Synaptic connectome of the Drosophila circadian clock

The master clock in the Drosophila brain is widely regarded to be comprised of approximately 150 neurons. This number is derived from neuronal expression of different clock genes¹⁴. Clk856-Gal4, based on the Clk promoter, faithfully recapitulates expression in most of these clock neurons (Fig. 1A). These neurons can be broadly classified into four classes each of LN and DN (Fig. 1B and Table 1). The LN comprise Lateral Posterior Neurons (LPN), dorsoLateral Neurons (LN_d), Ion Transport Peptide (ITP)-expressing LN (LN^ITP), and Pigment-Dispersing Factor (PDF)-expressing ventroLateral Neurons (LN_v^PDF). Conversely, the DN include anterior Dorsal Neurons 1 (DN_1a), posterior DN₁ (DN_1p), DN₂ and DN₃. These clock neuron classes can be further subdivided into different cell types (Fig. 1C and Table 1). Morphological characterization of some DN_1p and DN₃ subtypes is currently lacking¹⁸. Moreover, the estimated number of clock neurons in Drosophila is likely an underestimation as the precise number of DN₃ has not been determined thus far²⁵. This is partly because drivers like Clk856-Gal4 only include a small proportion of DN₃¹⁴. In addition, most DN₃ have small somata that are densely packed together, making it difficult to count. The majority of these small DN₃ project to the central brain, and hence they are aptly called small Central Projecting DN₃ (s-CPDN₃) (Table 1)²⁶. Similarly, a pair of DN₃ with large somata project to the central brain (large Central Projecting DN₃; l-CPDN₃), whereas about 6 neurons per brain hemisphere project to the anterior brain (Anterior Projecting DN₃; APDN₃)²⁶. These latter cells also have larger somata than the s-CPDN₃¹⁸. Despite the large number of DN₃, previous studies suggest that they are less important for behavioral rhythmicity compared to some of the LN, which are regarded as the master pacemaker neurons^27–29.

As a first step in determining the synaptic connectivity within the clock network, we used a systematic approach (Supplementary Fig.1) to identify most, if not all, of the clock neurons in the FlyWire connectome generated in our companion papers^9,10. In total, we successfully identified 242 clock neurons based on a combination of morphology, previously determined connectivity, and location of their cell soma (Fig. 1C, D, Supplementary Movies 1, 2, Supplementary Data 1)^{17–19,26,30}. The number of clock neurons that we identified in the connectome was considerably higher than the expected number of clock neurons (~152) in the adult brain (Table 1). This was mainly due to the presence of more s-CPDN₃ in the connectome than estimated, which prompted us to accurately quantify the total number of DN₃ in adult brains. To comprehensively quantify DN₃, we used antibodies against clock proteins Period (PER) and Vrille (VRI) in tim-Gal4 > GFP expressing flies. Since PER and VRI exhibit peak expression at different Zeitgeber times (Fig. 2A), the inclusion of three clock markers (PER, VRI, and timeless (tim)) can provide a reliable estimate of the total number of DN₃ (Fig. 2B, C, Supplementary Movie 3). Indeed, our analyses revealed that there are more than 70 DN₃ per hemisphere in both males and females (Fig. 2D). Thus, the number of DN₃ identified in the connectome (171 neurons) is in line with our DN₃ estimate (~166 neurons) based on anatomical analysis. The majority of these are s-CPDN₃ (157 neurons) which we now classified into five subtypes (s-CPDN₃A-E) based on morphological similarity (Fig. 2E–I). These five DN₃ subtypes are comprised of 26 distinct cell types^9,10 and could thus be subdivided further in the future. During the revision of this manuscript, another study reported the existence of 12 DN₃ clusters/subtypes based on single-cell RNA sequencing³¹, further supporting our analyses.

Additionally, we also identified several candidate DN_1p which could not be classified into previously characterized DN_1p cell types (Table 1). To verify if these neurons identified in the connectome are indeed DN_1p, we characterized them anatomically. Specifically, we performed multi-color flip-out (MCFO) analysis of DN_1p using Clk4.1M-Gal4 to elucidate the morphology of this densely packed neuronal cluster. In total, about 14 DN_1p per hemisphere are labeled with the Clk4.1M-Gal4 (Fig. 2J). MCFO analysis revealed DN_1pA (4-5 neurons) and DN_1pB (2 neurons) subtypes characterized previously (Table 1 and Fig. 2K). Our analysis additionally revealed the morphology of previously uncharacterized DN_1p subtypes (Fig. 2L–N). DN_1pC and DN_1pE are each comprised of two neurons per hemisphere (Fig. 2L, N). DN_1pD is comprised of four neurons per hemisphere (Fig. 2M). Two of these project over the midline while the other two remain ipsilateral, suggesting that the DN_1pD is comprised of two morphologically distinct subtypes. Notably, our analysis confirmed that the candidate DN_1p identified from the connectome are in fact morphologically similar to DN_1pC-E subtypes labeled by Clk4.1M-Gal4. Taken together, our identification and morphological characterization of novel DN₃ and DN_1p subtypes provide a solid framework to comprehensively examine the connectivity of clock neurons.

The FlyWire connectome combines automatically detected chemical synapses with proofread neurons. These synapses represent an additional anatomical feature that could potentially distinguish neuronal groups. Consequently, we asked whether the classification of clock neurons based on differences in their synaptic connectivity aligns with the traditional anatomical and recent gene expression-based classification. To address this, we clustered clock neurons based on cosine similarity between their total synaptic inputs and outputs. Our clustering analysis shows that neurons of a given clock cell type (Table 1) usually cluster together, suggesting that neurons from the same group are more similar (in terms of synaptic connectivity) to each other than to other clock neurons (Supplementary Fig. 2A, B). For example, all three LN_d^CRY- from one hemisphere are part of the same clade. Similarly, the two DN₂ are part of a clade. Exceptions to this are the clades containing DN_1p and DN₃. For DN_1p, this can be explained by our findings which show that this group comprises five morphologically distinct subtypes. In the case of DN₃, while some of them form their own cluster, other DN₃ cluster together with different clock neuron subtypes. On one hand, this is not unexpected since it is unlikely for such a large group of neurons to have similar connectivity patterns. On the other hand, this is quite interesting since it provides insights into their possible function. For instance, heterogenous clusters comprising clock neurons of different classes have similar synaptic inputs and outputs, and may thus play similar roles in the clock network and beyond. Moreover, the connectivity-based clustering does not resolve the five subtypes of the s-CPDN₃. This suggests that these five s-CPDN₃ subtypes comprise synaptically heterogeneous cell populations. Nonetheless, synaptic connectivity-based classification of clock neurons largely aligns with the ones determined based on anatomical and gene expression differences.

Having identified all the clock neurons, we next sought to determine their synaptic interconnectivity which could facilitate intercellular coupling within the network. Generally, we regarded >4 common synapses per neuron as significant connections and >9 synapses as strong connections. We first wanted to validate our analysis by comparing it with previously reported connections. In agreement with previous reports^18,19, we observed strong synaptic connectivity from DN_1a to LN^ITP, from DN_1pA to LN^ITP and LN_d, and from DN_1pB to DN₂ clusters (Fig. 1E and Supplementary Fig.3), highlighting the robustness of our approach. Importantly, our analysis also uncovered novel connections between the different subgroups. Specifically, we observed strong contralateral and ipsilateral connectivity from DN_1pA to LN^ITP, as well as additional significant connections with s-CPDN₃C, s-CPDN₃D, and LN_d^CRY+ across both hemispheres. Similarly, LN^ITP also provide synaptic inputs to s-CPDN₃ clusters in both hemispheres (Fig. 1E and Supplementary Fig.3). This raised the question of whether DN_1pA represents a heterogeneous population where one subgroup forms ipsilateral connections and the other contralateral. To address this, we examined their connectivity at cellular resolution (Supplementary Fig. 4,5) which revealed that individual DN_1pA indeed form both ipsilateral and contralateral connections (Fig. 1F). Our analysis thus identified DN_1pA as an important center which links the clock network across the two brain hemispheres. These results are in line with previous reports of contralateral projections of DN_1p³². In contrast, there are virtually no synaptic connections between the s-LN_v and the LN_d neurons (Fig. 1E and Supplementary Fig.3-5), which control morning and evening activities, respectively. This is consistent with previous analyses using the hemibrain connectome¹⁹.

To assess the extent of inter-individual differences in the numbers, neuronal projections, and synaptic connectivity of clock neurons, we next performed comparisons with the partial hemibrain connectome³³. Several groups of clock neurons were previously identified in the hemibrain connectome including all s-LN_v, l-LN_v, LN_d, LN^ITP, LPN, DN_1a, and some DN₂, DN_1p, and DN₃^17–19. Here, we identified additional DN_1p and DN₃ (Supplementary Fig.6A–D). In total, 64 clock neurons can be identified in the hemibrain connectome, with the majority of missing neurons belonging to the DN₃ subgroups (Supplementary Fig.6A–D, Supplementary Data 2). Comparison of different subgroups revealed stable neuronal numbers across the two connectomes (Table 1 and Supplementary Fig.6D). Similarly, there is a high degree of stereotypy in the connectivity between the clock clusters (Supplementary Fig.6E, F). For instance, l-LN_v and DN₂ form the least synaptic contacts with other clock clusters. At the opposite end of the spectrum, s-CPDN₃A are connected to all the clock clusters except for s-LN_v, l-LN_v, DN_1a, and DN_1pE. Given its partial nature, the hemibrain connectome lacks information about all contralateral connections, reiterating the significance of characterizing information flow across entire networks. Taken together, our analyses revealed hitherto unknown connectivity between the clock neurons which could contribute to the robustness of the master clock. Moreover, the identification of the complete circadian neuronal network in the FlyWire connectome underscores the power of the fruit fly in pushing forward the frontier of our understanding of chronobiology.

While the FlyWire and hemibrain connectomes exhibit a high degree of stereotypy, we further used an independent approach to validate our connectivity analyses. We performed light microscopy-based trans-synaptic circuit tracing by expressing trans-Tango³⁴ using specific driver lines for different populations of clock neurons (Supplementary Figs. 7,8). Upon driving trans-Tango with Clk4.1M-Gal4 which labels most DN_1p (Supplementary Fig. 7B), we observed post-synaptic signals in DN_1p, DN₂, DN₃, LPN, LN_d, 5th-LN_v, and s-LN_v (Supplementary Fig. 9A). However, l-LN_v and DN_1a were not post-synaptic to DN_1p. Thus, our trans-Tango analysis of DN_1p agrees with the connectivity of DN_1p based on the connectomes (Supplementary Fig. 3 and Supplementary Fig. 6F). Similarly, a split-Gal4 line targeting DN₃ drives post-synaptic signals in DN_1a, DN_1p, DN₂, DN₃, LPN, LN_d, and l-LN_v (Supplementary Fig. 9B), which mirrors the connectivity seen in the connectomes. While post-synaptic signal was not detected in most clock neurons of control flies, occasionally, a false post-synaptic signal was detected in two clock neurons (LN_d) from the entire network (Supplementary Fig. 10A). Hence, any potential synaptic output to LN_d should be treated with caution. Overall, we observed similar congruency between the two approaches with other Gal4-lines including those targeting DN₂ (Supplementary Fig.10B), LPN (Supplementary Fig. 10C), and LN^ITP (Supplementary Fig.10D).

Differences compared to the connectomes were observed when driving trans-Tango with Pdf-Gal4 (for s-LN_v and l-LN_v) (Supplementary Fig. 10E) and with the DN_1a-specific split-Gal4 line (Supplementary Fig.10F). In both cases, trans-Tango generated post-synaptic signals in more clock neurons than anticipated based on the connectomes. This discrepancy could be explained by: (1) the presence of additional neurons in the Gal4-line (e.g. PDF tritocerebrum neurons³⁵), (2) daily remodeling of neural circuits, as shown previously for s-LN_v and DN_1a^36,37 and/or (3) connections persisting through development. In summary, our trans-Tango analysis is largely in agreement with the clock network generated using the connectomes.

Following the successful validation of our connectivity data, we next identified all the major classes of neurons providing inputs to the clock network. To this end, we utilized the annotation scheme of our companion paper⁹, which provides a hierarchical classification of all neurons in the connectome (Fig. 3A). We found that neurons intrinsic to the brain provide the majority of the inputs to the clock network (Fig. 3A–C). This includes visual centrifugal neurons projecting from the central brain to the optic lobes, visual projection neurons projecting from the optic lobes to the central brain, as well neurons intrinsic to the optic lobes and central brain (Fig. 3C). Examining inputs to specific clock clusters, we observed differential inputs across all the subgroups (Fig. 3B). As expected, s-LN_v and l-LN_v receive most of their input from optic lobe and visual centrifugal neurons as they have a large number of input synapses in the optic lobes and the accessory medulla (AME) (Supplementary Fig. 11). In contrast, APDN₃, l-CPDN₃, and LN^ITP populations receive major inputs from visual projection neurons. The remaining clock clusters receive most of their inputs from central brain neurons (Fig. 3B, Supplementary Figs.11, and 12). In some cases, such as DN_1pC-E, DN₂, and s-CPDN₃A-E, a significant portion of these central neurons are clock neurons themselves (Supplementary Fig.13), confirming prominent intercellular synaptic connectivity between some clock clusters. Interestingly, only 4 sensory neurons provide direct inputs to the clock network. These are anterior cells (aDT4) (Dr. Gregory Jefferis, personal communication) (Fig. 3A, D) which provide temperature inputs to LPN, DN_1pC, and DN_1pE^38,39. Therefore, most of the strong thermosensory input to clock neurons is indirect via thermosensory projection neurons⁴⁰.

Having broadly classified the inputs from different neuronal super classes to the clock network, we probed further and identified individual cells providing the strongest synaptic inputs to clock neurons. For this purpose, we used a stringent threshold of 80 synapses to obtain a narrow list of candidate inputs. Our analysis discovered 13 neurons, including 7 aMe neurons (a subset of visual projection and visual centrifugal neurons) that are strongly connected to specific downstream clock neurons (Fig. 3E). For example, individual aMe3 and aMe6a neurons can form more than 79 synapses per neuron with APDN₃, while aMe8 are similarly connected to LN_d^{CRY+ & ITP} clock neurons (Fig. 3E). The unifying feature of these aMe neurons is their dense arborization in the AME and posterior lateral protocerebrum (Fig. 3E), where they anatomically interact with clock neuron dendrites (Supplementary Figs. 11 and 12)^18,41. Interestingly, the aMe neurons themselves receive strong inputs from the extraretinal photoreceptors. Specifically, aMe3 and aMe6a neurons receive strong inputs directly from the Hofbauer-Buchner (HB) eyelets (Fig. 3E). Conversely, aMe8 receive indirect inputs from ocellar retinula cells via the ocellar ganglion neurons (OCG) type 2c (OCG02c) (Fig. 3E). Therefore, the clock receives strong light inputs from extrinsic photoreceptor cells, albeit indirectly. This is not surprising since light is the most important Zeitgeber for circadian clocks⁴². Flies synchronize their circadian clocks with the light-dark cycles using these extrinsic photoreceptor cells as well as via the blue-light photoreceptor Cryptochrome (CRY), which is expressed in about half of the clock neurons (Fig. 3F)⁴². While CRY interacts with the core clock protein Timeless and can quickly reset the clock, the different photoreceptor cells are important for sensing dawn, dusk, high light intensities, and day length, and for adapting morning and evening activities to the appropriate time of day. Regardless, we found little direct inputs from the photoreceptor cells and other sensory cells to the clock neurons (Fig. 3A). This is consistent with previous findings which revealed that most of the light input to the clock appears to be indirect⁴³. In light of this and our in-silico circuit tracing analysis described above, we comprehensively characterized indirect connectivity between photoreceptor cells and clock neurons. Therefore, we traced all the disynaptic connections between them. Using the normal threshold of >4 synapses, we again recovered the strong connections from the H-B eyelets via the aMe3/aMe6a to the APDN₃, and additional weaker connections to the s-CPDN₃A, LN^ITP, l-LN_v, and LN_d^CRY+ (Fig. 3G). Furthermore, we revealed connections from R7/R8 compound eye photoreceptors to several clock neurons via aMe12⁴⁴ and other interneurons. While we did not observe any disynaptic connections from the ocellar retinula cells to clock neurons using the normal threshold (Fig. 3G), reducing the threshold to >2 synapses revealed connections from the ocelli to APDN₃ and l-CPDN₃ via OCG02c (Fig. 3H). The synaptic connections from the ocelli to OCG and beyond are extensively characterized in our companion paper¹⁰ and demonstrate interesting details that may also be valid for the other photoreceptor inputs to clock neurons. The majority of ocellar photoreceptors are synaptically connected to ocellar ganglion neurons with thick axons (OCG01a-f) or directly to descending neurons (DNp28). These connections likely enable fast behavioral responses. In contrast, axons of OCG02c that connect to the clock neurons are rather thin and not suited for fast neurotransmission. Instead, these neurons appear suited for collecting light information over time – a property needed for entraining the circadian clock. Further, collecting light information over larger time intervals may not require a high synapse density. Thus, 3 to 4 synapses between retinula cells and the relevant downstream OCG observed here could be sufficient for this purpose (Fig. 3H). The same is also true for the photoreceptor cells of the compound eyes. Reducing the threshold of significant connections from 5 to 3 synapses revealed indirect clock input from additional photoreceptor cells, including those that project from the dorsal rim area of the eye (Fig. 3H). These photoreceptor cells are involved in polarized vision and might contribute to time-compensated sun compass orientation⁴⁵. Whether the connectivity observed with a lower threshold of >2 synapses is functional in vivo remains to be seen; however, this is very likely since there are usually many photoreceptor cells that synapse onto only a few aMe neurons. For example, theoretically, the ~300 pale R8 cells project to only 3–4 aMe12 neurons⁴⁴, resulting in ~100 connections on average per aMe neuron. Even if each of these connections were mediated via only 3 synapses, each aMe neuron could potentially receive inputs from R8 cells via 300 synapses, which is quite substantial (Fig. 3H).

Delineating the output pathways that translate daily 24-hour oscillations of the molecular clock into physiological and behavioral rhythms remains a major focus in chronobiology. Using the same strategy as above to identify the inputs, we systematically classified all the neurons downstream of the clock network. Most synaptic output from the clock network is directed to intrinsic brain neurons, and in particular, the central brain neurons (Fig. 4A–C). Except for l-LN_v, all clock clusters have a majority of their output onto central brain neurons. l-LN_v mostly provide inputs to Medullary intrinsic neurons in the optic lobe (Mi 1, Fig. 4B, C), consistent with their role in adapting the sensitivity of the visual system to the time of day^46,47. Further, examination of individual postsynaptic partners of clock neurons (Supplementary Fig. 14) reveals that the majority of the output from DN_1pA is onto other clock neurons, namely LN^ITP, LN_d^CRY+, and s-CPDN₃C-D (Fig. 4B and D). s-CPDN₃C-D, in turn, provide strong output to DN1pC-E (Fig. 4D, E), thus linking CRY-negative DN_1p clusters to CRY-positive DN_1p clusters. After broadly classifying clock outputs, we next focused on specific cell types which receive the strongest synaptic inputs (>49 synapses) from clock neurons using an approach similar to the one used earlier for clock inputs. Our analysis identified the enigmatic Clamp neurons (Fig. 4F), which receive strong synaptic inputs from APDN₃. While the functions of most of these clamp neurons are still unknown, some of them output onto descending neurons, while others promote sleep²⁶. Moreover, DN_1pB provide strong inputs to Tubercle-innervating neurons (Fig. 4G), which are part of the anterior visual pathway⁴⁸. Lastly, several clock neurons are strongly connected to diverse neurons from different neuropil regions (Fig. 4H).

The circadian clock is also known to modulate behaviors such as activity/sleep, spatial orientation, and learning and memory^49,50. These behaviors are regulated by higher brain centers such as the central complex and mushroom bodies. The clock also influences systemic physiology via modulation of endocrine/neurosecretory cells (NSC)^16,21. Consistent with previous results^17,18, we found few direct connections from the clock to the central complex (Fig. 4I), mushroom bodies (Fig. 4J), and NSC (Fig. 4K). Consequently, we predicted that the clock output to these brain regions is either indirect or paracrine via neuropeptides. In line with this prediction, we found prominent disynaptic connectivity from clock neurons to central complex neurons (mainly fan-shaped body (FB) and ellipsoid body neurons (EB)), Kenyon cells (KC), dopaminergic neurons (DANs), mushroom body output neurons (MBONs) and NSC (Fig. 4I–K). Taken together, circadian modulation of neurons regulating diverse behaviors and physiology is largely indirect.

Next, we focused on clock output pathways to descending neurons which could influence diverse behaviors such as locomotion, feeding, and egg-laying. Interestingly, clock neurons provide direct inputs to 18 descending neurons (Fig. 5A). These descending neurons include those that have not yet been classified (Fig. 5B), as well as Allatostatin-C (AstC) and SIFamine (SIFa) peptidergic neurons (Fig. 5C), the latter of which modulate feeding, mating, and sleep⁵¹. When considering disynaptic connections, the connectivity between the clock network and descending neurons increased drastically, with approximately 24% of all descending neurons receiving indirect inputs from most of the clock neurons (Fig. 5A).

In order to obtain a better understanding of how the clock modulates descending neurons, and other downstream neurons in general, we first sought to characterize the neurotransmitters expressed in clock neurons. To address this, we used a combination of neurotransmitter predictions based on electron microscopy⁵² and anatomical expression mapping of neurotransmitter markers (Fig. 5D–F), which are largely in agreement^53,54. For some clock neurons such as DN_1pA and LN^ITP, anatomical mapping (Fig. 5F) allowed us to fill the gap left by uncertain neurotransmitter predictions based on electron microscopy (Fig. 5D). Remarkably, most of the lateral clock neurons express the excitatory neurotransmitter acetylcholine, while most of the dorsal clock neurons express glutamate which can either be excitatory or inhibitory in Drosophila⁵². The only exception to these are the cholinergic l-CPDN₃. None of the clock neurons are predicted to be GABAergic. We also examined the neurotransmitters predicted to be expressed in presynaptic and postsynaptic partners of clock neurons (Fig. 5D). As seen for clock neurons, acetylcholine and glutamate appear to be the major neurotransmitters expressed in their synaptic partners. In addition to neurotransmitters identified here, previous work has revealed that s-LN_v and l-LN_v utilize the inhibitory transmitter glycine⁵⁵. It is noteworthy that most of the postsynaptic partners of s-LN_v and l-LN_v are glutamatergic and cholinergic, respectively. Taken together, this comprehensive classification of neurotransmitters expressed in clock neurons allows valence assignment to the clock output pathways such as those to descending neurons.

We next utilized this information to characterize output pathways from the clock to downstream descending neurons. Direct synaptic output to descending neurons predominantly derives from s-CPDN₃A-D (glutamatergic), and LPN (cholinergic) (Fig. 5G). Three descending neuron types, namely DNpe048, DNpe033, and DNpe041, receive the majority of clock inputs. These descending neurons receive a substantial portion (about 25%) of their total input from clock neurons (Fig. 5H). We focussed further on DNpe048 which receives strong cholinergic and glutamatergic inputs from clock neurons. DNpe048 express crustacean cardioactive peptide (CCAP) and have recently been shown to regulate sugar and water ingestion by integrating inputs from other modulatory interneurons⁵⁶ (Fig. 5I). Allatostatin-A-expressing Janu-AstA neurons which regulate hunger and thirst⁵⁷ also provide inputs to DNpe048. It is thus tempting to speculate that DNpe048 integrate time and temperature cues via s-CPDN₃A-D and LPN along with hunger and thirst signals from diverse modulatory interneurons to regulate daily feeding rhythms (Fig. 5I, J).

Having elucidated major direct clock output to descending neurons, we decided to focus on the 312 descending neurons that receive clock inputs indirectly. These include oviposition descending neurons (oviDNa), vaginal plate opening descending neurons (vpoDN), and DNg30 that have previously been implicated in the regulation of reproductive behaviors. oviDNa regulate egg-laying⁵⁸, vpoDN control female sexual receptivity during courtship⁵⁹, and DNg30 express the neuropeptide natalisin which regualtes mating⁶⁰ (Fig. 5K). A recent study elucidated the pathway from LN_d^CRY+ to oviDNa which controls rhythmic egg-laying⁶¹. Here, we show that glutamatergic DN_1a and cholinergic l-CPDN₃ provide strong inputs to AVLP594 which in turn provide strong inputs to mating-regulating DNg30. Moreover, several types of clock neurons including s-CPDN₃C and APDN₃ provide strong inputs to vpoDN via pC1 neurons which integrate inputs regarding the female mating status^58,59. Lastly, SIFa neurons, which receive both direct and indirect clock signals, also independently affect sexual receptivity. It remains to be seen if mating and sexual rhythms exist in Drosophila, and the contribution of these pathways towards them. In summary, our connectivity analysis indicates that the clock can have a major influence on diverse behaviors via outputs to descending neurons.

Given the large repertoire of neuropeptides previously shown to be expressed in clock neurons^12,14,15, we predicted that peptidergic paracrine signaling is crucial in mediating intercellular coupling within the clock network as well as output to downstream neurons such as NSC. First, we determined if any additional neuropeptides are expressed in clock neurons. To address this, we used a publicly available single-cell transcriptome dataset of clock neurons¹⁴ combined with immunohistochemical localization and T2A-Gal4 lines. Unsupervised clustering of all clock neuron transcriptomes using t-SNE analysis yields 32 independent clusters (data not shown), 16 of which have high expression of clock genes (tim and Clk) and can be reliably identified based on known markers (Fig. 6A, B). Our analysis revealed that most clock clusters express at least one neuropeptide (Fig. 6B). Consistent with previous studies, l-LN_v express high levels of Pdf, whereas s-LN_v express both Pdf and short neuropeptide F (sNPF) (Fig. 6B). Similar coexpression of neuropeptides is also observed in other clusters including the DN_1p^{CNMa & AstC} cluster which coexpresses CNMamide (CNMa), AstC, and Diuretic hormone 31 (Dh31) neuropeptides. In total, at least 12 neuropeptides are highly expressed in the clock network (Fig. 6B, C). Importantly, this includes novel clock-related neuropeptides, namely DH44 and Proctolin (Proc). Dh44 is expressed in several clock clusters including DN_1a, DN_1p^AstA, DN_1p^sNPF, DN₃^VGlut, LPN and LN_d^NPF (Fig. 6B, C). We independently confirmed the presence of DH44 peptide in these clusters using a combination of DH44 antibody or DH44-T2A-Gal4 (Fig. 6D, Supplementary Fig. 15, and Supplementary Data 3). Proc, on the other hand, is strongly expressed in DN_1p^CNMa and weakly in DN₂ clusters which was verified by driving GFP using a Proc-T2A-LexA driver (Fig. 6E and Supplementary Fig.15). Proc expression in other clock clusters such as LN_d^NPF and DN₃^VGlut, remains to be validated. Lastly, we also detected AstC expression in additional clock neurons. AstC immunoreactivity was previously localized in DN_1p, DN₃, and LPN⁶², which is in agreement with AstC transcript expression in DN_1p^Rh7 and DN_1p^{CNMa & AstC} clusters (Fig. 6B, C). Here, we show that AstC is additionally expressed in DN₂, which were labeled using an antibody against VRI (Fig. 6B, F, Supplementary Fig.15, and Supplementary Data 3). Our expression analyses revealed the comprehensive neuropeptide complement of clock neurons (Fig. 6B–F, Supplementary Fig.15, and Supplementary Data 3) and provides the basis to explore paracrine targets of clock neurons.

As a first step in this direction and to validate our approach, we focused on select NSC which have been extensively characterized previously²³. We predicted that NSC are targeted by clock-related neuropeptides since they receive sparse monosynaptic inputs from clock neurons despite being closely associated with them anatomically. To investigate potential paracrine signaling between clock neurons and NSC, we again turned our attention to single-cell transcriptomics. We identified single-cell RNA transcriptomes of medial NSC expressing DH44 (m-NSC^DH44) and insulin-like peptides (m-NSC^DILP), as well as lateral NSC expressing corazonin (l-NSC^CRZ), DH31 (l-NSC^DH31) and ITP (l-NSC^ITP)⁶³ based on previously identified markers, and quantified the expression of clock peptide receptors (Fig. 6G). Consistent with our prediction, our analysis indicated that multiple receptors for clock peptides are indeed expressed in different populations of NSC. Modulatory inputs to m-NSC^DILP have been examined extensively and our analysis is in agreement with previous expression and functional studies^16,64–67. Taken together, our analysis uncovers the molecular substrates of paracrine signaling between clock neurons and NSC.

We next explored the magnitude of peptidergic paracrine signaling between clock neurons themselves. To predict putative paracrine connections, we extensively mapped the expression of clock neuropeptide receptors within the clock network. Consistent with promoter and genomic fragment-based Pdfr-Gal4 expression⁶⁸, Pdf receptor (Pdfr) transcript is highly enriched in most clock clusters (Fig. 7A, B), which we verified independently by expressing GFP using Pdfr[RA]-T2A-Gal4 (Fig. 7C, D). Other receptors are more sparsely expressed within the clock network (Fig. 7A, B). However, most clock clusters express at least two receptors, with DN_1p^sNPF expressing at least 7 receptors. We validated our single-cell transcriptome analysis by mapping the expression of select receptors using T2A-Gal4 knock-in lines. Our anatomical mapping of receptors (Fig. 7D and Supplementary Data 4) is largely in agreement with transcriptome data. In some cases, however, receptor mapping can provide additional insights. For instance, there are four transcript variants (A, B, C, and D) encoding the Drosophila Neuropeptide F receptor (NPFR). Using Gal4 lines specific for NPFR-A/C and NPFR-B/D isoforms, we showed that these isoforms are differentially expressed across the clock clusters, with A/C isoforms expressed more broadly than B/D isoforms (Fig. 7D and Supplementary Data 4).

Finally, we utilized our expression data of neuropeptides and their cognate receptors in clock neurons to delineate putative paracrine signaling pathways within the network. For this, we utilized an approach (Supplementary Fig. 16) similar to the one used recently to predict the Caenorhabditis elegans neuropeptide connectome⁶⁹. Briefly, we used expression data based on independent methods to conservatively localize the expression of neuropeptides and their receptors across all the clock clusters. We also utilized the connectome to factor in the distance between neurons of different clusters. This was done to ensure that the cells releasing the peptide and those expressing its receptor are not further apart than a cut-off of 14μm, which was set based on a previous paracrine connectivity study¹⁶. Lastly, we only considered strong peptide-receptor interactions by disregarding ligands with EC₅₀ values for receptor activation higher than 500 nM. This stringent in silico approach allowed us to predict paracrine connectivity within the clock network with high confidence. Taking s-LN_v as an example, this cluster expresses both PDF and sNPF. Following our expression thresholding, PDFR is expressed in most clock clusters, whereas sNPF receptor is expressed in DN_1p, DN₃, LPN, and l-LN_v (Fig. 7 and Supplementary Fig.16). Thus, in addition to providing synaptic inputs to DN₃, s-LN_v can potentially provide paracrine inputs to most clock clusters across both hemispheres (Fig. 7E). Connectivity from l-LN_v is also enhanced by paracrine signaling, although not to the same extent as it is for s-LN_v. Expanding this analysis to other clock clusters allowed us to comprehensively identify putative peptidergic signaling pathways between clock clusters (Fig. 7E). These pathways, however, can only be considered putative for three main reasons: (1) all clock neuropeptides are also expressed in other non-clock neurons⁷⁰, suggesting likely inputs from neurons extrinsic to the clock network, (2) we don’t account for peptide efficacies and receptor affinities since these values were independently determined in distinct systems thus making comparisons difficult and (3) despite our best efforts to account for it, distance of peptide diffusion may vary. To put our peptide diffusion threshold of 14μm into perspective, we performed an additional stringent analysis where we lowered this threshold to 1μm. In this scenario, long-range peptide diffusion is not necessary and two neurons could communicate via paracrine signaling only if they are touching or almost within touching distance and express the peptide-receptor pair. Even using these highly stringent criteria, we see significant putative paracrine connections (~92% of those seen with the 14μm threshold) between different clock classes (Supplementary Fig. 16C). We speculate that the actual magnitude of paracrine signaling is somewhere between no peptide diffusion and brain-wide peptide diffusion, and our threshold of 14μm appears to be a good starting point to elucidate paracrine signaling pathways. In summary, peptidergic signaling greatly enriches the connectivity between different subsets of clock neurons. Additional investigations are necessary to determine which of these putative connections are functional in vivo.

Using a multipronged approach centered around the FlyWire connectome^9,10, we describe the whole-brain neural connectome of an animal circadian clock. Our analysis using the complete brain connectome eluded identification of one DN_2, and a couple of s-CPDN₃. Nevertheless, given the fact that we identified almost all of the expected clock neurons as well as several additional DN₃ (Table 1), our clock neuron synaptic wiring diagram is complete enough to be regarded as a connectome. Our clock connectome is also a significant upgrade (10-fold larger numerically) compared to the partial connectivity diagram based on the hemibrain connectome reported earlier¹⁹. The previous analysis was based on only 24 clock neurons and largely focused on LN clusters, while excluding l-LN_v, and several DN_1p, DN₂, and DN₃ clusters due to the incomplete nature of the dataset. However, as evident from our analysis here, the DN in fact represents an important hub in the clock network and displays high synaptic connectivity. In particular, DN_1p plays a large role in clock cluster interconnectivity and DN₃ appears important for clock output pathways. Our analysis also sheds light on the precise number of DN₃ in the clock network. Although approximately 80 DN₃ were previously estimated in the entire clock network, there are in total about 170 DN₃ based on our connectome and anatomical analyses. We anticipate that resources such as NeuronBridge⁷¹, which allow for comparisons between electron and light microscopy datasets, will facilitate the identification of Gal4 drivers that target the novel s-CPDN₃ subtypes identified here. In addition, we also characterized the molecular basis for neuropeptide connectivity between the clock neurons, consequently highlighting putative peptidergic pathways within the clock network. Similar to vertebrates^22,72, the Drosophila clock network is highly peptidergic, with all clock neuron clusters expressing at least one neuropeptide. Notably, a majority of the clock clusters express two neuropeptides, and several express three, while the LPN expresses four neuropeptides. Similar neuropeptide coexpression is also evident in SCN neurons and thus appears to be a common feature of clock neurons²². Surprisingly, there is little to no overlap in the neuropeptide complement of the Drosophila and vertebrate clock neurons (Fig. 8). Orthologs of vertebrate clock neuropeptides including vasoactive intestinal peptide, arginine vasopressin, neuromedin S, cholecystokinin, gastrin-releasing peptide, and prokineticin 2 are either absent in the Drosophila genome or expressed outside the clock network. Hence, Drosophila and vertebrates have evolved to utilize different signaling molecules while still conserving the diversity of neuropeptide signaling within the clock networks. Remarkably, except for PDF, there appears to be little conservation in neuropeptide identities of clock neurons across different insects^49,73. This suggests that it is more important to conserve the mode of communication (paracrine signaling) rather than the messenger (specific neuropeptide).

Analysis of interconnectivity within the clock network revealed extensive contralateral synaptic connectivity between the clock neurons, which is largely mediated by DN_1pA and to a lesser extent by s-CPDN₃C, s-CPDN₃D, l-CPDN₃, and LN^ITP. Furthermore, paracrine peptidergic signaling amongst clock neurons has the potential to further strengthen this contralateral connectivity. Such a strong bilateral coupling of clock neurons could prevent the internal desynchronization of clock neuron oscillations between the two hemispheres – a phenomenon that can happen in other insects and even in mammals, but so far has not been observed in fruit flies⁷. In addition, consistent with previous analysis using the hemibrain connectome¹⁹, we did not observe any synaptic connectivity between the s-LN_v and the LN_d neurons, which control morning and evening activities, respectively. In line with this observation, the phase relationship between morning and evening oscillators is plastic, consequently facilitating seasonal adaptations as in mammals²⁸. Our data may also explain how the morning and evening oscillators in flies internally desynchronize under certain conditions⁷⁴. One such relevant condition is increased PDF signaling during long days²¹, which was shown to delay the evening oscillators^13,75,76 and may lead to internal desynchronization⁷⁷. Here, we confirm the presence of PDFR in the evening neurons. Thus, enhanced PDF signaling could delay the evening neurons and bring them out of phase with the morning neurons, especially because the two sets of neurons are not connected via synapses. Taken together, the lack of interconnectivity between s-LN_v and LN_d could potentially be a factor promoting adaptation to different seasons and contexts.

Our analyses reveal that extrinsic light input from the photoreceptor cells of the compound eyes, HB eyelets, and ocelli to the clock neurons is largely indirect, with the former two transmitting light inputs via aMe neurons. This situation may appear to be different from mammals where intrinsically photosensitive retinal ganglion cells in the retina project directly through the retinohypothalamic tract onto the SCN neurons (Fig. 8)⁷⁸. However, even the mammalian clock receives indirect photoreceptor inputs from rods and cones via bipolar cells and retinal ganglion cells. Furthermore, indirect light inputs may reach the SCN also via the intergeniculate leaflets of the thalamus⁷⁹. Altogether, this suggests that the fly and mammalian systems are not fundamentally different. One apparent difference is the presence of CRY, a cell-autonomous circadian photoreceptor, in subsets of Drosophila clock neurons that is sufficient for light entrainment in eyeless mutants⁸⁰. Mammals lack light-sensitive CRY. Instead, they possess light-sensitive melanopsin in retinal ganglion cells, and mice lacking rods and cones can still entrain to light/dark cycles due to melanopsin⁸¹. Thus, flies and mice possess several redundant and partly parallel light-input pathways to entrain their clocks. The similarity is even higher when comparing flies with vertebrates in general. For instance, fish, birds, and reptiles possess additional photoreceptors in the pineal gland, which is reminiscent of the extraretinal HB eyelets or even the ocelli of flies (Fig. 8). Most importantly, vertebrates and flies use their eyes for both vision and entraining their circadian clocks, tasks that require completely different properties of light inputs. Vision requires image formation and fast neurotransmission, whereas circadian entrainment is dependent on integrating light collection over a longer time that can be at a slower rate. The connectome reveals that the number of synapses as well as the axon thickness of the neurons mediating this connectivity are very different between the two types of photoreception. Hence, they are aptly suited to perform their required functions.

The synaptic and putative paracrine connections reported here can only be considered predictions until they are functionally verified. In the case of paracrine connectivity, functional connectivity experiments (in normal and peptide mutant flies) using electrophysiological methods or genetically encoded secondary messenger sensors (calcium and cAMP) are needed to confidently establish functional paracrine connectivity. Additionally, the synaptic connectivity reported here is likely an underestimation due to several factors: (1) we did not explore connectivity via gap junctions, (2) proofreading of photoreceptor axons was not complete in the version utilized here¹⁰, and (3) we generally used a connectivity threshold of >4 synapses. Preliminary expression analysis using the single-cell transcriptomes of clock neurons suggests that gap junction genes are enriched in the clock network (data not shown) and they can influence activity-rest rhythms⁸². It remains to be seen which clock neurons are additionally coupled via gap junctions and how this electrical connectivity complements synaptic and peptidergic connectivity detailed here. Moreover, as discussed earlier, fewer than 5 synapses could also represent functional connections which were largely disregarded in our analyses. Further, during the revision of this manuscript, additional synapses were predicted in the FlyWire connectome by Princeton University. Some of these additional synapses increase the connections between clock neurons, specifically those mediated by LN^ITP (not shown). Crucially, these additional synapses and connections do not change the conclusions reported here. While the FlyWire and hemibrain connectomes exhibit a high degree of stereotypy, they are both based on an adult female brain. The lack of a male brain connectome currently prevents any comparisons on sex-specific differences within the circadian network and its output pathways which could influence sexually dimorphic behaviors and physiology. Further, the connectome provides a singular snapshot of connectivity which could change depending on the time of day, the age of the animal as well as its internal state.

In conclusion, our circadian clock connectome, the first of its magnitude, is a significant milestone in chronobiology. Given the high conservation of circadian network motifs between Drosophila and vertebrates, this connectome provides the framework to systematically investigate circadian dysregulation which is linked to various health issues in humans including sleep, metabolic, and mood disorders. Moreover, it will also facilitate the development and experimental validation of novel hypotheses on clock function.

Drosophila melanogaster strains used in this study are listed in Supplementary Table 1. T2A-Gal4 and T2A-LexA knock-in lines generated previously ^83,84 were obtained from the Bloomington Drosophila Stock Center (BDSC) and Dr. Shu Kondo. Flies were maintained under LD12:12. Flies for peptide and receptor mapping were reared at 25 °C on Drosophila medium containing 0.7% agar, 8.0% glucose, 3.3% yeast, 4.0% cornmeal, 2.5% wheat embryo, and 0.25% propionic acid. Flies for trans-Tango and Gal4 verification were raised at 18 °C and 25 °C, respectively, on a standard medium containing 8.0% malt extract, 8.0% corn flour, 2.2% sugar beet molasses, 1.8% yeast, 1.0% soy flour, 0.8% agar and 0.3% hydroxybenzoic acid. 2-week-old flies were used for trans-Tango analysis.

Further information on research design is available in thelinked to this article. Nature Portfolio Reporting Summary

Nature Communications thanks Xitong Liang, Amita Sehgal and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Connectivity analyses can be performed using the cell IDs provided at https://codex.flywire.ai/↗ and with the scripts provided (see below). Source data for plots are provided with this paper. Source data are provided with this paper.

Code used for the analyses and data visualization is publicly available on GitHub: https://github.com/Zandawala-lab/Synaptic-connectome-of-the-Drosophila-circadian-clock_Reinhard-and-Fukuda-et-al.-2024↗.