Introduction

The central complex (CX) of the adult Drosophila melanogaster brain consists of approximately 2,800 cells that have been divided into 257 cell types based on morphology and connectivity (Scheffer et al., 2020; Hulse et al. 2021; Wolff et al., 2015). These cell types are themselves organized into a set of highly structured neuropils (see Figure 1). The CX plays a key role in the flow of information between sensory inputs and motor outputs. It is particularly important in orientation and navigation, and much progress has been made in defining the cell types and circuits involved in these behaviors (reviewed in Fisher 2022; Green and Maimon 2018; Keram 2022; Turner-Evans and Jayaraman 2016). The CX also appears to participate in sleep and/or activity regulation (reviewed in Dubowy and Sehgal 2017; Shafer and Keene 2021). But these are unlikely to be the only functions performed by the CX.

Figure 1.

The connectome has provided a detailed wiring diagram of the CX, information that will be critical for understanding how it performs its functions (Hulse et al., 2021). However, the functions of most cell types in the CX remain unknown. Gaining this knowledge will likely require measuring and manipulating the activity of individual cell types. Such experiments are greatly facilitated by the availability of cell type-specific genetic driver lines. Such lines also allow biochemical approaches for determining the neurotransmitters and neuropeptides used by individual cells. This is particularly relevant for the CX, which is one of the most peptidergic brain areas (Kahsai C Winther, 2011; Nässel and Zandawala, 2019). The roles of neuropeptide signaling in the CX are largely unexplored.

Results and Discussion

Generation and analysis of split-GAL4 lines for CX cell types

We generated cell type-specific split-GAL4 lines for CX cell types using the same general approach that we previously applied to the mushroom body (Aso et al., 2014a; Rubin and Aso, 2023) and the visual system (Tuthill et al., 2013; Wu et al., 2016; Nern et al., 2024); see methods for details. In a previous report, we described the generation of split-GAL4 lines for cell types innervating the protocerebral bridge (PB), noduli (NO) and asymmetrical body (AB) (Wolff and Rubin 2018). Here we extend this work to the rest of the CX and include some improved lines for the PB, NO and AB.

Figure 2 and Figure 2—figure supplements 1-4 show the expression of 52 new split-GAL4 lines with strong GAL4 expression that is largely limited to the cell type of interest. Together with the other lines generated in this study and our previous work, we generated high-quality lines for nearly one-third of CX cell types that were defined by analysis of the connectome (Hulse et al., 2021). We also generated lines of lesser quality for other cell types that in total bring overall coverage to more than three-quarters of CX cell types. These additional lines often show some combination of expression in more than one CX cell type, unwanted expression in other brain areas, or weak or stochastic expression. Supplementary File 1 lists the two best split-GAL4 lines we generated for each CX cell type with comments about their specificity as well as the enhancers used to construct them. Additional split-GAL4 lines used in the sleep and NP/NT studies are also included in this file. Images of all lines are shown at www.janelia.org/split-GAL4 and the original confocal stacks of key imaging data can be downloaded from that site. For a subset of lines, images revealing the morphology of individual cells using MCFO (Nern et al., 2015), and higher resolution images, are also available (see for examples, Figure 3 and Figure 3—figure supplement 1 for examples). Additional split-GAL4 lines that may be useful for further studies are listed in Supplementary File 2.

Figure 2.

Figure 3.

Neurotransmitter expression in CX cell types

To determine what neurotransmitters are used by the CX cell types, we carried out fluorescent in situ hybridization using EASI-FISH (Eddison and Irkhe, 2022; Close et al., 2024) on brains that also expressed GFP driven from a cell-type-specific split GAL4 line. In this way, we could determine what neurotransmitters were expressed in over 100 different CX cell types based on which members of a panel of diagnostic synthetic enzymes and transporters they expressed: for acetylcholine, ChAT (choline O-acetyltransferase; acetylcholine synthesis) and, in most cases, VAChT (vesicular acetylcholine transporter); for GABA, GAD1 (glutamate decarboxylase; GABA synthesis); for glutamate, vGlut (vesicular glutamate transporter); for dopamine, ple (tyrosine 3-monooxygenase; dopamine synthesis); for serotonin, SerT (serotonin transporter); for octopamine, Tbh (Tyramine β-hydroxylase; converts tyramine to octopamine); and for tyramine, Tdc2 (Tyrosine decarboxylase 2; converts tyrosine to tyramine) accompanied by lack of Tbh.

Figure 4 shows two examples of this approach. In panels A-D we provide evidence that the PFGs use the less common neurotransmitter, tyramine. Panels E-H show an example of apparent co-transmission. Here, the FB tangential neuron FB4K expresses RNAs suggesting it can release both acetylcholine and glutamate. Cases of co-transmission using two fast acting neurotransmitters have been described in many organisms (reviewed in Svensson et al., 2019) including Drosophila, but are rare and may be post-transcriptionally regulated (Chen et al., 2023). Our full results are summarized below, together with our analysis of neuropeptide expression in the same cell types.

Figure 4.

Methods for using machine learning to predict neurotransmitter from EM images show great promise (Eckstein et al., 2024). However, they are unlikely to fully replace the need for experimental determination and validation for two reasons. First, rarely used transmitters such as tyramine are problematic due to limited training data. Second, accurate prediction of co-transmission is challenging for current computational approaches.

Survey of neuropeptide and neuropeptide receptor expression in the adult central brain

Neuropeptides provide a parallel mode of communication to wired connections and can act over larger distances using volume transmission (reviewed in Nässel 2009; Bargmann and Marder, 2013). Neuropeptides are widely expressed in the CX (Kahsai et al., 2011; Nässel and Zandawala, 2019) and are likely to play important roles in its function. However, information on the expression of neuropeptides and their receptors is not provided by the connectome. To look for expression of neuropeptides in the CX, we took a curated list of 51 neuropeptide-encoding genes from FlyBase (FB2024_02, released April 23, 2024; Öztürk-Çolak et al., 2024) and eliminated 12 genes based on their not having been detected in RNA profiling studies of the adult brain. Trissin and Natalisin were added to the FlyBase list based on evidence summarized in Nässel and Zandawala (2019). We only examined a small subset of neuropeptide receptors, selecting those whose cognate neuropeptides we thought might play a role in the CX. We used EASI-FISH to determine the expression patterns of these genes in the adult central brain of females (Figures 5-7). The list of 41 neuropeptides and 18 neuropeptide receptors we explored is presented in Figure 5—figure supplement 1.

Figure 5.

The neuropeptide expression patterns we observed fell into two broad categories. Some neuropeptides, like those whose expression patterns are shown in Figure 5, appeared to be highly expressed in only a few relatively larger cells. Several of these neuropeptides—for example, SIFa (Terhzaz et al., 2007) and Dsk (Wu et al., 2020)—are expressed in broadly arborizing neurons that appear to deliver them to large areas of the brain and ventral nerve cord. In contrast, neuropeptides like those shown in Figure 6 appear to be expressed in dozens to hundreds of cells and appear poised to function by local volume transmission in multiple distinct circuits. NPF and SIFa appear to act in both these modes. As we show below, most of the neuropeptides shown in Figure 6 are expressed in the CX, each in distinct subsets of cell types.

Figure 6.

Neuropeptide receptors (Figure 7) are more broadly, but not uniformly, expressed. In cases where more than one receptor has been identified for a given neuropeptide, such as Dh44 (Figure 7G; reviewed in Lee et al., 2023) and Tk (Figure 7T; see Wohl et al., 2023), the different receptors have distinct expression pattens.

Figure 7.

Neuropeptide expression in CX cell types

We selected 17 neuropeptides whose transcripts were observed in cell bodies located in the same general brain areas as those of the intrinsic cells of the CX (see Figure 5—figure supplement 1). We used probes for these 17 genes to perform EASI-FISH on brains that also expressed GFP in a specific cell type. In this way, we could score neuropeptide expression in individual cell types as we had done for neurotransmitters. Figure 8 shows examples of this approach.

Figure 8.

Figure 9 presents a summary table of neurotransmitter and neuropeptide use by individual cell types based on our EASI-FISH results. Supplementary File 1 contains a list of the individual stable split lines that were used for each cell type and how they were scored. We also characterized six CX cell types by RNA profiling (Figure 9—figure supplements 1 and 2) and additional RNA profiling of CX cell types is provided in Epiney et al. (2024). These RNA profiling results are largely congruent with the EASI-FISH data and allow a comparison of transcript number and in situ signal strength.

Figure 9.

A few general features emerge from these data. First, more than half of the cell types assayed express a neuropeptide. This frequency is perhaps not surprising given that the CX is considered one of the most peptidergic regions of the adult brain (Kahsai et al., 2011; Nässel and Zandawala, 2019); nevertheless, the fraction of cells expressing an NP appears to be several fold higher in the CX than observed in the adult brain as a whole, as judged by single cell RNA profiling studies (Davies et al., 2018). Second, in every case cell types expressing a neuropeptide also express a small molecule neurotransmitter (see Nassel 2018 for a discussion of other cases of co-expression). This co-transmitter is most often acetylcholine or glutamate, but we observed cases of GABA, dopamine, tyramine, octopamine, and serotonin. Third, co-transmission of two small molecule, fast-acting transmitters does occur but is rare. Conversely, co-transmission of a fast-acting transmitter and a modulatory transmitter such as serotonin is common: nine of ten cell types expressing the serotonin transporter also appear to express another small transmitter, most often glutamate or acetylcholine. Octopamine is often, but not always, co-expressed with glutamate (see also Sherer et al., 2020).

Screen for cell types whose activation modifies sleep

Sleep is a complex behavior that has been widely studied in Drosophila (reviewed in Dubowy and Sehgal 2017; Shafer and Keene 2021). The phenotypic description of sleep and its relationship to activity, as well as the cell types that play a role in sleep regulation are under active study in many labs. The CX has been documented to be a significant brain region for sleep regulation. But many cell types in the CX have never been assayed for a role in sleep due to the lack of suitable genetic reagents. Therefore, we used our genetic drivers for CX cell types to screen for those whose activation by thermogenetics or optogenetics strongly influenced sleep or activity. As described in methods, we used three metrics: sleep duration; P(Doze), the probability that an active fly will stop moving; and P(Wake), the probability that a stationary fly will start moving. These assays were carried out over several years in parallel with our building the collection of lines, so many of the lines we assayed did not make it into our final collection of selected lines. Conversely, we did not assay all our best lines as many only became available after our behavior experiments were completed.

Our screen identified several cell types not previously associated with sleep and/or activity regulation. For example, hDeltaF was found to be strongly wake promoting (Figure 10). We also identified PEN_b (Figure 1—figure supplement 1), PFGs (Figure 1—figure supplement 2), EL (Figure 1—figure supplement 3) and hDeltaK (Figure 1—figure supplement 4) as likely to play a role. In most of these cases, we were able to assay multiple independent driver lines for the cell type. We also assayed several lines that each contained a mixture of dorsal FB cell types (Figure 1—figure supplement 5) but were otherwise free of contaminating brain or VNC expression. In addition to intrinsic components of the CX, we evaluated several cell types that, based on the connectome, we thought likely to convey information from the circadian clock to the CX. Figure 11 (SMP368) and Figure 11—figure supplement 1 (SMP531) present two such cases of strongly wake promoting cell types.

Figure 10.

Figure 11.

Results for lines not discussed in detail in the main paper are provided as Supplementary Files. Supplementary Files 3 and 4 give results for 600 split-GAL4 lines assayed by thermogenetic activation with TRPA1 in both males and females, respectively. Supplementary Files 5 (males) and 6 (females) present results on over 200 lines, selected based on the results of thermogenetic activation, that were also assayed by optogenetic activation with CsChrimson. Images of the expression patterns of these lines and their genotypes can be found at https://flylight-raw.janelia.org/. The set of lines is anatomically biased with the dorsal FB overrepresented, and many lines contain multiple cell types and non-CX expression. Nevertheless, we believe these data might be useful as a starting point for further exploration.

The goal of our screen was to identify candidate cell types that warranted further study. While we identified several new potential sleep regulating cell types within CX, we did not perform the additional characterization needed to elucidate the roles of these cell types. For example, we did not examine the effects of inhibiting their function. Nor did we examine parameters such as arousal thresholds or recovery from sleep deprivation. Finally, by using only a 24-hour activation protocol we might have missed features only observable in shorter activation protocols. On the other hand, we assayed all lines in both males and females with identical genetic and environmental manipulations and many lines were evaluated by both optogenetic and thermogenetic activation. We observed that many lines showed phenotypes that differed between sexes, even though the expression patterns of the split-GAL4 lines did not obviously differ across sexes. Lines that showed strong effects generally did so with both activation modes and with both beam crossing-based activity measures and video-based locomotion tracking.

Connections between the CX and the circadian clock

Not surprisingly, the connectome reveals that many of the intrinsic CX cell types with sleep phenotypes are connected by wired pathways (Figure 12 and Figure 12—figure supplement 1). The connectome also suggested pathways from the circadian clock to the CX. Some of these have been previously noted. Links between clock output DN1 neurons to the ExR1 have been described in Lamaze et al. (2018) and Guo et al. (2018), and Liang et al. (2019) described a connection from the clock to ExR2 (PPM3) dopaminergic neurons. We found two SMP cell types, SMP368 and SMP531, that were very strongly wake-promoting when activated suggesting they might be components of previously undescribed wired pathways from the clock to the CX.

Figure 12.

In addition to these wired pathways, our work supports the possibility of signaling from the clock over considerable distances to the CX using neuropeptides. Our RNA profiling of ER5 cells (Figure 9— figure supplement 1), which are known to be regulators of sleep and sleep homeostasis (Liu et al., 2016), revealed expression of receptors for both PDF and Dh44. The presence of the PDF receptor in ER5 cells was suggested by prior work (Im C Taghert, 2010; Parisky et al., 2008; Pirez et al., 2013). We confirmed these observations and showed that ER5 cells make Dh31 (video 1). Dh44 has been implicated as a clock output that regulates locomotor activity rhythms (Barber et al., 2021; Cavanaugh et al., 2014) and the DH44R1 receptor has been shown to function in sleep regulation in non-CX cells (King et al., 2017). However, the presence of the Dh44R2 receptor in the EB was unexpected. Dh31 is expressed by many cells in the fly brain (see Figure 6F) including DN1s (Kunst et al., 2014) and has been shown to play a role in sleep regulation; the cellular targets of Dh31 released from ER5 are unknown, however previous work (Goda et al., 2017; Mertens et al., 2005) has shown that Dh31 can activate the PDF receptor raising the possibility of autocrine signaling. Andreani et al. (2022) also showed a functional link between the clock and ER5 cells, but the circuit mechanism was not elucidated.

Concluding Remarks

We provide a greatly enhanced set of genetic reagents for manipulating the intrinsic cell types of the CX that will be instrumental in fully elucidating the many functions of the CX. We illustrate their use in discovering cell types involved in activity regulation, and uncovered new potential wired and peptidergic connections between the circadian clock and the CX. We surveyed neuropeptide and neuropeptide receptor gene expression in the adult central brain. Neuropeptides fell into two broad categories, those at are expressed in only a few cells and those that are expressed in dozens to hundreds of cells. We observed that neuropeptide receptor genes were much more broadly expressed than those of their cognate neuropeptides. Finally, we generated the largest available dataset of co-expression of neuropeptides and neurotransmitters in identified cell types. Unexpectedly, we found that all neuropeptide-expressing cell types also expressed a small neurotransmitter. Our data reveal the pervasive potential for peptidergic communication within the CX—more than half of the cell types we examined expressed a neuropeptide and one-third of those expressed multiple neuropeptides.

Materials and methods

Generation of split-GAL4 lines

Split-GAL4 lines were generated as previously described (Dionne et al. 2018). Databases of expression patterns generated in the adult brain by single genomic fragments cloned upstream of GAL4 (Jennet et al., 2012; Tirian et al., 2012) were manually or computationally (Meissner et al., 2023) screened. Individual enhancers that showed expression in the desired cell type were then cloned into vectors that contained either the DNA-binding or activation domain of GAL4 (Luan et al, 2008; Pfeiffer et al., 2012). These constructs were combined in the same individual and screened for expression in the desired cell type by confocal imaging. Over 15,000 such screening crosses were performed to generate the new split-GAL4 lines reported here. Successful constructs were made into stable lines.

The lines listed in Supplementary File 1 are currently being maintained at Janelia and the majority of these have also been deposited in the Bloomington Drosophila Stock Center.

Characterization of split-GAL4 lines

Lines were characterized by confocal imaging of the entire expression pattern in the brain and VNC at 20X. Most lines were also imaged at higher magnification (63x) and/or subjected to stochastic labelling (MCFO; Nern et al., 2015) to reveal the morphology of individual cells. Split-GAL4 images are shown as MIPs after alignment to JRC2018 (Bogovic et al., 2020). Over 1,800 confocal stacks derived from over 450 lines generated during this work are presented in, and can be downloaded from, an on-line database (janelia.org/split-GAL4). Images for the additional lines used in the sleep screen are available from flylight-raw.janelia.org.

Determining the correspondence between the cell-types present in each split-GAL4 line and those described in the connectome (Hulse et al., 2021) was based solely on morphology. Even when assigning correspondence between cells in two different connectomes, where information on connectivity can also be employed, the process is not always straightforward (see Schlegel et al. 2024). Because of the similarity in morphology of many of the CX cell types it was often challenging to assign correspondence to the cell types defined by connectome analysis. For this reason, we rated our confidence in our assignments as Confident, Probable or Candidate and include this information for each line at janelia.org/split-GAL4. To be considered confident, we judged our opinion had a >95% chance of being correct. Such assignments were generally only possible for cell types which had morphological features clearly distinct from those in other cell types. Most assignments were rated as Probable indicating 70-95% confidence. Lines whose cell type assignments are listed as Probable have been rigorously examined and the assignments are the most accurate that the available data allow. In the absence of single-cell data available in MCFO brains (the case for many lines) or additional data (for example, physiological data on connectivity), cell types that are morphologically very similar cannot be distinguished with complete confidence. Lines whose cell type assignments are listed as Candidate are even less certain (30-70% confidence).

RNA in situ hybridization

Adult females (5 to 7 days post eclosion) were expanded, probed and imaged using the EASI-FISH method as described in (Eddison and Ihrke, 2022; Close et al., under review). The oligo probes and HCR hairpins were designed by, and purchased from, Molecular Instruments, Inc. Imaging was performed on a Zeiss Z7 microscope equipped with a 20X objective. Laser power and exposure time was optimized to maximize the signal-to-noise ratio.

RNA profiling

The data shown in Figure 9—figure supplement 1 were generated as described in Aso et al. 2019. See NCBI Gene Expression Omnibus (accession number GSE271123) for the raw data and additional details.

Sleep measurement and analysis: Thermogenetic activation screen

Split-GAL4 flies were crossed to 10X UAS-dTrpA1 (attP16) (Hamada et al. 2008) and maintained at 21–22°C in vials containing standard dextrose-based media (7.9g Agar, 27.5g Yeast, 52g cornmeal, 110g dextrose, 8.75 ml 20% Tegosept and 2 ml propionic acid/ liter).

Virgin female progeny or male progeny (as specified in the figures), 3–7 days post-eclosion, (n = 16-32/trial) were placed in 65 mm × 5 mm transparent plastic tubes with standard cornmeal dextrose agar media and placed in a Drosophila Activity Monitoring system (Trikinetics). Food composition was kept consistent between rearing and experimentation. Locomotor activity data were collected in 1-min bins for 5-7 days. Activity monitors were maintained with a 12 hr:12 hr light–dark cycle at 40-65% relative humidity. Total 24-hr sleep amounts (daytime plus nighttime sleep/sleep duration), Pwake, and Pdoze were extracted from the locomotor data as described in (Donelson et al., 2012; Wiggin et al., 2020) using MATLAB-based SCAMP program. (Sitaraman et al., 2024; Vecsey et al., 2024)

Sleep duration was defined as 5 min or more of inactivity (Hendricks et al., 2000; Shaw et al., 2000). Representative sleep profiles were generated representing average (n = 24–32) sleep (min/30 min) for day 1 (baseline), day 2 (activation), and day 3 (recovery/post activation). In addition to permissive temperature controls, split-pBDPGAL4U /dTrpA1 were used as genotypic controls for hit detection. pBDPGAL4U (attP40, attP2), has enhancerless GAL4-AD construct and GAL4-DBD constructs inserted on chromosome II and III (Dionne et al., 2018), as is the case for split-GAL4 driver lines in behavioral assays. Each split-GAL4 line was tested at least twice in independent trials. For all screen hits, wake activity was calculated as the number of beam crossings/min when the fly was awake. Statistical comparisons between experimental and control genotypes were performed using Prism (Graphpad Inc) by Kruskal Wallis One way ANOVA followed by Dunn’s post-test. Pairwise comparisons between the empty (split-pBDPGAL4U) control and experimental lines were made using Mann-Whitney U test.

Sleep measurement and analysis: Optogenetic activation screen

Split-GAL4 flies were crossed to 20XUAS-CsChrimson-mVenus-trafficked (attP18) (BDSC:55134) and maintained at 21–25°C in vials containing standard cornmeal food supplemented with 0.2% retinal. Male and virgin female progeny were collected into separate vials containing standard cornmeal food with 0.4% added retinal and kept in a 25°C incubator on a 12:12 light:dark schedule for 3-5 days before loading. Typical sleep experiments lasted 6-7 days. Flies were loaded into 96 well plates or individual tubes using CO2 anesthesia. Flies were allowed to recover from anesthesia and acclimatize to the experimental chambers for 16-18 hours prior to starting the experiment.

The behavioral setup for video recording system was adapted from (Guo et al., 2016). Flies were briefly anesthetized and loaded into 96 well plates (Falcon^™ 96-Well, Non-Treated, Fisher Scientific Inc) containing 150 ml per well of 5% sucrose, 1% agarose and 0.4% retinal. The plates were covered with breathable sealing films. Small holes (one per well) were poked with fine forceps into the film to further ensure air exchange and prevent condensation. The entire setup was housed in an incubator to control light/dark conditions and temperature. The 96-well plates with flies were placed in holders, constantly illuminated from below using an 850 nm LED board (Smart Vision Lights Inc) and imaged from top using a FLIR Flea 3 camera (Edmund Optics Inc). 635 nm red light (for optogentic activation) was provided using an additional backlight, low levels of white light (to provide a light-dark cycle) were supplied from above. Optogenetic activation was for a 24 h period (starting in the morning at the same time as the white light was turned on for the day) and was delivered in pulses of 25 ms duration at 2 Hz frequency. Each experiment also included at least one full day without the red light preceding and following the activation day (matching the general design of the thermogenetic experiments).

Fly movement was tracked using single fly position tracker (GitHub - cgoina/pysolo-tools) and processed for sleep duration and other sleep parameters using SCAMP. In addition to the 5 min criteria, used to define total duration of sleep, P(Doze), the probability that an active fly will stop moving, and P(Wake), the probability that a stationary fly will start moving provide key additional measures of inactivity and activity and were included in our analyses (Wiggin et al., 2020). These sleep measures are presented in Figure 10 and 11 (and supplements) and supplementary files 3,4,5, and 6.

Supplemental Files for sleep phenotypes

The supplemental files present data on day1 (baseline) and day 2 (activation). Given the large effects of environmental conditions on activity, we calculated p values between experimental and control group within the same environmental conditions. However, we also present mean differences between the days as a complementary way to identify lines that modified sleep when activated.

Acknowledgements

We thank Y. Aso (SS32244), J. Goldammer (SS51128), and M. Ito (SS49931, SS48762) for providing GAL4 lines. Robert Ray performed the RNA profiling experiments summarized in Figure 9—figure supplements 1 and 2 and the meta-analysis of RNA profiling data to identify which neuropeptide genes were expressed in the adult brain. Marisa Dreher (Dreher Design Studios) performed connectomic analyses and figure generation. We thank Janelia’s Project Technical Resources led by Gudrun Ihrke for assistance: ME, NC; Kari Close and Yisheng He performed EASI-FISH experiments, and Claire Managan scored results; Dan Bushey helped with earlier versions of python scripts to collate sleep assay data sets. Jennifer Jeter imaged and scored EASI-FISH experiments. Janelia’s FlyLight Project Team and Project Pipeline Support team, especially Allison Vannan, Jennifer Jeter, Joanna Hausenfluck, Zachary Dorman, Kelley Lee, and Geoffrey Meissner, performed CNS dissections, staining, and imaging. Janelia’s Invertebrate Shared Resource and Scientific Computing contributed to stock generation and image processing, respectively. Geoffrey Meissner and Rob Svirskas contributed to the split-GAL4 website. Michael Kunst, Preeti Sareen and Michael Nitabach contributed to early screening of split-GAL4 lines for effects on sleep. Wyatt Korff helped with establishment of the 96-well sleep assay. Heather Dionne, Martin Reyes, Anisha Ali and Matthew Finger helped with conducting sleep experiments and analyzing data. We thank Brad Hulse and other colleagues for comments on earlier drafts of the manuscript and for helpful discussion. This work was supported by the Howard Hughes Medical Institute, NIH 2R15GM125073-03 (to D.S.), and NSF IOS 2042873 (to D.S.).

Additional information

Author Contributions

TW, ME, DS and GMR conceived the study. TW, AN and GMR created split-GAL4 lines. TW analyzed lines for CX cell types, assigned them to cell types, and scored EASI-FISH experiments. ME designed, conducted, and scored EASI-FISH experiments. NC, PS and DS designed and conducted behavioral assays. GMR wrote the paper with input from all authors.

Additional files

Supplemental File 1.

Supplemental File 2.

Supplemental File 3.

Supplemental File 4.

Supplemental File 5.

Supplemental File 6.

Video 1.