Segmentally repeated ventral nerve cord circuits drive different leg rubbing behaviors in Drosophila grooming

TL;DR

Flies use repeated nerve circuits to coordinate different leg rubbing behaviors based on which leg is stimulated.

Contribution

Identification of segmentally repeated command-like interneurons and their role in distinct leg rubbing behaviors in Drosophila.

Findings

Activation of LegPNs induces segment-specific leg rubbing.

LegPL activation causes leg flexion, while LegPC activation triggers multi-segmental leg rubbing.

Circuit differences in T2 LegPCs correlate with distinct leg recruitment patterns.

Abstract

Animals recognize mechanosensory stimuli and generate targeted responses. Flies perform distinct leg rubbing movements depending on which of the legs is stimulated. While the leg that receives stimulus is always involved in cleaning, different adjacent legs are recruited. For front and hind legs, the contralateral homolog or ipsilateral middle leg is used, but for the middle leg, one or both hind legs are engaged. Here, we identify six segmentally repeated command-like interneurons, LegPNs, whose activation induces leg rubbing. Sensorimotor circuits, repeated in each neuromere, include mechanosensory inputs and reciprocal excitatory connections with pre-motor LegPLs. Activation of LegPLs causes leg flexion. There are segmental differences in the circuits downstream of LegPNs for the middle legs—T2 LegPCs lack commissural connections but include additional ipsilateral intersegmental projections. LegPC activation causes simultaneous front and back leg rubbing. These behavioral and anatomical results demonstrate how shared and modified serially homologous neural circuits coordinate segmentally distinct grooming movements.

•Leg stimulation or LegPN activation triggers segment-specific leg rubbing•Segmentally repeated circuits link leg mechanosensory and motor neurons via LegPNs•Differences in commissural and intersegmental circuits correlate with leg recruitment•LegPL activation induces flexion while LegPC activation triggers multi-segmental leg rubbing

Leg stimulation or LegPN activation triggers segment-specific leg rubbing

Segmentally repeated circuits link leg mechanosensory and motor neurons via LegPNs

Differences in commissural and intersegmental circuits correlate with leg recruitment

LegPL activation induces flexion while LegPC activation triggers multi-segmental leg rubbing

Neuroscience; Behavioral neuroscience

Figures

Leg usage in rubbing response depends on which leg is physically stimulated or which LegPNs are optogenetically activated(A) Leg usage in rubbing response upon contact with the left front leg (top), left middle leg (middle), and left hind leg (bottom). The targeted, touched leg is indicated in red. Bars represent the mean value of each group, and error bars indicate the standard error of mean (SEM) for each group. The average percentages of each leg rubbing combination for stimulation of the left front leg (top), left middle leg (middle), and left hind leg (bottom). The legs involved are indic

Local sensorimotor circuits of LegPNs(A) Reconstruction of T1L-LegPN from FANC EM data. Left, the skeleton with pre-synaptic and post-synaptic sites of T1L-LegPN; orange, post-synaptic sites; blue, pre-synaptic sites. Right, pre-synaptic and post-synaptic site distribution across different ganglia in the ventral nerve cord.(B) Reconstruction of example sensory neurons that synapses onto T1L-LegPN, including tactile bristles (*n* = 3), proprioceptor (*n* = 3), and bitter sensory neurons (*n* = 1).(C) Reconstruction of an example LegPL neuron (T1L-LegPL1) and five T1 leg motor neurons post-synap

Identification of segment-specific pre-motor circuits(A) Connectivity between LegPN command-like neurons, pre-motor neurons, and motor neurons in each thoracic segment based on FANC connectome data. The LegPLs are grouped together, and commissural pre-motor neurons are shown in blue text; the other pre-motor neurons post-synaptic to LegPNs are listed individually by the names indicated in Figure S3. Synapse counts for the left-side reconstruction are indicated.(B–D) EM reconstructions of LegPCs. (B) EM reconstructions in T1L. (C) EM reconstructions in T3L. (D) EM reconstructions in another T3L

Activation of LegPCs evokes simultaneous leg rubbing(A) Expression pattern of SS53029 (*R24E12-p65ADZp* [*attP40*]; *VT019345-ZpGdbd* [*attp2*]) targeting LegPCs (green) with nc82 synaptic neuropil counterstain (magenta).(B) Ethograms showing the leg rubbing phenotypes of 8 individual, intact flies during optogenetic activation (10 s indicated by the red bar below). Note the cream-colored regions indicating simultaneous front and back leg rubbing, a new behavior, shown on the still photograph (C) and quantified in (D) in purple, as compared to the percent time spent performing front leg rubbin

Full text

Introduction

An animal’s nervous system detects relevant mechanosensory stimuli1^,^2^,^3 and coordinates limbs for walking, jumping, flight, and grooming. Sensory and motor circuits in the vertebrate spinal cord or insect ventral nerve cord (VNC) can produce leg movements, often under control from the brain via descending neurons. The neurons capable of commanding coherent motor programs and the circuits generating rhythmic actions are now being identified by anatomical, computational, and behavioral experiments.

Genetic tools to label specific neurons of interest,4 methods for precise quantification of limb movements,5 and large-scale anatomical datasets showing neuronal connectivity6 make Drosophila a powerful system to investigate neuronal coordination of behaviors. Mechanosensory and proprioceptive neurons from the flies’ legs have been characterized,7 and motor neurons innervating each leg muscle have been mapped.8^,^9 Some of the descending neurons10^,^11^,^12 and interneurons located in the VNC13 proposed to contribute to motor control have been described, and recent modeling experiments based on connectivity data have reported candidates for central pattern generating circuits.14

Grooming behavior is initiated by mechanosensory stimulation15^,^16^,^17 and requires coordinated leg movements.18 Sensory neurons, interneurons, and motor neurons that contribute to grooming have been identified in neuronal silencing and activation screens.19^,^20^,^21^,^22 Here, we describe neurons whose activation induces leg rubbing behavior and use their connectivity to identify additional neurons that contribute to grooming.

Results

Leg rubbing employs different combinations of legs depending on which leg is stimulated

Proper response to tactile stimuli on different body parts is crucial for animal survival. Peripheral receptors sense mechanosensory inputs, which are then processed in the central nervous system to generate appropriate motor outputs.15^,^16^,^23^,^24^,^25^,^26 Flies rub their legs together in response to bristle deflection because of light touch or debris.19^,^27 Even decapitated flies can perform this sensory-motor reflex.28 When we deflected leg bristles, we observed location-dependent behavioral responses: the stimulated leg became the target for rubbing by the other legs (Figures 1A and S1A). The stimulated leg is always involved, but other “helper” legs are also employed. These can be the leg from the same segment on the other side (contralateral) or the leg on the same side from the adjacent segment (intersegmental or ipsilateral), and the most common combination of legs differs among segments. For example, when the left front leg L1 was stimulated, the right front leg R1 was recruited alone 23% of the time, the left middle L2 leg was recruited alone 56% of the time, and both the R1 and L2 participated 21% of the time for three-leg rubbing. Similarly, when bristles on the hind leg L3 were deflected, R3 and L2 legs cleaned L3 (a three-leg combination) 49% of the time, R3 rubbed L3 33% of the time, and L2 rubbed L3 18% of the time. In contrast, when the left middle leg L2 was touched, the contralateral right middle leg R2 was never recruited. Instead, either the left hind leg L3 (27%) or both left and right hind legs (69%) were engaged (Figure 1A). Thus, contralateral combinations are common in front and hind leg rubbing but not when the middle legs are stimulated. Figure 1A shows the outcomes when the left legs were touched; comparable results were observed during right leg stimulation (Figure S1A). The relative frequency of the usage of different leg combinations is also similar to what has been described during spontaneous, unstimulated grooming.29^,^30Figure 1. Leg usage in rubbing response depends on which leg is physically stimulated or which LegPNs are optogenetically activated(A) Leg usage in rubbing response upon contact with the left front leg (top), left middle leg (middle), and left hind leg (bottom). The targeted, touched leg is indicated in red. Bars represent the mean value of each group, and error bars indicate the standard error of mean (SEM) for each group. The average percentages of each leg rubbing combination for stimulation of the left front leg (top), left middle leg (middle), and left hind leg (bottom). The legs involved are indicated both by the fly drawings and the color key. Note the absence of the L2-R2 combination. See also Figure S1A.(B) Expression pattern of R17F11-p65ADZ (attp40); VT029514-ZpGDBD (attp2) labeling 6 LegPNs (green) in the ventral nerve cord (CNS, magenta). Scale bars are 100 μm, unless otherwise indicated.(C) Average leg rubbing responses upon simultaneous activation of all 6 LegPNs (R17F11AD, VT029514DBD > CsChrimson; n = 16). Smooth line shows the mean trend using LOESS. Optogenetic stimulation was given between 60 s and 120 s, indicated by the red line.(D) Individual behavioral records (ethograms) for intact and decapitated flies.(E and F) Expression pattern of individual LegPNs (E) and behavioral phenotypes (F). Genotype: VT029514-Gal4, R57C10-Flp2::PEST, UAS > stop > CsChrimson.See also Figure S1B for the right-side clones.

Optogenetic activation of individual LegPNs induces segment-specific leg rubbing responses

The vertebrate spinal cord and the arthropod VNC have segmentally repeated units for somatosensation and motor control of limbs.31 In an adult fly, these are located in the three thoracic segments (T1–T3) and called the leg neuropils.32 The T1, T2, and T3 neuropils are associated with front, middle, and hind legs, respectively; each segment has both left and right neuropils. Each leg neuropil receives sensory information in the ventral region, and the dendrites of the motor neurons controlling that leg’s muscles are located in the dorsal region.9^,^13^,^33^,^34 Leg sensory neurons usually do not synapse directly onto the leg motor neurons (except the bilateral campaniform sensilla, bCS).9 Most sensory to motor connections occur through local interneurons, while some pass through commissural, bilateral, or intersegmental neurons.11^,^34 Descending neurons can also influence the choice of motor programs.35^,^36^,^37 Flies use their legs for walking and grooming behaviors,38 and we were interested in identifying neural circuits specific to grooming movements. Our previous screen19 of over 1,500 Gal4 lines identified only two circuits, R17F11-Gal4 and R53E05-Gal4, whose activation induced leg grooming. We searched for additional candidate lines by visual observation of Gal4 expression patterns in the FlyLight database,39 selecting 9 lines that label neurons directly downstream of leg mechanosensory neurons. We tested for optogenetic activation phenotypes, and only VT029514-Gal4 activation induced leg grooming. Interestingly, R17F11-Gal4 and VT029514-Gal4 labeled a common set of neurons. The intersection of R17F11-AD and VT029514-DBD lines revealed a class of command-like interneurons, which we named “LegPNs” (leg projecting neurons) and whose optogenetic activation induced leg rubbing movements (Figures 1B and 1C). We consider the LegPN command-like neurons in the sense that their activation can evoke a natural behavioral program, using the definition proposed by Wiersma and Ikeda.40 Subsequent discussions of the command neuron concept41^,^42 have debated additional criteria, including that candidate neurons should be necessary for the behavior and active during it.43 LegPNs are not absolutely required for leg rubbing: silencing LegPNs using GtAcr2 or Kir2.1 does not eliminate grooming behaviors induced by bristle deflection or dust (Figure S4B), and we do not yet know if the LegPNs are normally active during dust-induced grooming or leg rubbing. The silencing experiments indicate that there must be additional circuits that can support leg rubbing during grooming, but we propose that the activation phenotype of the LegPNs makes them an effective entry point to identify neural circuits normally controlling this behavior.

There are six LegPNs in the VNC, and they are segmentally repeated or serial homologs,34 meaning that a pair of similar neurons can be found in each segment, with one neuron in each leg neuropil. Most of each neuron’s axon terminals and dendrites remain local, restricted to the hemi-neuromere associated with a single leg; so, we named them T1L-, T1R-, T2L-, T2R-, T3L-, and T3R-LegPNs, respectively. In addition to the branches in the VNC, LegPNs also extend neurites (presumably axons) into the brain, where they project exclusively in the suboesophageal zone (SEZ). While this makes them anatomically classified as ascending neurons capable of conveying information from the VNC to the brain, their projections within the VNC seem to be sufficient for their role in leg rubbing because activation of the LegPNs in decapitated flies could still induce leg rubbing (Figure 1D).

Activation of all six LegPNs triggered multiple types of leg rubbing movements in alternation (Figure 1D), which prompted us to investigate whether activation of individual LegPNs can elicit different leg rubbing combinations. We used a recombinase-based strategy that generates expression in random subsets within the GAL4 expression pattern (Figure 1E). Activation of each individual LegPN led to one or two dominant leg rubbing types, similar to the combinations of legs recruited by manual mechanosensory stimulation of the corresponding leg (compare Figures 1A and 1F). For example, optogenetic activation of the T1L-LegPN induced leg rubbing that always involved the L1 leg and either the R1 leg or both the R1 and L2 legs (Figure 1F). Clonal activation of the T2-LegPN recruits hind legs for intersegmental leg rubbing. T3 clones are rare, but consistent results were seen in clones on the left and right sides (Figure S1B). Examples of clonal activation phenotypes are shown in Video S1. Thus, the segmental differences in leg involvement from individual LegPN activation are reminiscent of rubbing responses to mechanical stimulation on each leg.

Video S1. Activation phenotypes of individual LegPNs show leg rubbing behaviors targeted to different legs

LegPNs participate in local sensorimotor circuits

To investigate the sensorimotor circuits underlying leg rubbing in different legs, we identified and reconstructed the six LegPNs in the electron microscopy (EM) dataset of the female Drosophila VNC, known as FANC9 (Figures 2A and S2A). We focused on the left LegPNs where the dataset is more complete.33 Consistent with our observations from light microscopy, automatic synapse detection in EM44 revealed that each LegPN has enriched projections in the corresponding leg neuropil. T1L-LegPN showed that 95% of post-synaptic sites and 50% of pre-synaptic sites are in T1L neuromere, indicating predominantly local input and output connections (Figure 2A). Similar synaptic distributions were observed in T2L- and T3L-LegPNs (Figure S2B).Figure 2. Local sensorimotor circuits of LegPNs(A) Reconstruction of T1L-LegPN from FANC EM data. Left, the skeleton with pre-synaptic and post-synaptic sites of T1L-LegPN; orange, post-synaptic sites; blue, pre-synaptic sites. Right, pre-synaptic and post-synaptic site distribution across different ganglia in the ventral nerve cord.(B) Reconstruction of example sensory neurons that synapses onto T1L-LegPN, including tactile bristles (n = 3), proprioceptor (n = 3), and bitter sensory neurons (n = 1).(C) Reconstruction of an example LegPL neuron (T1L-LegPL1) and five T1 leg motor neurons post-synaptic to this LegPL. Five T1-LegPLs are identified.(D) The proportion of T1L-LegPL synapses that target T1L motor neurons.(E) Venn diagram of T1L-MNs that T1L-LegPLs target. Note that each LegPL targets some common and some specific T1 leg motor neurons.(F) Recurrent or reciprocal connection motifs between LegPNs and LegPLs in each segment with the synapse numbers indicated on the arrows.(G and H) Expression pattern of R79C09-p65ADZ (attp40). (G) R24D12-ZpGDBD (attp2) labeling LegPL neurons (green), and (H) quantification of activation phenotype as the average flexion angle induced immediately after optogenetic activation.(I) Still photographs illustrating the normal and induced leg flexion angles (red lines).(J) Summary of local sensorimotor circuits around LegPNs.See also Figures S2 and S3.

To determine whether LegPNs receive sensory inputs, we reconstructed the most highly connected pre-synaptic partners based on synapse number, an approximate indicator of connectivity strength.45 In Figure S2C, we present the pre-synaptic partners of T1L-LegPN as an example; T2L- and T3L-LegPN connection numbers are given in Figure S2D. Although LegPNs receive many synapses, most pre-synaptic partner neurons contribute only a few synapses each. Only 30 neurons could make six or more synapses onto T1L-LegPN, and among these, seven were sensory neurons from the left front leg (Figure 2B). Of the 1,099 predicted synapses onto T1L-LegPN, 184 (16.7%) were from sensory neurons. Further categorization of sensory neuron morphology9 shows that these come from mechanosensory tactile bristle neurons or proprioceptive campaniform sensilla and hair plates. Previous studies have shown that thermo- or opto-genetic activation of these sensory neuron classes triggers grooming responses, including leg rubbing.16^,^17^,^46 A few contacts come from bitter sensory neurons, which have also been shown to evoke grooming.47 Direct sensory inputs to LegPNs are also observed in T2L and T3L (Figure S2D). It is difficult to infer the functional significance of this number of synaptic connections from mechanosensory bristle neurons. There is an estimate of 350–450 tactile bristle neurons for each leg,48^,^49^,^50 and they synapse onto many VNC interneurons other than LegPNs,24^,^25 including some targets with higher connection convergence.7^,^15^,^24^,^25^,^51 The mechanosensory neurons are not the only neurons pre-synaptic to the LegPNs: the most highly connected partners are uncharacterized excitatory and inhibitory interneurons, which might provide indirect sensory information as well. Thus, while LegPNs do receive direct mechanosensory information from tactile bristle neurons, circuit connectome analysis suggests that there are additional pathways that convey mechanosensory information into the central nervous system and may also contribute to driving a normal leg rubbing behavior.

Next, we examined the circuits post-synaptic to LegPNs to understand how activating these neurons evokes appropriate leg movements. The most connected downstream partners of each LegPN were reconstructed in the VNC EM (FANC9) dataset (Figures S3A–S3C). The LegPNs do not connect directly to leg motor neurons, but they do synapse onto a group of highly interconnected, anatomically similar pre-motor neurons (pre-MNs) in each segment. We named these neurons “LegPLs” (leg projecting local neurons).

We used T1L-LegPLs for illustration purpose (Figures 2C–2E); the connectivity between LegPNs and LegPLs was repeated in each segment (Figures 2F and S3A–S3C). Eleven of the seventeen most highly connected post-synaptic partners of T1L-LegPN were pre-motor neurons, and five of those were T1L-LegPLs (122 synapses; Figures 3A and S3A; Table S1). These LegPLs make approximately 50% of their synapses onto motor neurons and share target motor neurons in common, suggesting coordinated function (Figures 2D and 2E).Figure 3. Identification of segment-specific pre-motor circuits(A) Connectivity between LegPN command-like neurons, pre-motor neurons, and motor neurons in each thoracic segment based on FANC connectome data. The LegPLs are grouped together, and commissural pre-motor neurons are shown in blue text; the other pre-motor neurons post-synaptic to LegPNs are listed individually by the names indicated in Figure S3. Synapse counts for the left-side reconstruction are indicated.(B–D) EM reconstructions of LegPCs. (B) EM reconstructions in T1L. (C) EM reconstructions in T3L. (D) EM reconstructions in another T3L commissural pre-motor neuron.(E) Reconstruction of LegPC serial homologs in MANC EM data. Scale bars, 100 μm.EM, electron microscopy.

LegPLs are synaptically connected to each other and also make reciprocal feedback connections onto LegPNs, forming highly interconnected local networks (Figure 2F). Both LegPNs and LegPLs are likely to be cholinergic and excitatory based on their lineages and automatic neurotransmitter class predictions.34^,^52^,^53 Our immunohistochemistry data support this: LegPNs could be labeled with cholinergic reporter lines but not with glutamatergic or GABAergic ones (Figures S1C and S1D). We propose that LegPNs and LegPLs form a positive feedback loop to amplify local sensory signals and reinforce selection of the stimulated leg as the target for leg rubbing. Alternatively, they may contribute to the sensory-motor mapping required to bring stimulated legs into contact54 or constitute part of a circuit generating rhythmic leg rubbing movements.55

We obtained genetic reagents for targeting LegPLs (Figure 2G), and optogenetic activation in intact flies showed leg flexion in all segments (Figures 2H and 2I; Video S2)—an essential component of leg rubbing movements. As with LegPNs, silencing LegPLs with GtACR2 did not significantly reduce leg rubbing in dusted flies (Figure S4D), indicating that other partially redundant circuits supporting leg rubbing must exist.

Video S2. Optogenetic activation of LegPNs (alternating leg rubbing actions), LegPLs (leg flexion in all segments), and LegPCs (both front and hind leg rubbing)

In summary, we discovered a neural circuit within each VNC neuromere that bridges from local sensory perception to motor outputs in each leg (Figure S4A). Tactile sensory neurons are connected to a command-like LegPN neuron, which, in turn, synapses onto LegPL pre-motor neurons (Figure 2J). There are reciprocal excitatory connections among LegPLs and between LegPLs and LegPNs. This local connectivity pattern is repeated in each segment to build a sensory-motor circuit that can induce leg rubbing.

Segment-specific neural architecture may explain why different leg combinations are recruited

Leg rubbing relies on the involvement of both the target leg and the rubbing partners, either contralateral or segment-adjacent legs. We observed segmental difference when the T1 front legs or T2 middle legs were stimulated, either by mechanical touch or command neuron activation (Figures 1A, 1F, S1A, and S1B). Front leg activation typically recruited the contralateral leg, while middle leg activation never led to contralateral leg recruitment but instead engaged the hind legs. These findings suggest that the circuits connecting LegPNs to motor neurons ought to diverge in each segment to account for these differences in leg selection.

Commissural connections, linking the left and right leg neuromeres, might be involved in coordinating leg rubbing. Commissure disruption can affect grooming,56 and the role of excitatory and inhibitory commissural neurons is well established in the control of gait in vertebrates.57^,^58 Given the LegPNs’ ability to recruit left and right legs for rubbing, we hypothesized that either the LegPNs themselves or their post-synaptic partners might form connections across the midline. Among the highly connected post-synaptic partners of T1L-LegPN, we identified a commissural pre-MN, “T1-LegPC” (leg projecting commissural neuron), that synapses onto motor neurons in both left (6 synapses) and right (64 synapses) T1 leg neuropils (Figures 3A and 3B). There are at least two commissurally projecting pre-MNs post-synaptic to T3-LegPN as well (T3-LegPC and T3-D16; Figures 3A–3D and S3C).

In contrast, we did not find commissural pre-MNs post-synaptic to the LegPNs in the T2 segment associated with the middle leg. T2L-LegPN formed strong connections (partners with more than 10 synapses) with nine pre-MNs. Five pre-MNs, including four LegPLs, synapse exclusively onto local T2L motor neurons. The other four pre-MNs project ipsilaterally to T3L motor neurons (Figures 3A and S3B)—T2L-D1, T2L-D8, T2L-D17, and T2L-D18—none showed commissural projections. We named them “T2-LegPI” (leg projecting intersegmental neurons) to indicate their ipsilateral connectivity and speculate that they may contribute to the recruitment of T3 hind legs to help with cleaning the middle legs, as shown in Figures 1 and S1.

We identified a genetic reagent (SS53029) to target the LegPCs. The splitGAL4 line labels a pair of neurons, putative serial homologs, in each segment (Figure 4A). The T1 and T3 neurons show commissural projections, but as predicted from the EM, the T2 homologs do not. Instead, T2-LegPCs extend neurites ipsilaterally from T2 to T1 (Figures 3E and S4F). These segmental circuitry differences align with the behavioral output when the different legs are stimulated or the different LegPNs are activated: front/hind legs rub against their left-right partners, while each middle leg is cleaned by the hind legs instead.Figure 4. Activation of LegPCs evokes simultaneous leg rubbing(A) Expression pattern of SS53029 (R24E12-p65ADZp [attP40]; VT019345-ZpGdbd [attp2]) targeting LegPCs (green) with nc82 synaptic neuropil counterstain (magenta).(B) Ethograms showing the leg rubbing phenotypes of 8 individual, intact flies during optogenetic activation (10 s indicated by the red bar below). Note the cream-colored regions indicating simultaneous front and back leg rubbing, a new behavior, shown on the still photograph (C) and quantified in (D) in purple, as compared to the percent time spent performing front leg rubbing (red) or back leg rubbing (blue) alone when LegPNs (left histogram) or LegPCs (right histogram) are activated.

The splitGAL4 line allowed us to test the effect of LegPC activation: it causes leg rubbing in intact flies (Figure 4B). Although the phenotype is similar to LegPN activation, there are some interesting differences (compare Figures 1D and 4B). LegPC activation had a faster behavior onset, elicited more two-leg rubbing using left right pairs, and could induce attempts to rub both the T1 and T3 legs simultaneously (Figures 4C and 4D; Video S2). Thus, we identified one leg rubbing command-like neuron (LegPN) from large-scale behavioral screens to initiate this circuit investigation, and now we have found anotherone (LegPC) by connectome analysis.

Examination of LegPN circuits in additional connectomes

Since this work was initiated, we obtained access to a second VNC connectome for an adult male, MANC,34^,^59 where we confirmed key findings (Table S1). We identified the six LegPNs in MANC by their morphological similarity to the FANC reconstructions and light level images. The MANC LegPNs are now annotated as belonging to a serial set, 11,844, and as ascending neurons from the 5B hemi-lineage AN05B100.34 Pre-synaptic partners of the MANC LegPNs also include local leg mechanosensory neurons. A synapse threshold of 10 revealed 32 sensory neurons that directly synapse onto LegPNs, while a synapse threshold of 3 increased this to 145 neurons (from the ch, ta, pp, and unclassified xx mechanosensory and proprioceptive types). Indirect connections between mechanosensory neurons and LegPNs via excitatory interneurons were also observed (Table S1 and data not shown). Thus, there are circuits connecting mechanosensory neurons to LegPNs in both FANC and MANC connectomes.

Post-synaptic partners of the LegPNs in MANC include pre-motor neurons corresponding to the LegPLs (IN03A073), which also make feedback and reciprocal connections. Although the exact synapse counts differ between datasets, the rank order of the most highly connected partners is similar, and the core local sensory-to-motor circuit around the LegPNs is also segmentally repeated in MANC, confirming this central conclusion of the FANC circuit analysis presented above (Table S1). Furthermore, we were able to match the LegPCs identified in FANC to homologous neurons in MANC based on similar morphology and shared connectivity to LegPNs and motor neurons (Figure 3E). Our assignment agrees with that of Marin et al.,34 who had identified these neurons as a serial set 12,289 from the hemi-lineage group IN10B014 (Table S1). The T2 instances of these neurons were not found to project commissurally in MANC either, thus supporting the difference in circuit connectivity in T2 that we had observed in FANC.

Identification of the LegPNs in the annotated MANC connectome allowed us to investigate how the LegPNs themselves might transmit information across the midline to recruit “helper” leg combinations. For example, T1L-LegPN could make most of its synaptic outputs onto neurons in the T1L neuropil, but some of its synapses were in T1R, T2L, and T2R neuropils (Figure 3A). The LegPNs do not synapse directly onto leg motor neurons, but they make indirect connections through both excitatory (cholinergic) and inhibitory (GABAergic) post-synaptic partners (Figure S4A and Table S1). Recent annotation of the leg motor neurons in the FANC33 and MANC60 connectomes allowed us to evaluate which motor neurons receive ipsilateral and contralateral LegPN and LegPC connections, but understanding the architecture for recruiting specific combinations of “helper” legs will require further biomechanical modeling and behavioral tests.

Codex61 searches for LegPN and LegPC neurons in the BANC62 and MCNS63 female and male brain and VNC connectomes showed six neurons of each type with similar morphology and connectivity. The top 10 input and output partners include the same neuronal types, and the class with the highest influence score on the LegPNs is leg mechanosensory neurons. These datasets also show sensorimotor circuits local to each leg neuropil with more commissural connections in T1 and T3 segments than in the T2 segment, in agreement with the initial FANC circuit analysis.

Discussion

Segmental organization

Segmentation is a fundamental organizational feature of both mammalian and insect development.23^,^64 In the nervous system, segments are often modular processing units, with local sensory inputs evoking local motor reflexes.42^,^65^,^66^,^67 Pre-motor circuits can control rhythmic and flexible limb movements,68^,^69 with segmental homologies and modifications allowing appendages to move differently.70^,^71 More complex behaviors can involve multiple segments and the circuits that coordinate among segments are not fully understood. Previous studies have explored the different intersegmental connections that enable forward and backward crawling in C. elegans, leech, and Drosophila larva,72^,^73^,^74^,^75^,^76 and components of circuits underlying coordination of forward vs. backward walking in limbed animals have been identified,77 but we lack a full understanding of how segmental circuits diverge to enable flexible limb coordination for their movement repertoire. Serially homologous neurons identified in the Drosophila adult VNC connectome34 are a way to initiate investigation of segmentally repeated circuits. Here, the leg rubbing subroutines of fly grooming and the connectome wiring diagrams allowed us to unpack conserved local sensorimotor circuits and uncover segmental differences in commissural projections that may allow recruitment of specific combinations of legs during grooming behavior.

Action selection

We noted that when all LegPNs are simultaneously activated, diverse leg rubbing movements are induced, but flies exhibit only one movement at a time, even in the absence of the brain. This highlights the VNC’s capacity to resolve conflicting commands21^,^78 and ensure actions are executed sequentially. In contrast, optogenetic activation of LegPCs in intact flies causes them to attempt to rub multiple legs at once (compare ethograms in Figures 1D and 4B and Video S2). We speculate that circuits mediating action selection may act between these command-like neurons. We examined connections between LegPNs, LegPLs, and LegPCs with the newly published Brake (BRK) neurons.79 The LegPCs, along with the LegPNs and other grooming command-like neurons (wPN120 and MGT22), synapse onto the BRK neurons (Table S1). These connections may contribute to the selection of legs that may be lifted and those that must remain immobile to stabilize balance and posture. There are also candidate inhibitory neurons post-synaptic to LegPNs and pre-synaptic to LegPCs, as suggested by the MANC connectome analyses (AN05B007 and others), that await behavioral experiments.

Commissural coordination

Leg rubbing requires immediate coordination among at least two legs. The post-synaptic connections of both LegPNs and LegPCs suggest how sensory stimulation on one leg initiates movement of that leg while also recruiting adjacent helper legs. While the majority of synapses from a given LegPN or LegPC are local—within their own leg neuropil—the LegPNs and T1 and T3 homologs of the LegPCs also send axon across the midline and between segments, making synaptic connections in other leg neuropils (Figure S4B). These connections could make excitatory and inhibitory pre-motor connections to coordinate multiple leg movements. The circuits observed in the connectome suggest mechanisms for leg recruitment with potential delay lines that could result in alternating extension/flexion cycles.

In contrast, thoracic sweeps can be ipsilateral. Activation of a single MGT command-like neuron evokes one leg to sweep that side of the body.22 The neuroanatomy of MGT corresponds with this: the thoracic mechanosensory neuron inputs and T3 pre-motor outputs synapse with an ipsilaterally restricted command-like neuron. The bilateral hind leg rubbing that follows thoracic sweeps requires contact between the leg and thorax, recruiting additional pre-motor circuits. Whether these include LegPNs, LegPCs, and LegPLs is not yet known.

Redundancy and complexity

In this study, we identified the LegPNs as neurons whose activity can induce leg rubbing behavior and explored their sensory and motor connections, using anatomical connectome datasets. This analysis showed a path from mechanosensory bristle neurons to the LegPNs and from the LegPNs to the leg motor neurons via LegPLs and LegPCs, consistent with this circuit playing a role in normal grooming behavior. The connectome analysis also revealed greater complexity—there are other pathways that could connect mechanosensory bristle neurons to leg motor neurons, indicating that the circuit we mapped may not be exclusively required. Indeed, silencing LegPNs, LegPLs, or LegPCs does not prevent leg rubbing during dust-induced grooming or reduces reflexive responses to leg bristle touch (Figures S4C and S4E), supporting redundancy. Even though the activation of both LegPNs and LegPCs can evoke leg rubbing behavior, they do not form an obligate series—silencing LegPCs while activating LegPNs did not suppress leg rubbing (Figure S4F). Therefore, additional circuits that can support leg rubbing behavior remain to be determined.

The roles of the other neurons highly connected to the LegPNs, both upstream and downstream, remain unknown. These partners include uncharacterized descending neurons and intersegmental ascending neurons, as well as local excitatory and inhibitory neurons (Table S1). The most highly connected partners of the LegPNs and those for which genetic reagents can be generated will be prioritized for future optogenetic activation and silencing experiments. Additionally, LegPNs are ascending neurons with brain projections to the SEZ—a region receiving mechanosensory inputs from the head and housing the dendrites of descending neurons related to anterior grooming.12^,^21^,^80^,^81^,^82 Analysis of the new BANC62 and male CNS63 connectomes that link the brain and nerve cord should facilitate future assessment of the role of these brain projections in coordinating the grooming behavior. Our behavioral screening and connectome analysis identified some components of neural circuits that contribute to targeted leg rubbing and demonstrated that there is more to be learned.

Limitations of the study

We identified the LegPNs by behavioral screening and the LegPCs and LegPLs through their connectivity. We were able to test the phenotypes of LegPC and LegPL activation because we could obtain genetic targeting reagents. None of these neuronal types showed strong silencing phenotypes, indicating that there are other neurons that can supply some redundant functionality to support leg rubbing and grooming. Anatomical analyses of the connectomes can suggest possibilities, but developing additional genetic tools for behavioral tests remain critical. The circuits that control flexible, coordinated, and rhythmic leg movements are not immediately obvious from their connectivity; thus, functional experiments will continue to be important.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Julie H. Simpson (jhsimpson@ucsb.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Primary data to reanalyze any behavior shown here are available from the lead contact upon request.
• •Representative code for grooming analysis is available on GitHub: https://github.com/AutomaticBehaviorRecognitionSystem/ABRS.
• •Neuronal reconstructions are publicly available through Codex and Neuroglancer websites, using annotation terms from Table S1.

Acknowledgments

For pre-publication access to the FANC VNC EM connectome and analysis tools, we thank Wei-Chung Allen Lee, John Tuthill, Jasper Phelps, and Brandon Mark. For automatic segmentation, we thank Zetta Ai LLC (Ran Lu, Nico Kemnitz, Kisuk Lee, Akhilesh Halageri, Manuel Castro, Dodam Ih, Sven Dorkenwald, Forrest Collman, Casey Schneider-Mizell, Derrick Brittan, Chris S. Jordan, and Thomas Macrina). We also thank the FANC Slack community for advice and discussion. For pre-publication access to the MANC EM connectome data and analysis, we thank the Janelia Research Campus FlyEM Project Team and the Cambridge Connectomics Community, especially Greg Jefferis, Lisa Marin, Sebastian Cachero, Tomke Stürner, and Kathi Eichler for sharing data and expertise. We thank members of the Simpson lab for valuable feedback, especially Diego De Alba and Varun Aniruddha Sane for help with MANC connectome comparisons in R and Cytoscape, Durafshan Sakeena Syed for advice on inhibitory neuron identification, and Carla Ladd for graphical abstract design. We also thank all the UCSB undergraduate students—Ethan Zhang, David McNeil, Aaron George, Aleena Ghumman, Andrew Roos, Avery Zinner, Christopher Denning, David Berry, Gabe Bello, Juan Gutierrez, Kai Thomas, Michelle Wan, Mansi Shah, Natalie Walker, Omeed Hashemi-Nejad, Shreya Prabu, Sofia Easton, and Tanya Sangani—who have contributed to proofreading and editing the LegPN connectivity. The study was supported by an 10.13039/100000001NSF Career award 1943276 and 10.13039/100000002NIH grants NS110866 and NS132900 to J.H.S.

Author contributions

L.G., N.Z., P.T., S.Y., and J.H.S. conceptualized the study; L.G., N.Z., P.T., S.Y., J.D., L.K., and J.H.S. performed experiments and analyzed data; L.G. and J.H.S. wrote the manuscript; L.G., N.Z., S.Y, and P.T. generated figures; J.H.S. acquired the funding and supervised the study.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesRabbit polyclonal anti GFPInvitrogenCat #A11122; RRID: AB_221569Rabbit monoclonal anti HACell signaling TechnologyCat #3724; RRID: AB_1549585Rat monoclonal anti FLAGNovus BiologicalsCat #NBP1-06712; RRID: AB_1625981Mouse monoclonal brp antibodyDSHBCat #4b_314866; RRID: AB_2314866Goat anti-rabbit Alexa Fluor 488InvitrogenCat #A27034; RRID: AB_2536097Goat anti-mouse Alexa Fluor 568InvitrogenCat #A11004; RRID: AB_2534072Goat anti-mouse Alexa Fluor 647InvitrogenCat #A-21236; RRID: AB_2535805Experimental models: Organisms/strainsCantonSBloomington Stock CenterRRID: BDSC_64349Control-spGal4: BPp65ADZp(attp40); BPZpGDBD(attp2)Bloomington Stock CenterRRID: BDSC_79603LegPNs split gal4: R17F11-p65ADZ (attp40); VT029514-ZpGDBD(attp2)Bloomington Stock CenterRRID: BDSC_68844RRID: BDSC_75062LegPNs LexA: R17F11-lexABloomington Stock CenterRRID: BDSC_52520LegPNs Gal4: VT029514-Gal4Vienna Drosophila Resource Center–LegPLs split gal4: R79C09-p65ADZ (attp40); R24D12-ZpGDBD(attp2)Bloomington Stock CenterRRID: BDSC_70815RRID: BDSC_68750LegPC split gal4 SS53029*: R24E12-p65ADZp (attP40);VT019345-ZpGdbd (attp2)Bloomington Stock CenterRRID: BDSC_8875720XUAS-CsChrimson-mVenus (attp18)Bloomington Stock CenterRRID: BDSC_5513413XLexAop-CsChrimson-mVenusBloomington Stock CenterRRID: BDSC_55138UAS-GtACR2 (attP2)Bloomington Stock CenterRRID: BDSC_9298710XUAS-IVS-eGFPKir2.1 (attP2)von Reyn et al.83N/A20XUAS-FRT>STOP>FRT-CsChrimson-mVenus (attP2)Shirangi et al.84N/AR57C10-Flp2::PESTNern et al.85N/AChat-lexABloomington Stock CenterRRID: BDSC_84379Vglut>FlpBloomington Stock CenterRRID: BDSC_84708GAD1-ADBloomington Stock CenterRRID: BDSC_60322LexAop2-mCD8::GFP, UAS-mCD8::RFPBloomington Stock CenterRRID: BDSC_32229Multicolor Flip-Out85: pBPhsFlp2::PEST in attP3;; pJFRC210-10XUAS-FRT>STOP>FRT-myr::smGFP-OLLAS in attP2, pJFRC201-10XUAS-FRT>STOP>FRT-myr::smGFP-HA in VK0005, pJFRC240-10XUAS-FRT>STOP>FRT-myr::smGFP-V5-THS-10XUAS-FRT>STOP>FRT-myr::smGFP-FLAG in su(Hw)attP1/ TM2*Bloomington Stock CenterRRID: BDSC_62118RRID: BDSC_64094Software and algorithms**Rhttps://www.r-project.org/RRID: SCR_001905Matlabhttps://www.mathworks.com/products/matlab.htmlRRID: SCR_001622Pythonhttps://www.python.org/RRID: SCR_008394Fijihttps://imagej.net/FijiRRID: SCR_002285Adobe Illustratorhttps://www.adobe.com/products/illustrator.htmlRRID: SCR_010279Vcodehttp://social.cs.uiuc.edu/projects/vcode.htmlN/AAutomatic Behavior Recognition System (ABRS)Ravbar et al.86N/AFANCPhelps et al., Azevedo et al.9^,^33N/AMANCMarin et al., Takemura et al.34^,^59N/A

Experimental model and study participant details

Fly husbandry

Flies were reared on standard cornmeal food at 25°C with 50% relative humidity and a 12hr light/dark cycle. Behavioral assessments were conducted on male flies aged 3-8 days. Optogenetic experiments involved collecting male adults shortly after eclosion, subjecting them to a 12-hour starvation period, and then transferred to the retinol food 3-5 days before testing. Retinol food contains 0.4mM all-trans-retinal. The flies were maintained in darkness until testing. Clonal experiments were initiated at 25°C: individual testing and subsequent anatomical characterization were performed after setting crosses.

Method details

Mechanical stimulation of leg bristle

Experiments employed 2-day old wild type CantonS male flies. Flies were anesthetized on ice and decapitated using fine-tip scissors. Decapitated flies were transferred to a 100mm Petri dish with Whatman filter paper, saturated with 200ul 1X Phosphate Buffered Saline (PBS) to recover for 1-2 hours at room temperature (n=21). Individual flies were transfer to a 60mm Petri dish under a dual stereo microscope (Motic DSK-500 Dual Stereo Microscope System). Behavior was recorded at 30Hz with an iPhone 11, connected by a microscope smartphone camara adapter attached to a microscope eyepiece. A single brush bristle was used to mechanically stimulate the femur and tibia bristles on each leg. A single stimulation trial consisted of 5 consecutive brush bristle sweeps. To ensure clarity in results, flies were only stimulated if they had been inactive, not grooming, for a minimum of 5 seconds. The sequence of leg stimulations was consistent: L1, L2, L3, R1, R2, R3, repeated 3 times to obtain 18 trials for each fly. Behavioral responses to the leg stimulation were scored manually. Only flies that displayed a reaction to the stimulation at least once were considered in leg usage analyses. The behavior responses are classified as leg rubbing if two or three legs come into contact with each other, and the legs move back and forth in a coordinated manner. For statistical considerations, we applied an all-or-none analysis approach without factoring in the duration of the action. For instance, if a fly exhibited two distinct reactions post a single stimulation, each response was assigned a value of 0.5. The response was assigned a value 1.0 if there was only one reaction so that the sum of all the reaction value equals to the number of responsive trials. The total percentage of each distinct reaction was calculated as the total value of a single reaction divided by the total number of responsive trials. For flies with LegPL or LegPN interneuron inhibition, each fly received 18 bristle deflection stimuli (3 per leg). The response rate was calculated as the number of stimuli that elicited a response divided by the total number of stimuli (18). For instance, if a fly responded to 9 stimuli with leg rubbing, the response rate would be 9/18, or 0.5.

Optogenetic experiments in free-walking flies

Males were anesthetized on ice and rested in the recording chamber for a minimum of 15 mins before testing. Light activation from below was achieved using custom-made LED panels (LXM2-PD01-0050, 625nm) with an intensity of 3.4 mW/cm^2^. For high-resolution recordings of individual LegPN activation, a single fly was placed inside a transparent 10-mm diameter quartz chamber. The recordings were captured from the ventral side of the fly. An FLDR-i132LA3 red ring light (626 nm) was used for optogenetics activation. Flies were flipped into fresh retinol food every 12h to prevent excessive humidity in the vial due to the light exposure. 30Hz videos were recorded by a FLIR Blackfly S USB 3 camera. Grooming movements were manually annotated by Vcode. Clones expressing in single LegPN were obtained by low-levels of Flp recombinase expressed from the R57C10 ElaV enhancer. The UAS>stop>csChrimson-mVenus enabled behavioral tests for selection of male flies displaying leg rubbing, followed by immunohistochemistry to determine clone location. We used VT029514-Gal4, rather than the splitGAL4 combination, because empirically we obtained a higher frequency of clones. To quantify leg flexion after LegPL activation: we measured the femur–tibia joint angle of the T1 left and right legs for each fly using ImageJ. Joint angles were extracted from images captured immediately before and 1 second after light activation of LegPL. The statistical significance for the joint angle before and after LegPL activation was assessed using a paired t-test.

Immunofluorescence and confocal imaging

For antibody staining, adult males were fixed rotating with 4% paraformaldehyde at room temperature for 2 hours. Subsequently, samples were dissected in 1xPBS, followed by three washes of 1 minute each in 1xPBST (1xPBS and 1% Triton X-100). CNS tissues were blocked with 5% normal goat serum (NGS) in PBST for 30mins at room temperature. Primary antibody staining was conducted overnight at 4°C on a nutator in 300ul 1xPBST with following dilutions: 5% NGS, rabbit polyclonal anti GFP (Invitrogen, 1:1000), mouse monoclonal anti brp (DSHB nc82, 1:200). Post-staining, samples were washed 4 times with 1xPBST for 15mins each. Secondary antibody staining was performed at room temperature on a nutator in the dark. CNS samples were incubated in 300ul 1xPBST with secondary antibodies with following dilutions: 5% NGS, goat anti-rabbit Alexa Flour 488 (Invitrogen, 1:500), goat anti-mouse Alexa Flour 568 (Invitrogen, 1:200). Afterwards, samples were washed again with 1xPBST 4 times for 15 mins each and mounted in VectaShield. All confocal images were taken on a Zeiss LSM710 microscope using sequential scanning mode and processed by Image J. Immunohistochemistry for Multicolor Flp clones was performed according to Nern et al.85

EM reconstruction

Neuronal skeletons were reconstructed using a serial section transmission electron microscope volume of an adult female ventral nerve cord, employing both auto-segmentation and manual proofreading within FANC. Key neuroanatomical features derived from confocal images, such as position, backbone orientation, and projection boundaries, were utilized to identify potential candidates for LegPNs.

For accurate analysis with minimized human bias, automatic synapse detection was employed to identify both upstream and downstream partners. Both LegPNs and their associated partners underwent proofreading within FANC. We designated preferred partners based on their connections and synapse numbers, serving as a proxy estimate for connection strength. Genetic reagents targeting LegPLs were computationally identified by generating Color-depth Maximum Intensity Projections (MIPs)87 from EM neuron skeletons and searching the MCFO library.88 The original protocols, available publicly on GitHub, were sourced from Jasper Phelps and Hideo Otsuna.

• •FANC auto-recon: https://github.com/htem/FANC_auto_recon/tree/main/colormips
• •ColorMIP Mask Search: https://github.com/JaneliaSciComp/ColorMIP_Mask_Search

To define pre-motor neurons downstream of LegPN, we assessed whether these downstream partners formed more than five synapses with leg motor neurons. Subsequently, we analyzed the connectivity of these pre-motor neurons to motor neurons. Cumulative connectivity was computed for motor neurons belonging to the same neuropils. We then compared the strength of pre-motor neuron projections to motor neurons across all six leg neuropils based on this cumulative connectivity. The EM connectivity data was processed using Navis Python and the Natverse R package.89

To further verify the EM connectivity data, we compared the FANC reconstructed data with neurons in MANC. We identified the 6 LegPNs homologs in MANC based on the known neuroanatomical features such as soma location, dendrite arborization, and backbone orientation characterized in the FANC connectome, and the neurons labeled in confocal images of the splitGAL4 line labeling LegPNs. Using the MANC connectivity matrix with the Neurprint interface,90 we analyzed the top 35 downstream partners of each LegPN. Candidates for LegPL, LegPC, and LegPI matches were evaluated by visualization in Neuroglancer, largely as described in.34 The Table S1 summarizes this matching analysis using stable neuronal identifiers.

Quantification and statistical analysis

Data analysis was performed in MATLAB and R. Mann-Whitney U test was used to evaluate significance. For progression analysis, behavioral probabilities were calculated every 3s (video 100 Hz video), or 6s (30Hz video). Data was plotted with ggplot2 R package.