We reconstructed all axons in the left ADMN using an EM dataset of the VNC of a female adult fly (FANC; Figure 1D; also see Methods; Azevedo et al., 2024; Phelps et al., 2021). For each automatically segmented neuron, we used the software interface Neuroglancer to manually proofread the major branches, as well as all branches that could be reliably attached (Methods). In the left ADMN, we identified 490 sensory axons and 14 motor axons. Axons were identified as sensory if they did not attach to a cell body in the VNC (Methods). The total number of axons is slightly higher than previously reported counts from cross sections of the wing nerve (455–465 axons, Edwards et al., 1978). Of these afferents, we classified 364 as wing margin bristle axons based on their ventral projections (Palka et al., 1979). Of the 126 non-bristle afferents, we identified 64 axon morphologies from published images of dye-fills (Appendix 1—table 1; Burt and Palka, 1982; Ghysen, 1980; Ghysen, 1978; Kays et al., 2014; Koh et al., 2014; Lu et al., 2012; Palka et al., 1986; Palka et al., 1979; Thistle et al., 2012; Whitlock and Palka, 1995). Of the 62 remaining axons previously unidentified in the literature, we identified sparse GAL4 lines in the FlyLight collection (Jenett et al., 2012) that labeled axon morphologies that resembled the reconstructed axons from FANC, crossed these lines to a fluorescent reporter, and then imaged the wing and wing hinge to visualize expression (Figure 1—figure supplement 1). Using this strategy, we successfully identified 50 of the 62 previously unidentified morphologies.

We reconstructed postsynaptic partners of sensory neurons until at least 70% of the output synapses from each sensory neuron were attached to proofread neurons (Figure 2A). Sensory axons make direct synapses onto motor neurons, other sensory neurons, VNC intrinsic neurons, and interneurons that ascend to the brain. To identify clusters of sensory axons with similar postsynaptic connectivity, we used a pairwise measure of cosine similarity, where a score of 1.0 indicates that the two neurons contact the same partners with the same proportion of synapses. We then ordered the neurons via agglomerative clustering, which revealed clusters of neurons with similar morphologies (Figure 2B). The cosine similarity of axon pairs within each cluster was significantly higher than across clusters (Figure 2B, inset (permutation test; 10,000 permutations, observed difference = 0.34, p<0.05)). Figure 2C–E shows the axon morphology of each cluster, organized by peripheral class. In the remainder of the paper, we focus on identifying the novel sensory neuron classes in Figure 2C.

Proximal CS axons are characterized by three branches: one short branch projects to the tectulum and two long branches project anteriorly to the brain and posteriorly to the haltere neuropil (Ghysen, 1980). There are ~36 proximal CS on the wing, and we found 38 axons in the EM dataset that followed this pattern. Previous dye fills of distal CS revealed axons that do not ascend to the brain and instead send two processes to the posterior VNC (Ghysen, 1980). There are ~17 distal CS on the wing, and we found 15 axons that match this pattern. We also identified five ascending axons that resemble the small CS morphology, although they are missing a posterior branch. Overall, our comprehensive reconstruction revealed many morphological subgroups with overlapping postsynaptic partners, suggesting a high degree of integration within wing sensorimotor circuits.

Due to the need for rapid sensory feedback necessary for flight control, we were especially interested in identifying axons with monosynaptic connections onto wing motor neurons. We found that 34 of 62 previously uncharacterized axons synapse onto wing steering motor neurons (Figure 3A–B). Of these, one group of axons synapses directly onto the well-characterized b1 motor neuron, which innervates the b1 muscle to help stabilize pitch during flight (Whitehead et al., 2022). The b1 motor neuron fires on nearly every wing stroke, and input from wing afferents sets the phase of its activation (Fayyazuddin and Dickinson, 1999; Heide and Götz, 1996). Notably, the input to the b1 motor neuron from ipsilateral wing and haltere axons is clustered around the putative spike initiation zone (Figure 3B), as has previously been reported based on axonal spatial overlap (Chan and Dickinson, 1996). This synaptic organization may be a structural mechanism for facilitating rapid modulation of b1 activity based on sensory feedback.

The wing sensory axons that synapse onto the b1 motor neuron have not been previously characterized (Figure 3C). They terminate shortly after entering the VNC and do not branch more extensively. We observed an unexpected ultrastructural feature in these axons: their terminals contain very densely packed mitochondria compared to other cells (Figure 3D). This feature is also present in interneurons that make electrical connections (Trimarchi and Murphey, 1997) to the b1 motor neuron (Figure 3E). We speculate on this ultrastructure further in the Discussion.

To identify the peripheral identity of these axons (Figure 3F), we found a driver line that labeled this population (Figure 3G; Jenett et al., 2012) and crossed it to a fluorescent reporter. Imaging the wing revealed that the population of short axons that directly synapse onto a subset of wing steering motor neurons originates from a field of CS on the tegula (Figure 3H). This finding suggests that the tegula may play a previously underappreciated role in flight control, particularly in regulating the tonically firing muscle b1 and a tonically firing muscle from another motor module, i2 (Figure 3A).

A group of five axons branch dorsally and ventrally as they enter the VNC and cross the midline (Figure 4A). We identified a sparse driver line that labeled these neurons (Figure 4B) and found that their corresponding cell bodies were in the tegula. There is a row of short stubby hairs on the dorsal face of the tegula (Figure 4C), resembling the HPs found at leg joints that are activated at extreme joint positions (Pratt et al., 2026; Pringle, 1938b; Trimarchi et al., 1999). The role of this tegula HP in wing sensation is unknown, although their peripheral morphology had been previously described (Fudalewicz-Niemczyk, 1963).

Two groups of axons with similar postsynaptic partners branch broadly throughout the tectulum without crossing the midline (Figure 5A). Using a sparse driver line (Figure 5B), we found that the cell bodies belong to an internal structure within the tegula (Figure 5C). We counted ~14 neurons in this structure, which separate into two bundles that attach to different points on the distal, anterior end of the tegula. Neurons in the chordotonal organ (CO) in the tegula are not labeled by iav-GAL4, unlike many other COs elsewhere in the body (Figure 5D; Kwon et al., 2010). They do, however, have actin-rich cap cells that are characteristic of other COs and not present in other mechanosensory neurons such as CS or HPs (Figure 5E; Field and Matheson, 1998). A subset of tegula CO neurons are labeled by NompC-GAL4, suggesting that they do express mechanosensory channels other than iav (Figure 5F). Because we only identified nine axons, vs. 14 cell bodies, some tegula CO axons might have a different morphology.

Axons with three distinct morphologies share a characteristic branch that passes laterally through the wing neuropil (Figure 6A). One group of axons extends a long process into the haltere neuropil, and another crosses the midline. By imaging sparse driver lines, we found that these axons come from neurons that make up a CO in the radius (Figure 6B–C). These neurons are distinguishable from the CS neurons in the radius because they do not have a dendrite that reaches toward the surface of the vein to innervate a dome. Instead, the cell bodies sit on the posterior side of the radius, and their dendrites and cap cells insert on the anterior side of the radius. These neurons all attach to the same point on the wing vein (Figure 6D), so they are likely subject to the same mechanical forces, allowing the CO to send parallel information to multiple regions of the VNC.

Five axons each extend a single process through the dorsal tectulum, and two of the axons cross the midline (Figure 7A). These axons originate from a cluster of five neurons in the thorax beneath the wing hinge near the parascutal shelf, just medial to the anterior nodal wing process (Figure 7B–C). Other than the three CS on the anterior nodal wing process, these are the only cells near the wing hinge labeled by the ChAT-GAL4 driver line, which targets nearly all peripheral sensory neurons (Figure 7C; Yasuyama and Salvaterra, 1999). As with all the anatomically defined populations of axons, the function of these novel wing hinge sensory neurons will require physiological measurements, but based on their location, they may signal wing opening and closing. We found no evidence for sensory neurons innervating pterale C (Figure 7D), a wing hinge sclerite that was previously thought to contain sensory receptors (Miyan and Ewing, 1984), although axons from the radius travel directly beneath pterale C.

Previous work uncovered morphological diversity across CS axons — subsets of CS axons follow different tracts and some ascend to the brain while others do not (Palka et al., 1979). It was unclear, however, whether there is morphological diversity of axons that originate from the same CS field. A field of CS is defined as having the same size, circularity, orientation, and general location on the wing (Cole and Palka, 1982). Single CS can be either rapidly or slowly adapting (Dickinson and Palka, 1987), but whether all CS within a field have the same firing properties remains unknown. It is also not known whether all CS within a field connect to the same postsynaptic partners. Answering this question could provide insight into the function of spatially clustered CS, for example, whether they underlie a population code or transmit signals in parallel to distinct downstream circuits.

To determine the morphological similarity of axons that innervate CS in the same field, we identified GAL4 driver lines that sparsely expressed in the wing nerve (less than five ADMN axons) and imaged their expression in the wing. We found many lines that label subsets of CS across multiple fields (Appendix 1—table 3) and fortuitously found three driver lines that label one to two separate CS in the ventral radius C field (v.Rad.C, Figure 8A). For these three lines, we compared VNC expression of axons from the ADMN using the FlyLight Multi-Color Flp-Out (MCFO) collection (Figure 8B–D; Meissner et al., 2023). In a driver line that expresses in two v.Rad.C neurons (CS 2 and 4), the VNC contains two distinct axon morphologies originating from the ADMN nerve (Figure 8D, row 1). Both axons possess a process that ascends to the brain, but one also projects down to the haltere neuropil. In a second driver line that also labels the second CS in v.Rad.C, we observed the same axon morphology that ascends to the brain but does not reach the haltere neuropil (Figure 8D, row 2). A third driver line that expresses in the third CS in v.Rad.C contains a non-ascending wing axon with two posterior projections (Figure 8D, row 3). These results show that neurons that innervate adjacent campaniform sensilla within the same field can have different axon morphologies. Each of the three v.Rad.C axons falls into a cluster of morphologically similar axons that connect to similar postsynaptic neurons (Figure 8E, clusters 2, 20, and 21, respectively, from Figure 2). The additional axons in each cluster likely originate from CS in other fields, on different parts of the wing. Overall, our results suggest that adjacent CS neurons in the same field connect to different target neurons in the VNC, and that the CS from different fields can connect to common postsynaptic targets.

Sensory neurons in the Drosophila wing have been a useful model system for investigating the relative contributions of intrinsic regulation vs. extrinsic signaling in the determination of axon morphology (Ghysen, 1980; Palka et al., 1979). Past work identified the axonal morphologies of wing CS and differences in the genetic programs guiding axon development for CS in fields and single CS (Ghysen, 1980; Palka et al., 1979). Here, we build on this knowledge using cell-type-specific genetic tools and connectomics to create a more complete map of the wing sensory apparatus and central sensorimotor circuits. By reconstructing axons from an electron microscopy volume, we found novel wing sensory axon morphologies, many of which originated from internal sensory structures that were inaccessible with previous dye-fill techniques, such as the previously uncharacterized CO axons.

Most wing CSs are organized into fields, in which domes of roughly the same size and shape cluster together. By finding sparse driver lines that label only one or two CS in the ventral radius C field (v.Rad.C), we found that CS in the same field can have distinct axon morphologies. This finding is consistent with our unsuccessful attempts to build split-GAL4 driver lines that specifically label CS from a single field by intersecting lines that label the same axon morphology. This organization may offer a functional advantage by sending similar signals to multiple regions of the CNS in parallel; viewed from a different perspective, postsynaptic neurons may rapidly integrate information across different CS fields. One possibility is that CS axon morphology is more closely linked to neuronal intrinsic properties (e.g., slowly vs. rapidly adapting). We hypothesize that the morphological clusters we identified (Figure 8D–E) are similarly tuned CS neurons distributed throughout different fields across the wing. A concurrent study focused on the central and peripheral organization of haltere mechanosensory neurons found that common central axon morphologies map across multiple CS fields on the haltere (Dhawan et al., 2026), consistent with our prediction for the wing.

One advantage of the dye-fill technique is that it is sometimes feasible to repeatedly label the same exact neuron across individuals. By comparing axon morphologies across individuals, researchers have found differences in small branches that extend from the primary neurite of the neuron (Kays et al., 2014; Palka et al., 1986). As more connectomic datasets become available, we will be able to ascertain if this morphological variation is also reflected by variation in downstream connectivity. Understanding relationships between morphology and connectivity across individuals, and even across species, will provide a framework for deciphering the developmental logic that governs the formation of sensorimotor circuits.

Although it is the most comprehensive to date, our atlas of wing sensory neurons is not complete. There were six axon morphologies (12 total axons) from the connectome that we could not reliably map to peripheral structures. There were also several peripheral structures whose axon morphologies we could not identify, such as the three CS on the anterior nodal wing process and the two large CS on the tegula. Furthermore, some of the uncharacterized axon morphologies likely belong to the tegula and radius COs, both of which had more cell bodies than identified axons. These gaps highlight the need for complementary approaches, such as combining small-scale experimental approaches with large-scale comprehensive datasets, to fully characterize the wing’s sensory landscape.

The 3D reconstructed axons are from the FANC dataset (Phelps et al., 2021), for details on segmentation, see Azevedo et al., 2024. Only the left wing afferents were analyzed due to damage to the right side ADMN (Azevedo et al., Extended Data Figure 4). Following automatic segmentation, neurons were proofread to include primary neurites and as many branches as could confidently be reattached. Neurons were annotated using CAVE (Dorkenwald et al., 2025). Depth-colored reconstructions were created using braincircuits.io (Azevedo et al., 2024).

To group axons by similar connectivity, we computed the cosine similarity of synaptic weights onto postsynaptic partners. We included fragments (9.7% of total output synapses) and used a three-synapse threshold for connections. Cosine similarity and agglomerative clustering were computed with the Python library Scikit-learn (cosine_similarity, AgglomerativeClustering, and dendrogram packages) (Pedregosa et al., 2011). A permutation test to compare within- and between-cluster similarity was computed with the Python library SciPy (Virtanen et al., 2020). Information on synapse location predictions and error can be found in Azevedo et al., 2024.

We used Drosophila melanogaster raised on standard cornmeal, molasses, and yeast medium at 25 °C in a 14:10 hr light:dark cycle. We used female flies 2–7 days post-eclosion for imaging. The genetic driver lines screened are listed in Appendix 1—table 3. Originally, they were chosen by manually looking through the FlyLight database for driver lines with sparse expression in the ADMN. Later, the braincircuits.io ‘genetic lines matching’ tool was used to screen driver lines by inputting segment IDs for particular wing afferent morphologies, with the particular lines selected based on sparse ADMN expression.

The fly food recipe used was based on the Bloomington standard Cornmeal, Molasses, and Yeast Medium recipe, which can be found at https://bdsc.indiana.edu/information/recipes/molassesfood.html. Our recipe had only slightly different antifungal ingredients and included tegosept, propionic acid, and phosphoric acid.

Mounted wings and wing hinges were imaged on a Confocal Olympus FV1000. Images were processed in FIJI (Schindelin et al., 2012).

See Appendix 1—table 2 for a list of references we used to identify peripheral structures along the wing and near the wing hinge (Cole and Palka, 1982; Dinges et al., 2021; Fudalewicz-Niemczyk, 1963; Hartenstein and Posakony, 1989; Hertweck, 1931). Sensory structures were identified from confocal image stacks by closely scrutinizing the images to see exactly where GFP-labeled neurons were in relation to landmarks, such as wing veins and sclerites. Campaniform sensilla were the most straightforward sensory structures to identify thanks to a comprehensive atlas (Dinges et al., 2021). The chordotonal organs were identified by their actin-rich attachment cells labeled by phalloidin. The structure on the tegula was identified as a HP due to the appearance of the hairs.