The adult central brain (i.e. brain without optic lobes) is composed of neuronal and glial progeny derived from both T1 and T2 NBs. We first generated a transcriptomic atlas of all central brain neurons derived from both T1 and T2 NBs. The flies were of a genotype that gave permanent lineage tracing of T2 NB-derived progeny in the adult (Figure 1A and A'). This allowed us to first analyze all central brain T1- and T2-derived neurons and glia (Figure 1) and then focus on neurons and glia produced specifically by T2-derived lineages.

We dissociated nuclei from 7-day-old adult fly central brains with optic lobes removed using methods previously described (Li et al., 2022; McLaughlin et al., 2021). The cDNA libraries were then prepared from dissociated nuclei with split-pool-based barcoding for single-nuclei transcription profiling (Rosenberg et al., 2018). We sequenced a total of 30,699 nuclei at 589 median genes per nuclei from the T1+T2 central brain.

To distinguish T2-derived progeny from T1-derived progeny, we expressed worniu-Gal4,asense-Gal80 to drive expression of FLP recombinase (FLP) specifically in T2 NBs. This resulted in the flip out of a stop cassette and thus continuous expression of Gal4 under the ubiquitous actin5C enhancer. After removal of the stop cassette in T2 NBs, the actin5c-Gal4 can continuously drive expression of the reporter genes (RFP or GFP) and FLP in the T2-derived neuronal and glial progeny in adult brains (Figure 1A). We assigned 3,125 nuclei which have expression of RFP, GFP, or FLP as being derived from T2 lineages. We assigned 27,574 triple-negative nuclei as derived from the T1 lineages. We used Seurat for filtering, integrating, and clustering the atlas to provide transcriptionally unique cell clusters (Stuart et al., 2019). We used uniform manifold approximation and projection (UMAP)-based dimension reduction to visualize the 152 clusters (Figure 1B). The most enriched or least expressed genes (marker genes) for each cluster are shown in Supplementary file 1.

We used known markers (Supplementary file 2; Croset et al., 2018; Davie et al., 2018) to identify distinct cell types in the central brain, including glia, mushroom body neurons, olfactory projection neurons, clock neurons, Poxn+ neurons, serotonergic neurons, dopaminergic neurons, octopaminergic neurons, corazonergic neurons, hemocytes, and ocelli (Figure 1B; Supplementary file 1). We did not observe any cluster exclusively containing progeny from T1 or T2 NB lineages (Figure 1C, Figure 1—figure supplement 1). We conclude that both T1 and T2 generate transcriptionally similar cells, despite their different developmental origins. Alternatively, deeper sequencing may resolve a single cluster into two clusters with each derived from T1 or T2 lineages. We next searched for cluster-defining genes in the T2 enriched clusters (Figure 1C’) and found high expression of AstA (Allatostatin A), SerT (Serotonin transporter), Tk (Tachykinin), and Vmat (Vesicular monoamine transporter) (Figure 1D). These enriched marker genes suggest that AstA+ neurons, serotonergic neurons, and TK+ neurons may be primarily derived from T2 lineages.

We explored the central brain atlas for glial cell types and their gene expression. We focused on the 3409 glial nuclei from clusters that expressed the pan-glial marker repo (Campbell et al., 1994; Xiong et al., 1994) in the T1+T2 atlas (Figure 1E). Each cluster contained a mix of T1 and T2 glial progeny (Figure 1E’). We identified six known glial cell types (astrocytes, cortex, ensheathing, surface glial and the two subtypes: perineurial and subperineurial) based on canonical glial makers (Figure 1F; Supplementary file 3). Similar to a previous glial cell atlas (Lago-Baldaia et al., 2023), we found some glial subtypes (astrocytes, ensheathing, and subperineurial) mapping to multiple clusters (Figure 1E and F).We identified a glial cluster that expressed genes associated with extracellular matrix, vkg and Col4a1 (Figure 1F), which have previously been identified as pan surface glial markers (DeSalvo et al., 2014; Hindle and Bainton, 2014). Interestingly, the two surface glial subtypes, perineurial and subperineurial clusters, do not express these markers at cluster defining levels, and conversely the surface glia cluster 5 does not express perineurial or subperineurial specific makers (Figure 1F, Supplementary file 4). Differential gene expression analysis for all genes between T1 and T2 glial progeny did not show differences across any glial cell types or clusters (Supplementary file 5). We conclude that the adult central brain contains known glial cell types with no differences in gene expression between T1 and T2-derived glia.

We explored the diversity of T1- and T2-derived neurons by generating T1- and T2-specific cell atlases. We identified T1-derived neurons by bioinformatically excluding cells co-expressing T2-specific markers FLP+/GFP+/RFP+ plus repo+ glial clusters. We then generated a T1 neuron atlas containing 22,807 T1-derived neurons that form 114 clusters. Marker genes of each cluster are shown in Supplementary file 5. We identified the neurons that are known to be generated by T1 NBs, including MB neurons and olfactory projection neurons (Croset et al., 2018; Davie et al., 2018; Li et al., 2022; Supplementary file 2). We identified other T1-derived neurons, including clock neurons, Poxn+ neurons, and neurons that release dopamine, serotonin, octopamine/tyramine, and other neuropeptides (Figure 2—figure supplement 1, Supplementary file 6; Croset et al., 2018; Davie et al., 2018). We gathered a list of genes that represented the 10 most enriched genes from each cluster and calculated the scaled averaged expression of each gene from each cluster (Figure 2—figure supplement 1). Each unique combination of enriched genes could be referred to as cluster markers. We conclude that our T1-derived nuclei contain the expected abundance of neuronal diversity seen in previous scRNA-seq atlases (Croset et al., 2018; Davie et al., 2018; Li et al., 2022).

In our whole brain atlas described above, T2-derived cells represent a minor contribution to the atlas due to their low nuclei numbers relative to the T1-derived neurons (Raji and Potter, 2021). To generate a comprehensive T2 atlas, we needed to increase the percentage of T2-derived nuclei for RNAseq. We used fluorescence-activated cell sorting (FACS) to select nuclei that are labeled by T2-specific permanent lineage tracing (Figure 1A and B). We then selected the nuclei in silico that showed expression of UAS-transgenes (FLP, GFP, or RFP, see above) to eliminate contamination of T1 nuclei during FACS. In total, we sequenced 61,118 T2 nuclei. We included 3125 UAS transgene-labeled nuclei from the dissociated central brain without sorting from the T1+T2 atlas (see above). We used integration in Seurat to generate an atlas containing 64,243 nuclei that formed 198 clusters. T2 NBs have been estimated to produced ~5000 neurons/glia in the central brain (Ito et al., 2013; Yang et al., 2013), with ~1800 neurons in the central complex (Hulse et al., 2020; Schlegel et al., 2024), giving our atlas ~12 x coverage. We next filtered out the repo+ glial cell clusters to generate a T2 neuron cell atlas with 50,148 non-glial nuclei forming 161 clusters (Figure 2A). Marker genes of each cluster are shown in Supplementary file 8; Marsh, 2024.

We calculated the scaled average expression of the top 10 most enriched genes from each T2 neuron cluster (Supplementary file 8). Each set of genes could serve as cluster marker genes (Figure 2B). We conclude that each cluster within our T2 neuron atlas represents a transcriptionally unique cell type. For the remainder of the paper, we focus solely on the diversity of T2-derived neurons and glia.

An important aspect of neuronal identity and function is fast-acting neurotransmitter expression. We determined the neuronal populations that expressed seven fast-acting neurotransmitters: glutamatergic, cholinergic, GABAergic, tyraminergic, dopaminergic, serotonergic, and octopaminergic in T2-derived neurons (Figure 3A–G). In neurons, we found that cholinergic neurons were most abundant (21%), followed by glutamatergic neurons (12%), GABAergic neurons (9%), tyraminergic neurons (2.6%), dopaminergic neurons (1.2%), serotonergic neurons (1.2%), and octopaminergic neurons (0.7%). Additionally, 12% of neurons were co-expressing two or three neurotransmitters. The remaining 40% in our atlas were not expressing any neurotransmitter with a log normalized expression <2 (Figure 3H); this is similar to the ratios in the adult midbrain (Croset et al., 2018) and optic lobes (Konstantinides et al., 2022). The nuclei co-expressing two or more fast-acting neurotransmitters may reflect an authentic feature of these neurons, as see in other systems (see Discussion). Alternatively, our double- and triple-positive neurons may be false positives (see Discussion).

We next explored the complete T2 atlas for glial cell types and their gene expression. We sub-clustered from the T2 atlas for 12,315 glial nuclei from clusters that expressed the pan glial marker repo (Figure 4A; Campbell et al., 1994; Xiong et al., 1994). Similar to our T1+T2 glial atlas, we identified glial cell types: astrocytes, cortex, ensheathing, astrocyte-like, and two surface glial subtypes, perineurial and subperineurial, based on canonical glial makers (Figure 4B; Supplementary file 9). In line with our T1+T2 atlas and previous glia cell atlas (Lago-Baldaia et al., 2023), some subtypes mapped to several subclusters including ensheathing, astrocytes, and chiasm (Figure 4A–B).Interestingly, we identified chiasm glia that may represent the migratory chiasm glia generated from the DL1 lineage (Viktorin et al., 2013; see Discussion). We conclude that our T2 atlas contains the expected glial cell types.

When generating our atlas, we dissected both female and male adult brains that were processed as separate samples in order to differentiate the sex of nuclei in the snRNAseq atlas. We noticed that several cell types in the T2 glial atlas showed unequal number of male and female nuclei (Figure 5A–B and E–F). We identified sex-biased clusters by normalizing the number of input nuclei between male and female samples and compared the proportion of nuclei by sex within each cluster to identify sex-biased clusters (Figure 5B and F). We found that the T2 glial atlas contained two female enriched clusters: ensheathing/astrocyte (cluster 7) and chiasm (cluster 5), and one male enriched cluster: the astrocyte-like (cluster 3; Figure 5B). We determined differential genes expressed between male and female nuclei across all glia (Figure 5C; Supplementary file 10). We found female nuclei expressed higher levels of genes including the female-specific genes yp1, yp2, and yp3 (Figure 5C; Warren et al., 1979). Additionally, female nuclei were enriched for dsx (Supplementary file 10).Male glial nuclei expressed higher levels of genes including the male-specific genes lncRNA:rox1/2 and fru (Figure 5C; Supplementary file 10; Amrein and Axel, 1997; Meller et al., 1997; Ryner et al., 1996).We next looked at expression within the sex-biased clusters. We found similar results that male nuclei expressed high levels of the male-specific lncRNA:roX1/2 and low expression of the female-specific genes yp1, yp2, and yp3 compared to the female nuclei (Figure 5D; see Discussion). We conclude that male and female adult T2 glia have sex-specific differences in gene expression within the same glial cell type.

We next explored sex differences among neurons (Figure 5E–E’’). We identified 14 neuronal clusters with disproportionate amounts of male and female nuclei (Figure 5F). We first tested the differential genes expressed between male and female nuclei across all T2-derived neurons (Figure 5G; Supplementary file 11). Similar to the T2 glia sex differences, we found female nuclei expressed higher levels of the female-specific genes yp1, yp2, and yp3 (Figure 5G; Warren et al., 1979). Additionally, female nuclei were enriched for dsx (Supplementary file 11). Male neuronal nuclei expressed higher levels of the male-specific genes lncRNA:rox1/2 and fru (Figure 5G; Supplementary file 11; Amrein and Axel, 1997; Meller et al., 1997; Ryner et al., 1996). We next looked at expression levels within the sex-biased clusters. We found similar results that male nuclei expressed high levels of the male-specific lncRNA:roX1/2 and low expression of the female-specific genes yp1, yp2, and yp3 compared to the female nuclei (Figure 5H; see Discussion). We conclude that male and female adult T2 neurons have sex-specific differences in gene expression within the same neuronal subtype.

Neuropeptides are often used as markers to distinguish different neuronal populations. We examined whether individual neuropeptide-encoding genes were exclusively expressed within single clusters, representing single neuronal subtypes. Among the 49 neuropeptides in Drosophila, we identified 13 with enriched expression in a limited number of clusters, making them suitable as cluster-defining markers (Figure 6A). For all clusters significantly expressing cluster-defining neuropeptides, we assessed their co-expression with specific fast-acting neurotransmitters (Figure 6A). We found that all neuropeptide-expressing clusters, co-express one or more neurotransmitters (see Figure 3H).

These data raise several questions. What is the relationship between neuropeptide expression and cluster identity? Can two distinct clusters express the same neuropeptide? Can two neuropeptides be co-expressed by the same cluster? Our data show that all of these patterns exist. For example, in some cases, one cluster expresses two or more neuropeptide genes (e.g. clusters 100 and 128 express NPF and AstA); conversely, we also observe one neuropeptide expressed in multiple clusters (e.g. Ms, 2 clusters; Tk, 2 clusters; Figure 6A).

Next, we wanted to validate whether these neuropeptidergic cells indeed come from T2 lineages and target the CX. This required a method for specifically labeling of T2 peptidergic neurons, as antibody staining could be contaminated with T1-derived neurons. We developed a genetic approach to selectively visualize neuropeptide expression exclusively in neurons originating from T2 NB lineages. This technique involves driving FLP recombinase (FLP) under the control of the T2-specific driver wor-Gal4,ase-Gal80. The T2-specific FLP catalyzes the excision of a stop codon in the lexAop-FRT-stop-FRT-myr::gfp transgene, allowing existing neuropeptide-2A-LexA transgenes (Deng et al., 2019) to drive GFP expression specifically in neuropeptide-positive T2-derived neurons (Figure 6B). This method revealed the morphology of specific T2-derived peptidergic neurons.

We first assayed Ms+ neurons because Wolff et al., 2025 reported that at least two FB neurons, FB4Z and FB5R express Ms and they target to two distinct FB layers. However, we found that T2-derived Ms-2A-LexA-expressing neurons project to multiple layers of the dorsal fan-shaped body and the entire ellipsoid body, suggesting an unknown class of Ms+ neurons targeting to EB and/or FB (Figure 6C–C'’, Video 1). In our atlas, Ms is mainly expressed in clusters 157 and 160; however, we cannot distinguish the corresponding identity of each neuron.

Next, we assayed neurons in two clusters (100 and 128) that express the same two neuropeptides, AstA and NPF (Figure 6D–E). We wanted to determine if these clusters contain two cell types with similar gene expression, or a single cell type that co-expressed both neuropeptides. We used NPF-2A-lexA and AstA-2A-lexA to label each class of neurons. We found that they were generated at indistinguishable numbers of cells and projections: both had ~20 neurons projecting to a lateral domain of the ellipsoid body and the same two layers of the fan-shaped body (Figure 6D–E; Video 2 and Video 3). Whether these T2-derived NPF and AstA neurons are indeed the same would require antibody validation.

We investigated whether specific transcription factors (TFs), or TF combinatorial codes, might correlate with the expression of cluster-defining neuropeptides, and thus provide candidates for a regulatory relationship where each cluster may have a TF code that drives transcription of a specific neuropeptides. To address this, we performed an unbiased correlation analysis of TF and neuropeptide expression patterns across all cells in our dataset to identify potential regulatory relationships. Interestingly, we found high correlations of multiple TFs to each neuropeptide (Figure 6, Figure 6—figure supplement 1, Supplementary file 12). Our data provide a stepping-stone to the analysis of upstream TF combinatorial codes that drive neuropeptide expression.

It is generally thought that neuronal diversity is generated by TF combinatorial codes that are uniquely expressed in neuronal subtypes, with an emphasis on homeodomain TFs (Reilly et al., 2020; Sagner et al., 2021). Thus, we assayed each class of TFs to determine how many were expressed in cluster-specific (i.e. neuron-specific) patterns. We found that most clusters expressed a unique TF combination (Figure 7). This is true for all zinc-finger TFs (Figure 7A) and helix-turn-helix TFs (Figure 7B). Nearly, all clusters expressed a unique combination of homeodomain TFs (Figure 7C), with only a few clusters sharing a homeodomain code (Figure 7C'). In contrast, basic domain, unspecified domain, and high mobility group TFs were more promiscuously expressed (Figure 7D). We conclude that zinc finger, helix-turn-helix, and homeodomain TFs are good candidates for forming unique combinatorial codes that should be tested for a role in generating neuronal diversity.

We sought to link well characterized CX neurons to their transcriptomes within the T2 atlas. We chose columnar neurons due to known cell type markers and genetic access. We used four complementary approaches to identify them, described below.

The differential expression of TFs probably contributes to the morphological or connectivity diversity found within each class of CX neurons (Hulse et al., 2021; Turner-Evans et al., 2020; Wolff and Rubin, 2018; Wolff et al., 2015). We conclude that our T2 atlas can be used to link known neuron subtypes to their transcriptome (Figure 9). This is a necessary step in determining how cluster-specific gene expression regulates neuron-specific functional properties such as morphology, connectivity, and physiology.

We observed that most fast-acting neurotransmitters are uniquely expressed in neuronal populations, but we also observed a smaller group of neurons that express two or three neurotransmitters. This includes excitatory and inhibitory neurons, for example 2.9% of all neurons express genes necessary for cholinergic and GABAergic neurotransmitters (Figure 3). Co-expression of neurotransmitters has been reported in other systems (Granger et al., 2020; Granger et al., 2023; Lozovaya et al., 2018; Meye et al., 2016; Saunders et al., 2015; Shabel et al., 2014; Spitzer, 2015; Takács et al., 2018) and it may be biologically relevant in the T2-derived neurons.

Neuropeptides are profoundly important for a vast array of behaviors (Schoofs et al., 2017), and thus our mapping of neuropeptide expression to distinct neuronal subtypes may facilitate a functional analysis linking neuron identity, neuropeptide expression, and behavior.

We identified six classes of glia in the T1+T2 glial atlas that reflects previous scRNAseq datasets (Figure 1E–F; Konstantinides et al., 2018; Lago-Baldaia et al., 2023). Similar to a recent glial cell atlas (Lago-Baldaia et al., 2023), we found glial subtypes like astrocytes, ensheathing, and subperineurial glia mapped to several clusters (Figure 1E–F). It remains unclear if these clusters with the same cell type annotation represent distinct glial identities or different transcriptional states within these populations. Interestingly, we detected a small population of vkg+ surface glia specific to the whole atlas (T1+T2) and not the T2 glia atlas. The lack of differential gene expression between T1- and T2-derived glia suggests these cell types are similar despite different developmental origins. It will be for future work to understand how different progenitors produce similar glial fates.

Similar to the T1+T2 glial atlas, our T2 atlas captured the known glial subtypes found in other scRNAseq datasets (Figure 4; Konstantinides et al., 2018; Lago-Baldaia et al., 2023). The T2 glia atlas contains more clusters than the T1+T2 atlas that may be due to the nearly 4x the number of glia captured (T1+T2 atlas 3,409 nuclei; T2 atlas – 12,315 nuclei). Interestingly, we identified the astrocyte-like glia of the central brain previously described (Awasaki et al., 2008). Additionally, we captured chiasm glia that have been shown to be derived from the T2 lineage DL1 and migrate into the optic lobe (Viktorin et al., 2013). These nuclei could represent either chiasm glia that did not migrate out of the central brain or optic lobe tissue that was not completely removed during dissections. It will be interesting to investigate how these T2-derived migratory chiasm glia compared to the optic-lobe derived chiasm glia, because of their different developmental origins.

The T2 glia and neuron atlases contained clusters with disproportionate enrichment of female and male nuclei when normalized for sample input. We found the expected differential expression of yolk protein transcripts (yp1, yp2, yp3) enriched in female nuclei and the long non-coding RNAs rox1/2 and fru enriched in male nuclei (Amrein and Axel, 1997; Meller et al., 1997; Ryner et al., 1996). Interestingly, we found dsx to be enriched in both glial and neuronal female nuclei. Surprisingly, we found additional female and male-specific gene enriched in both neurons and glia including ATPsyndelta and undescribed, computationally predicted genes (CGs; Figure 5). It remains to be determined if these genes are driving sex-specific differences within glial and neuronal subtypes. These genes may reflect sex-specific differences in the adult central brain and may provide insight into how behavioral circuits are linked to sex-specific behaviors. Future work should aim to characterize and test these genes for functional roles.

Our atlas showed zinc finger TFs and homeodomain TFs as forming combinatorial codes specific for distinct cell types. This raises the possibility that these TF combinations determine neuronal functional properties, by establishing and maintaining neuronal identities. This is consistent with data from C. elegans (Hobert, 2021; Reilly et al., 2020), mammalian spinal cord (Briscoe et al., 2000; Sagner et al., 2021), Drosophila optic lobe (Holguera and Desplan, 2018; Konstantinides et al., 2018), leg motor neurons (Baek et al., 2013; Enriquez et al., 2015), and VNC (Soffers et al., 2024). Interestingly, we found more zinc finger TFs expressed in unique combinatorial codes compared to homeodomain TFs; this may be a caution to focusing too narrowly on homeodomain TFs function. We note that our results are mostly correlative and await functional analysis of the TF combinatorial codes found in our dataset. It is also possible that the current sparse TF expression patterns are due, in part, to false negative results due to low read depths. This can be resolved by increasing read depth or by performing functional assays.

In this study, we used several complementary approaches to map identified neurons to their transcriptomic UMAP cluster. First, we used bulk-seq data from specific neurons with our atlas to match neuron to cluster. Second, we used enhancer elements from neuron-specific split Gal4 lines to link enhancer-associated genes to each hemi-driver with common expression in a cluster. Third, we used known or novel TFs and other molecular markers that are expressed by a cluster as an entry point to look for other cluster specific genes. Each of these methods provides candidate neuron-cluster association, which requires experimental validation. It is not clear which method is more robust at identifying functional neuron-cluster associations. Nevertheless, we have validated a number of neuron-cluster associations using the third approach: beginning with a single gene-cluster correlation. In the future, our data can be used in two pipelines for linking neuron subtype to transcriptome. First, start with at least one validated marker for a cluster, then search all labeled clusters for candidate TFs that co-express with the validated marker, followed by validation via RNA in situ (or antibodies). Second, we can collect neurons of interest with FACS or antibody selection (Davis et al., 2020), and perform bulk RNA sequencing, followed by looking for clusters enriched for the TFs found in bulk sequencing data, and then validation via RNA in situ (or antibodies). Notably, the second pipeline requires no prior markers beyond the Gal4 or LexA line expressed in single neuronal populations. Both pipelines result in a transcriptome that can be used to identify (a) TFs for their role in neuronal specification and/or maintenance, (b) cell surface molecules that may regulate neuronal morphology and/or connectivity, or (c) functionally relevant genes encoding ion channels, neuropeptides, receptors, and signaling pathways.

We first used the split-pool method (Parse Bioscience, Seattle, WA, USA) to barcode RNAs from whole brain to generate a library which includes both T1 and T2 progeny. This library, T1+T2, was used for analysis in Figures 1 and 5. To increase the number of T2 nuclei for snRNAseq, we labeled T2 nuclei by crossing 20xUAS-FLPG5.PEST;worniu-Gal4,asense-Gal80; Act5c(FRT.CD2)Gal4 to UAS-RedStinger (RFP) or UAS-unc84-2xGFP (GFP) flies. The adult flies were aged for 1 week at 25 °C before dissection. Equal amounts of male and female central brains (excluding optic lobes) were dissected at room temperature within 1 hr. The samples were flash-frozen in liquid nitrogen and stored separately at –80 °C. Dissociation of nuclei from the frozen, dissected brains was performed according to published protocols (McLaughlin et al., 2022). RFP +or GFP +nuclei were collected by sorting dissociated nuclei with SONY-SH800 with 100 μm chip. We performed three rounds of sorting and snRNAseq. In the first round, we pooled male and female brains together to select GFP + nuclei and used particle-templated instant partitions to capture single nuclei to generate cDNA library (Fluent BioSciences, Waterton, MA). In the second round, RFP +nuclei from male and female were pooled together. In the third round, RFP +nuclei from male and female brains were collected separately. The split-pool method was then used to generate barcoded cDNA libraries from each individual nucleus from the second and third rounds. All libraries were sequenced with pair-ends reads 150 bp on Illumina Novaseq 6000 (University of Oregon’s Genomics and Cell Characterization Core Facility).