. This allows entry to the neural circuits underlying sex-typical behaviors and identification of the neuronal nodes that control component behaviors and behavioral sequencing (Nojima, 2021).

Dsx proteins, which are part of the structurally and functionally conserved Doublesex and Male-abnormal-3 Related Transcription factors (DMRT) protein family, are critical for sex-specific differentiation throughout the animal kingdom. In the insect phylum, Dsx proteins act at the interface between sex determination and sexual differentiation, regulating a myriad of somatic sexual differences both inside and outside the nervous system. The dsx gene has functions in both sexes: its transcripts undergo sex-specific alternative splicing to encode either a male- or female-specific isoform. dsx expression is highly regulated in both male and female flies, as shown by its temporally and spatially restricted expression patterns through development, with only a select group of neurons expressing dsx. The dsx gene is expressed in some 150 and 30-40 neurons per hemisphere in the male and female brains, which reside in 10 and 7 to 8 discrete anatomical clusters, respectively. This restricted expression of dsx in higher order neurons in the brain suggests these neurons may act as key sex-specific processing nodes of sensory information (Nojima, 2021).

To study the fundamental principles of neural circuit organization underlying sex differences in behavior, this study identified and mapped dsx⁺ sexual dimorphisms in the CNS. This analyses revealed that all dsx⁺ clusters are either sexually dimorphic or sex specific; none are sexually monomorphic. To examine higher order processing differences between the sexes, this study focused on the dsx⁺ anterior dorsal neuron (aDN) cluster, as it is present in both sexes yet has sexually dimorphic dendritic arborizations associated with sensory perception. These anatomical differences lead to sex-specific connectivity, with male aDN inputs being exclusively visual, while female inputs are primarily olfactory. Finally, this study shows that this unique sexually dimorphic neuronal hub that reroutes distinct sensory pathways gives rise to functionally distinct social behaviors between the sexes: visual tracking during courtship in males and communal egg-laying site selection in females (Nojima, 2021).

This study identified a small cluster of two neurons per hemisphere in the central brain, which reconfigures circuit logic in a sex-specific manner. Perhaps most surprising is the seemingly unrelated behaviors these equivalent neurons control in each sex-visual tracking during courtship in males and communal egg laying in females. Ultimately, these circuit reconfigurations lead to the same end result-an increase in reproductive success. These findings highlight a flexible strategy used to structure the nervous system, where relatively minor modifications in neuronal networks allow each sex to respond to their social environment in a sex-appropriate manner (Nojima, 2021).

The behavioral function of the male aDN cluster appears to be related to visual aspects of courtship behavior. A set of visual projection neurons, LC10a, was previously identified as involved in tracking and following behaviors in the male during courtship; however, no apparent sex differences in their anatomy or their physiological responses to visual stimuli were detected. It would seem these sex differences in behavior arise from the sex-specific downstream connectivity of LC10a neurons in the central brain. This study identified aDNs connecting downstream to LC10a in males only. aDN inactivation mirrors visual tracking defects displayed upon LC10a inactivation; therefore, the male aDN cluster confers sex specificity to visually guided tracking of females during courtship (Nojima, 2021).

This study also identified AL5a neurons to be downstream of LC10a in both sexes. Interestingly, it has been reported that AL5a is likely upstream of the fru⁺ cluster Lv2/pIP-b/pIP8 thought to exchange and integrate visual information from the right and left hemispheres of the brain. This male-specific connectivity is compatible with a potential role for AL5a in mediating visual information necessary for wing choice during courtship, a behavior these neurons have been shown to elicit when activated (Nojima, 2021).

The two LC10a downstream clusters that this study identified, aDN and AL5a, also show differences in their anatomical connectivity and physiological responses. Whereas AL5a is downstream of LC10a in both sexes, aDN is only connected to LC10a in the male. Despite direct anatomical connectivity between LC10a and aDN in males, functional connectivity was only uncovered under conditions of pharmacological disinhibition. This observation might hint at inhibitory modulation of aDN that depends on the male's internal state, e.g., his mating drive, or additional cues that influence his courtship arousal. A previous study found that, in sexually satiated males, calcium responses in courtship 'decision-making' P1 neurons were absent when stimulating upstream neurons but could be restored to the levels observed in naive males by application of PTX. It is tempting to speculate that inhibition in the LC10a -> aDN pathway is similarly linked to sexual arousal. In contrast, AL5a responses to LC10a stimulation occurred in the absence of PTX and were markedly larger in AL5a than in aDN. The variation in calcium signals could be due to the considerable difference in cell numbers comprising each cluster (2 aDN versus 24 AL5a) or due to inputs from different AOTu regions. aDNs sample from the whole glomerulus region, whereas the AL5a cluster is restricted to the dorsal part of the AOTu, suggesting they extract information from broad versus specific parts of the visual field, respectively. Future investigation will be aimed at linking the clusters' anatomical differences with their differential processing of visual information to facilitate distinct behavioral roles (Nojima, 2021).

In females, the aDN cluster does not receive visual information but appears to sample from a range of sensory modalities, with information received via the antennal lobe dominating its inputs, suggesting its involvement in a complex behavior requiring multisensory integration. One such behavior is female egg-laying site selection, which is critical to the success of offspring. For Drosophila, offspring survival rates depend on the selection of oviposition sites that are shared with conspecifics, a process known to rely on olfaction (Nojima, 2021).

This study has shown that aDNs are highly integrated into circuitry known to regulate oviposition. The excitatory oviEN, which is anatomically similar to the aDNs, responds to information about substrate suitability via gustatory and mechanosensory cues in the legs and directly influences aDN output. Silencing oviEN function suppresses egg laying itself, whereas silencing aDN does not affect the overall number of eggs laid. Instead, aDN-silenced females are no longer able to show a preference to lay eggs communally, losing a female-specific social behavior essential for offspring survival. While both oviEN and aDN output directly onto the oviposition motor program (through oviDNs), oviENs are the largest contributors to oviDN dendritic budgets, with aDN being relatively minor contributors. Thus, the aDN cluster acts as a modulator of egg laying choice, whereas the oviEN more generally affects the mechanics of egg laying (Nojima, 2021).

As the oviposition of fertilized eggs is a female behavior that can only be displayed after mating, the behavioral programs required are likely inhibited in virgin females. The activity of the inhibitory neuron oviIN depends on female mating status and thus appears to act as a general inhibitor of egg-laying circuitry in virgin females. oviINs form axo-axonic synapses with both the aDN and oviEN, suggesting they gate their outputs by presynaptic inhibition in a state-dependent manner. Intriguingly, as both oviEN and oviIN form axo-axonic synapses with aDN, this suggests a potential gating mechanism by which their relative strengths inhibit or facilitate output from aDN onto downstream targets (Nojima, 2021).

Consistent with aDNs' behavioral function in egg-laying site selection, a female post-mating behavior, this study found differences in the aDN physiological responses in mated versus virgin females. Stimulation of OSNs resulted in significantly stronger aDN calcium responses in mated females compared to virgins. This finding might hint at a state-dependent inhibition of olfactory inputs into aDN in females, potentially analogous to the inhibition of visual inputs to aDN observed in males. The difference in physiological responses between mated and virgin females was not observed when stimulating PNs, which are downstream of OSNs but upstream of aDN. There are different possible explanations for this discrepancy, including differences in the populations of neurons targeted by the driver lines used to target PNs versus OSNs or inhibition in virgin females occurring at the level of OSN to PN connectivity; therefore, activating PNs directly bypasses the state-dependent inhibition. In addition to state-dependent effects, there also seemed to be differences in the calcium responses in different neuronal compartments. This finding could be explained by the position of the input synapses of different upstream neurons into the aDN (e.g., dendritic versus axonic). The exact mechanism of how aDN integrates these different inputs and transforms them into an output that guides egg-laying site selection remains to be examined (Nojima, 2021).

The principal output of the female aDN is the previously undescribed SMP156 neuron, which itself outputs primarily in the IB, where its axons show cross-hemisphere connectivity, suggesting it acts as integrators of sensory information from different directions. The major SMP156 output neuron type (IB011) projects to the lobula in the opposite hemisphere, potentially integrating olfactory and visual information as observed in other flying insects during pheromone orientation. Olfactory navigation requires comparisons of left and right inputs, e.g., when male moths orient themselves toward conspecific females in response to sex pheromones. Determination of position and direction applies to males pursuing females and females following pheromonal cues to locate a communal egg-laying site. It is proposed that the aDN cluster in females selectively integrates sensory information, relaying it to SMP156, which confers directionality and processes information relevant to locating an appropriate egg-laying site. In the absence of a male connectome for comparison, it can only be speculated about potential shared downstream connectivity. As the male aDN output sites are mainly overlapping with female sites in the SMP, it is possible that the male visual pathway also inputs into SMP156, or a similar neuron associated with the IB, potentially feeding back onto visual pathways, supporting appropriate tracking of the female. A male connectome and more genetic tools will help reveal the full extent of downstream functional connectivity and convergence between the sexes (Nojima, 2021).

As fundamental features of most animal species, sexual dimorphisms and sex differences have particular importance for the function of the nervous system. These innate sex-specific adaptations are built during development and orchestrate interactions between sensory information and specific brain regions to shape the phenotype, including the emergent properties of the sex-specific neural circuitry. Evolutionary forces acting on these neural systems have generated adaptive sex differences in behavior. In Drosophila, males compete for a mate through courtship displays, while a female's investment is focused on the success of their offspring. These sex-specific behaviors are guided by the perception and processing of sensory cues, ensuring responses lead to reproductive success. This study has shown how a sex-specific switch between visual and olfactory inputs underlies adaptive sex differences in behavior and provides insight on how similar mechanisms may be implemented in the brains of other sexually dimorphic species (Nojima, 2021).

The Drosophila giant fiber (GF) escape circuit is an extensively studied model for neuron connectivity and function. Researchers have long taken advantage of the simple linear neuronal pathway, which begins at peripheral sensory modalities, travels through the central GF interneuron (GFI) to motor neurons, and terminates on wing/leg muscles. This circuit is more complex than it seems, however, as there exists a complex web of coupled neurons connected to the GFI that widely innervates the thoracic ganglion. This study defines four new neuron clusters dye coupled to the central GFI, which were named GF coupled (GFC) 1-4. New transgenic Gal4 drivers were identified that express specifically in these neurons, and both neuronal architecture and synaptic polarity were mapped. GFC1-4 share a central site of GFI connectivity, the inframedial bridge, where the neurons each form electrical synapses. Targeted apoptotic ablation of GFC1 reveals a key role for the proper development of the GF circuit, including the maintenance of GFI connectivity with upstream and downstream synaptic partners. GFC1 ablation frequently results in the loss of one GFI, which is always compensated for by contralateral innervation from a branch of the persisting GFI axon. Overall, this work reveals extensively coupled interconnectivity within the GF circuit, and the requirement of coupled neurons for circuit development. Identification of this large population of electrically coupled neurons in this classic model, and the ability to genetically manipulate these electrically synapsed neurons, expands the GF system capabilities for the nuanced, sophisticated circuit dissection necessary for deeper investigations into brain formation (Kennedy, 2018).