A reanalysis of previous genomic studies shows that dprs and DIPs are regulated by Fru^M and are expressed in fru P1 neurons (Figure 1B). These independent studies examined fru P1 loss-of-function (Goldman and Arbeitman, 2007) and Fru^M gain-of-function/overexpression (Dalton et al., 2013) gene expression changes, fru P1 cell-type-specific gene expression in males and females (Newell et al., 2016), and identified direct genomic targets of Fru^M (Neville et al., 2014; Vernes, 2015). The majority of the dpr/DIP genes are identified as regulated by Fru^M or expressed in fru P1 neurons in at least three of these independent genome-wide studies (Figure 1B). Furthermore, a computational DNA-binding site analysis confirms Fru^M regulation. There is alternative splicing at the 3’ end of fru P1 transcripts that results in one DNA-binding-domain-encoding-exon being retained out of five potential exons. The predominant isoforms of Fru^M contain either the A, B, or C DNA-binding domain that each bind a different DNA sequence motif (genome-wide analysis described in Dalton et al., 2013). When we search for the presence of the three DNA sequence motifs near/in the dpr/DIP loci, Fru^M binding sites were found near/in all but two dpr/DIP loci that are examined (Figure 1—source data 1). Further evidence that dprs/DIPs have a role in fru P1 neurons comes from a live-tissue staining approach, using epitope-tagged, extracellular regions of a Dpr or DIP to examine binding (as done in Fox and Zinn, 2005; Lee et al., 2009; Özkan et al., 2013). This revealed sexual dimorphism in binding of tagged Dpr/DIP proteins to fru P1 neurons in the subesophageal ganglion brain region, with more neurons with overlap detected in males (Brovero et al., 2020). Together, these results demonstrate that every dpr and DIP gene is either regulated by Fru^M or expressed in fru P1 neurons in either males and females.

The above results led us to examine the expression patterns in the central nervous system of neurons that express both fru P1 and a dpr or DIP, using a genetic intersectional approach (Figure 1C). This approach restricts expression of a membrane-bound-GFP marker to neurons with intersecting expression of fru P1 and a dpr or DIP (fru P1 ∩ dpr/DIP). This is accomplished using a UAS-membrane-bound GFP reporter transgene that requires removal of an FRT-flanked stop cassette for expression. Removal of the stop cassette is mediated by fru P1-driven FLP recombinase (Yu et al., 2010). This system is used in combination with a collection of dpr- and DIP-Gal4 transgenic strains (Figure 1C; Venken et al., 2011; Nagarkar-Jaiswal et al., 2015a; Nagarkar-Jaiswal et al., 2015b; Tan et al., 2015; Lee et al., 2018). We primarily focused the analysis on 4- to 7-day-old adults (Figure 2 and Figure 3), which are sexually mature, and 0- to 24-hr adults to determine if the expression patterns change during adult stages (Figure 2—figure supplement 1 and Figure 3—figure supplement 1). Additionally, behavioral studies were performed on 4–7 day adults (Figure 4, Figure 5, Figure 6), so the expression and behavioral data can be co-analyzed (Figure 7). At a gross morphological level, the patterns we observe in older 4- to 7-day-old adults are also present in 0–24 hr adults, though in some cases expression in the mushroom body was not robust at the earlier 0–24 hr time point.

Based on our examination of the expression patterns in 27 intersecting genotypes, we find that 24 genotypes showed clear and consistent, membrane-bound GFP expression in the central nervous system. Of these, only two fru P1 ∩ DIP genotypes have restricted and unique patterns (fru P1 ∩ DIP-δ and fru P1 ∩ DIP-α), whereas the other genotypes have broader expression, with many in similar regions/patterns (Figures 2 and 3). fru P1 ∩ DIP-δ neuronal projections are near the anterior surface of the protocerebrum and appear to be near the γ -lobe of the mushroom body, based on visual inspection (Aso et al., 2014). The fru P1 ∩ DIP-α expression pattern is described in detail below. The 22 intersecting genotypes with broad expression, in both males and females, have consistent expression in the brain lateral protocerebral complex, including within the arch, ring, junction, and crescent (for summary see Figure 7 and Source data 1). This region has been shown to have fru P1 neurons with sexually dimorphic arbor volumes (Cachero et al., 2010; Yu et al., 2010). Furthermore, the lateral protocerebral complex has inputs from sensory neurons and is predicted to be a site of sensory integration, to direct motor output necessary for coordinating courtship activity (Yu et al., 2010). We find eight intersecting genotypes have expression in mushroom bodies in both males and females. This region has a well-established role in learning and memory, including learning in the context of courtship rejection (Jones et al., 2018; McBride et al., 1999; Montague and Baker, 2016; Zhao et al., 2018). Overall, the majority of fru P1 ∩ dpr/DIP genotypes have expression in similar regions, suggesting that some may function in a combinatorial manner within a neuron to direct patterning and/or synaptic targeting, consistent with the single-cell RNA-sequencing data presented below.

We observe sex differences in morphological features and cell body number in regions we scored (Figures 2 and 3). These regions were largely chosen because they were previously reported to have fru P1 neurons that display sex differences downstream of sex hierarchy genes transformer and fru P1 (Cachero et al., 2010; Yu et al., 2010). For example, 18 intersecting genotypes show consistent presence of signal in the mesothoracic triangle neuronal projections in males, but only two genotypes do so in females. While both males and female have expression in the DA1 and VA1v antennal lobe glomeruli in several intersecting genotypes, we also observe sexual dimorphism, with four genotypes having consistent expression in only female DA1 glomeruli (fru P1 ∩ dpr3, dpr10, dpr17, DIP-θ). In the ventral nerve cord, neurons that cross the midline are consistently observed only in males and not females. Previous work found a midline crossing phenotype that was also male-specific for a set of gustatory neurons (Mellert et al., 2010). For all regions where cell bodies are counted, the trend was that there are more cell bodies in males than females. Thus, the differences in the patterns of expression between males and females are not large, with several genotypes having quantitative differences in the numbers of cell bodies present. It is possible that there are additional quantitative differences that are not detected based on the resolution of the analyses, including quantitative differences in expression level of dpr/DIPs, or their subcellular localization, or in regions/features that are not quantified here.

fru P1 neurons have critical roles in reproductive behaviors. Studies have already determined the function of small subsets of neurons that are responsible for different aspects of behavior (reviewed in Auer and Benton, 2016). The dpr/DIP tools in hand can further address if additional combinations and/or quantitative differences in the number of fru P1 neurons are important for behavioral outcomes, given the fru P1 ∩ dpr/DIP subsets and combinations we examine are distinct from those previously studied. We used the genetic intersectional strategy to selectively activate intersecting neurons by driving expression of TrpA1, a heat-activated cation channel (Figure 1C; von Philipsborn et al., 2011). This allows for temporal control of neuronal activation by an acute increase of temperature in the courtship chambers (32°C; controls at 20°C). We note that quantitative differences in TrpA1 expression levels may account for some behavioral differences, in addition to the differences in the spatial expression patterns observed (Figures 2 and 3).

We find that neuronal activation resulted in decreases in male following and wing extending toward females for several genotypes (Figure 4 and Source data 2). We also observe that neuronal activation of fru P1 ∩ dpr (13 of 16) and fru P1 ∩ DIP (2 of 8) genotypes caused atypical courtship behavior toward a female, including double wing extension, and continuous abdominal bending, even if the female had moved away (Figures 4 and 7; for abdominal bending phenotype see Figure 4—video 1). These atypical behaviors could account for some of the decreases in following and wing extension. For example, if a male is locked into abdominal bending, this would reduce courtship following behavior. Additionally, we found that some males ejaculated on the chamber in five intersecting genotypes: dpr5 (5 of 15 animals), dpr9 (3 of 15 animals), dpr10 (3 of 15 animals), dpr12 (2 of 15 animals), and DIP-θ (4 of 15 animals). Of note, fru P1 ∩ DIP-α is the only strain that showed a decrease in courtship activities without a concomitant increase in atypical courtship behaviors. This suggests that fru P1 ∩ DIP-α neurons may normally inhibit courtship behaviors when they are activated.

We next determined if the males require females to reach an arousal threshold needed to perform typical and atypical courtship behaviors, given that several of the courtship behaviors described above occur when the male was not oriented toward the female. To address this question, we examine courtship behaviors in solitary males, using the same temporal activation strategy as above. We find that activation of the fru P1 ∩ dpr/DIP neurons is sufficient to elicit single wing extension, double wing extension, and abdominal bending in fru P1 ∩ dprs (11 of 16 genotypes) and fru P1 ∩ DIPs (3 of 8 genotypes) (Figures 5 and 7). Similarly, activating the intersecting fru P1 neuronal populations of fru P1 ∩ dpr5 (5 of 10 animals), dpr9 (1 of 10 animals), dpr10 (1 of 10 animals), dpr12 (3 of 10 animals), and DIP-θ (1of 10 animals) causes males to ejaculate without a female present. Overall, activation of these subsets of fru P1 neurons is sufficient to direct reproductive behaviors, even if a female is not present, consistent with other neuronal activation experiments (reviewed in Auer and Benton, 2016).

Given that activation of fru P1 ∩ dpr/DIP neuronal subsets resulted in changes in courtship behaviors, we next determine how silencing these neurons impacts male-female courtship, to gain further insight into their roles. To test this, we use the genetic intersectional approach with a UAS > stop > TNT transgene (Figure 1C; Stockinger et al., 2005). The intersecting genotypes express tetanus toxin light chain, which cleaves synaptobrevin, resulting in synaptic inhibition (Sweeney et al., 1995). For the control conditions, we also examine courtship behaviors of flies expressing an inactive form of TNT (TNTQ), using the genetic intersectional approach.

In addition to scoring courtship behaviors, overall motor defects were also scored (Figure 6). Given that neuronal silencing in several genotypes resulted in motor defects, in which the male fell and was unable to quickly right himself (p<0.005 for strong motor defects; 0.05>p>0.005 for minor motor defects in Figure 6), we quantify the time when the fly could not right himself as ‘motor defect’ and subtracted this from the overall courtship time for behavioral indices (Figure 6). The intersecting genotypes that consistently demonstrate motor defects additionally show decreases in following and wing extension upon silencing, likely due to some motor defect (fru P1 ∩ dpr1, dpr3, dpr4, dpr5, dpr9, dpr10, dpr11, dpr12, dpr15, and DIP-η). Additional courtship behavioral indices and latencies were quantified and those with motor defects show additional courtship phenotypes (Source data 2). However, seven intersecting genotypes had a decrease in following/wing extension indices and only minor or no motor defect (fru P1 ∩ dpr2, dpr6, dpr17, dpr18, DIP-ε, DIP-θ, and DIP-γ). One genotype, fru P1 ∩ dpr7, has an increase in following/wing extension with neuronal silencing. In the case of fru P1 ∩ dpr7, we do not detect GFP expression in the central or peripheral nervous system in adults, so the neurons underlying this phenotype remain to be determined.

The seven intersecting genotypes with no or minor motor defects, in the courtship assay, were further analyzed for locomotor defects using the Drosophila Activity Monitor (DAM) Assay (Trikinetics; Source data 2), along with genotypes fru P1 ∩ dpr7, and fru P1 ∩ dpr10, which has a strong motor defect in the courtship assay. If there is a significant difference, the intersecting genotype with neuronal silencing has increased locomotor activity in the DAM assay, suggesting that the courtship phenotypes are not due to overall loss in locomotor activity.

As above in the neuronal activating experiments, silencing fru P1 ∩ dprs (13 of 17 genotypes) is more likely to cause a courtship defect than silencing fru P1 ∩ DIPs (4 of 9 genotypes). Given the large effect size of the courtship defects compared to the smaller effect size of the motor defect, it is clear that silencing fru P1 ∩ dpr/DIP neurons in the central nervous system, for most genotypes, suppresses courtship (Figure 6). This is consistent with previous studies that have found that silencing fru P1 neurons in males leads to decreased courtship toward a female (Manoli et al., 2005; Stockinger et al., 2005). Interestingly, fru P1 ∩ DIP-α is the only strain to demonstrate motor defects, but no change in courtship behaviors upon silencing, underscoring the previous hypothesis that these neurons may normally be inhibitory for courtship.

Next, we determine if intersecting genotypes with similar expression patterns also have similar behavioral outcomes in the neuronal activating and silencing experiments described above. This allows us to determine if behavioral phenotypes are most often seen when there are large numbers of neurons activated, or when certain combinations of neurons are activated. We used a heuristic approach and generated a heatmap that groups dprs and DIPs based on similarity of the fru P1 ∩ dpr/DIP membrane-bound-GFP expression data (Figure 7A). At the top of the heatmap is a dendrogram showing the relationships in expression data, grouping those that are most similar together (from data in Source data 1). The bottom has colored dots that indicate the behavioral changes observed in the three different behavioral perturbation data sets. The scoring key for the GFP expression phenotypes is shown (Figure 7B and Source data 1). Only the 24 intersecting genotypes with quantified GFP expression data are included in the heatmap.

There is a set of eight intersecting genotypes grouped together on the right of the dendrogram that all have expression in the mushroom body and several regions within the lateral protocerebral complex, but varied expression across the other morphological features (Figure 7A; fru P1 ∩ dpr4, dpr5, dpr8, dpr9, dpr10, dpr12, dpr14, and DIP-γ). Seven have similar types of atypical courtship behaviors in the activating experiments (excluding fru P1 ∩ dpr8), in the male-female courtship assays. These seven also have similar behavioral phenotypes in the male-alone condition, indicating that the activation threshold in these lines can be achieved without a female present (Figure 5).

Furthermore, among the eight genotypes, there are four intersecting genotypes that have male ejaculates in the chamber, in both the male-female and male-alone neuronal activation assays. Among the eight intersecting genotypes, the four with the male ejaculates have the highest cell body counts in the abdominal ganglion, a region in the ventral nerve cord that has previously been shown to drive ejaculation (Source data 1; Tayler et al., 2012). However, not all intersecting genotypes with expression in the abdominal ganglion show the ejaculation phenotype, even those with a relatively high number of neuronal cell body counts in the abdominal ganglion (see fru P1 ∩ dpr1 and dpr2), suggesting that it is the combination of neurons that are activated that is critical for this phenotype. Furthermore, there is an intersecting genotype that does not have mushroom body expression, but also has the ejaculation phenotype (fru P1 ∩ DIP-θ). These results reveal how different combinations and numbers of neurons can direct a similar behavioral outcome. Overall, the results point to a critical role for interactions between the mushroom body and protocerebral complex in directing courtship behaviors, which are modified by being activated in combination with other neuronal populations. This is consistent with an idea put forth previously that posited connections between these two brain regions may integrate diverse external stimuli with internal physiological state and previous behavioral experience (Yu et al., 2010).

Twenty-two intersecting genotypes have expression in different regions of lateral protocerebral complex, but no consistent expression in the mushroom body. An examination of the behavioral data reveals no phenotypes correlated with the lateral protocerebral complex expression data. While the lateral protocerebral complex is critical for higher order processing, the data further supports the idea that interactions across different combinations of activated neurons, in each intersecting genotype, are critical for the behavioral outcomes and underscores how different patterns of neuronal activity can direct similar behavioral outcomes.

As an additional heuristic tool, we plot the correlation of the GFP expression patterns for the male and female data (Figure 7C). One goal is to gain insight into whether Dprs with the same DIP interacting partners are co-expressed in the same regions of the central nervous system. This allows us to gain understanding into the mechanisms used by these IgSF molecules to direct cell adhesion. Another goal is to determine if protein-protein interactions may occur through cis (within the same neuron) vs trans (across neurons) interactions. For example, if protein-protein interactions are in trans, then the Dpr/DIP interacting partners will be expressed in different neurons and may not have correlated expression patterns. Experimental evidence suggests trans interaction are functionally important, but there is currently no evidence that interactions in cis are important (Xu et al., 2018; Ashley et al., 2019; Courgeon and Desplan, 2019; Menon et al., 2019; Venkatasubramanian et al., 2019; Xu et al., 2019). To address these questions, the plots are annotated with DIPs (colored dots) that each Dpr interacts with on the right (based on interactome from Cosmanescu et al., 2018).

It appears that some Dprs/DIPs that bind the same partner have the most similar expression patterns. For example, in males fru P1 ∩ dpr1 and dpr2 have highly correlated expression and both Dpr1 and Dpr2 interact with DIP-η, DIP- ι, and DIP- κ. In addition, the male fru P1 ∩ DIP- η expression pattern is highly correlated with fru P1 ∩ dpr1, and dpr2. Similarly, in females, fru P1 ∩ dpr1, dpr2, and dpr3 have highly correlated expression, with Dpr1, Dpr2, and Dpr3 interacting with DIP-η and DIP- ι, and Dpr1 and Dpr2 interacting with DIP- κ. Therefore, in both males and females, fru P1 neurons may express a reporter for both a DIP and a dpr that can bind to it. In both males and females fru P1 ∩ dpr11 has correlated expression with fru P1 ∩ dpr1 and dpr2, but Dpr11 interacts with DIPs that Dpr1, and Dpr2 do not interact with. This suggests that expression of different combinations of Dprs that interact with different DIPs is a mechanism used to direct the specificity and strength of interactions. fru P1 ∩ DIP-α and DIP-δ have the most restricted expression patterns and they are not highly correlated with the expression patterns of their interacting Dpr partners, in either males or females. Consistent with the single-cell RNA-seq analysis (below), these data show that reporters for Dprs that bind the same DIP can be expressed in the same neurons, as well as in different neurons. Additionally, some Dprs with different binding partners have correlated expression, which could be a mechanism to mediate the specificity and strength of neuronal adhesion.

To determine the repertoires of dprs/DIPs expressed in individual fru P1 neurons, we perform single-cell RNA-sequencing (10X Genomics platform). The analysis was performed on male central nervous system tissues (48 hr pupal stage), from flies expressing membrane-bound GFP in fru P1 neurons. We chose this stage to gain further insight into how the dprs/DIPs direct development of fru P1 neurons, as this is the stage where Fru^M has peak expression (~1,700 Fru^M neruons, Lee et al., 2000). The data were filtered to include only the fru P1 neurons, based on detection of the membrane-bound GFP mRNA, resulting in 5621 neurons. We find that all fru P1 neurons express at least one dpr/DIP. A principal component analysis (PCA) was performed using only dpr/DIP gene expression values, and the dimensionality was reduced with the UMAP algorithm (McInnes et al., 2018 and Stuart et al., 2019). Cells with similar dpr/DIP expression will cluster closely with one another in the UMAP plot (Figure 8A). A visual inspection of the UMAP plot reveals that the patterns of dpr/DIP expression are not distinct enough to generate clusters that have large separation in the UMAP plot. We note that previous single-cell RNA-seq studies from adult central nervous system tissues found that dprs/DIPs are cell identity marker genes (Croset et al., 2018; Davie et al., 2018; Allen et al., 2020), with higher, enriched expression in subsets of neurons, rather than pan-neuronal expression, consistent with our results.

For each cluster, we next determine if a combination of dprs/DIPs are responsible for each cluster identity. We examine the average expression and the percent of cells with expression of each dpr/DIP in each cluster (Figure 8B). The UMAP clustering patterning is not due to co-expression of Dpr/DIP interacting partners (Figure 8C), based on a visual inspection, which might be expected if Dprs/DIPs predominantly interact in cis. We find that the majority of the DIPs have high average expression in one cluster, with a large percent of the cells in the cluster having expression. This is distinct from the majority of the dprs, where the average expression and the percent of cells that express the dpr is moderate and similar across many clusters. Furthermore, the distribution of the expression patterns overlaid on the UMAP plot for each dpr/DIP shows that the DIPs have more restricted expression. For example, dpr21 is broadly detected across the UMAP plot, whereas DIP-α has restricted expression in cluster 10, at the upper left-hand side of the UMAP plot (subset of expression patterns in Figure 8D; for all dprs/DIPs see Figure 8—figure supplement 1). This observation is consistent with cell-type-specific RNA-seq data from the Drosophila visual system (Tan et al., 2015; Cosmanescu et al., 2018; Davis et al., 2020), where the DIPs were among the most specifically expressed genes and dprs were more broadly expressed. This suggests that DIPs may have different functional roles, compared to Dprs, in terms of directing synaptic specificity or cell adhesion properties of the neuron.

We also generated a dendrogram by hierarchical clustering to visualize which dprs and DIPs have the most similar expression patterns across the fru P1 neurons (Figure 8—figure supplement 2). We find that some dprs and DIPs that have shared interacting partners have the most similar expression to each other. This includes the following pairs: dpr2 and dpr3; dpr6 and dpr10; dpr16 and dpr17; and DIP-ζ and DIP-ε. We find that DIP-α, and DIP-ι have the most distinct expression patterns from the rest of the interactome, which may be due to the low number of cells in which they are detected (see Upset plot described below; Source data 3). For some neurons, co-expression of dprs and DIPs with the same interacting partners may be a mechanism to generate different adhesion properties, as noted above.

Next, we examine the number of different combinations of dpr and DIP expression repertoires. A dpr/DIP gene is considered expressed if the normalized and scaled expression data value is >1, excluding those that may have stochastic expression detection due to low expression levels (5218 neurons remain). There are 458 neurons that express only one dpr or DIP. The range of neurons that express 2–8 dprs and DIPs is between 451–653 neurons (4024 neurons total); that express 9–11 dprs and DIPs is between 105–332 neurons (657 neurons total); and that express 12–15 dprs and DIPs is between 5–45 neurons (79 neurons total; Source data 3). Next, we look at the number of neurons with the same expression repertoire. This can be ascertained using an ‘Upset’ plot, which is conceptually similar to a Venn Diagram, but accommodates a large number of conditions. Here, there are the 5218 single neuron expression conditions. The majority of expression repertoires that are detected in more than one neuron are those for which the neuron only expresses one dpr or DIP (single dots on bottom of Upset Plot, 457 neurons; Source data 3). There were also 466 neurons that had shared co-expression combinations due to expression of 2–5 dprs and DIPs. The majority of fru P1 neurons had a unique repertoire of dpr/DIP expression (4295 neurons), due to expression of different combinations of 2–15 dprs and DIPs (Source data 3). In developing fru P1 neurons, this range of expression patterns of the dprs and DIPs may provide a mechanism to generate different connectivity properties for each neuron.

After examining all the fru P1 ∩ dpr/DIP patterns, and the single-cell RNA-seq data, it became apparent that fru P1 ∩ DIP-α neurons are sexually dimorphic, and that this is one of the genotypes with the fewest cells among the genotypes that were scored (Source data 1). Additionally, the fru P1 ∩ DIP-α neurons have arborization patterns that facilitate analysis of sex-differences in fine-scale processes that would be obscured in intersecting genotypes with broad expression. There are sex-differences in the superior medial protocerebrum region (SMP; Figure 9A and B, subpanels I), where females have a longer (dotted-line) and broader projection (arrowhead), as compared to males. Moreover, in the medial part of the midbrain, an ‘M’ shaped peak forms (‘M’-like) in males that is not typically observed in females (curved dotted-line, Figure 9A and B, subpanels II and III). Additionally, in the ventral lateral protocerebrum region (VLP) there are neuronal cell bodies (arrowhead, Figure 9A and B, subpanels II and III), and projections in a ‘square’ shaped pattern that are more frequently observed in females (closed dotted-line, Figure 9A and B, subpanels II and III). There is also a greater frequency of neuronal cell bodies present in the subesophageal ganglion (SEG) in females, as compared to males (arrowhead, Figure 9A and B, subpanels IV). In the abdominal ganglia (AbG) of the ventral nerve cord there is a higher density of projections in males (Figure 9A and B, subpanels V). In contrast, females have a distinct ‘forceps’ shaped pattern in the AbG region (arrowhead, Figure 9A and B, subpanels V). Taken together, it appears that the sex differences are due to differences in the number of neurons and also in the morphology of projections and arborizations (Figure 9).

We next investigated whether perturbations of the sex hierarchy genes impact fine-scale sex differences in fru P1 ∩ DIP-α neurons (Figure 9 and Source data 4). In this screen, DIP-α-Gal4 drives broad expression of each transgene (see Source data 4 for DIP-α-Gal4 expression), and the fru P1 ∩ DIP-α patterns are visualized. First, we examine the phenotypes when we overexpressed the female-isoform of the sex hierarchy gene tra (tra^F). This is expected to feminize the neurons by switching to female-specific splicing of fru P1 (Figure 1). In males, the projections in the SMP became more female-like (Figure 9C–D, subpanels I). In the medial part of midbrain, the horizontal projections in half of the male samples were either more female-like or not detected (Figure 9C–D, subpanels II). Similarly, among half of the male samples, the neuronal patterns within the AbG are either more female-like or absent (Figure 9C–D, subpanels V). We observed unexpected phenotypes in females upon overexpressing Tra^F, which suggests that quantitative differences in Tra^F have biological outcomes, as we previously suggested (Arbeitman et al., 2016). For instance, a lateral ascending neuronal projection was observed more frequently in the VLP region (Figure 9C–D, subpanels III dotted line). However, the neuronal cell bodies in the VLP, the adjacent ‘square’ shaped projection patterns (closed dotted-line, Figure 9C–D, subpanels III) and the medial horizontal projection (Figure 9C–D, subpanels III) were less frequently observed, as compared to control females.

We also examined phenotypes after Fru^M over-expression, by driving broad expression in DIP-α cells and visualizing the fru P1 ∩ DIP-α neurons. We tested three isoforms of Fru^M (UAS-Fru^MA, UAS-Fru^MB, and UAS-Fru^MC) and found that they could effectively produce Fru^M in the expected DIP-α pattern (Source data 4). Overexpression of Fru^MB and Fru^MC had large phenotypic impacts, whereas Fru^MA did not, consistent with previous functional studies of the Fru^M isoforms (Nojima et al., 2014; von Philipsborn et al., 2014). Overexpression of Fru^MB resulted in a higher frequency of the ‘M’ shaped projection pattern in males (curved dotted-line, ‘M’-like, Figure 9E–F, subpanels II), while the ‘U’ shaped SEG projection was not observed as frequently (‘U’-like, Figure 9E–F, subpanels IV). The density of the neuronal projections in the AbG was also reduced. In females, the lateral ascending neuronal projection in the VLP region is observed more frequently (Figure 9E–F,III). The overexpression of Fru^MC led to substantial reduction of fru P1 ∩ DIP-α intersecting neurons in both males and females (Figure 9G–H, subpanels III), which could be due to a loss of neurons and/or their projections. The phenotype could also be due to reduced DIP-α-Gal4 expression, given overexpression of Fru^M was previously shown to reduce expression of some IgSFs (Dalton et al., 2013).

A loss of the Fru^MC isoform, only in fru P1 neurons, had less strong phenotypic consequences (fru^FLP/ fru^ΔC; Figure 9I–J). In males, the SMP projections appeared more female-like and there was an increase in neurons with a lateral projection, due to loss of the Fru^MC isoform. Therefore, overexpressing Fru^MC isoform in the broad DIP-α-Gal4 pattern impacts fru P1 ∩ DIP-α neurons more substantially than loss of Fru^MC isoform in only fru P1 neurons. This suggests that the wildtype Fru^MC spatial expression pattern is critical for function. It is also possible that overexpression of Fru^MC interferes with other Fru isoform functions, since all Fru isoforms have a BTB domain thought to mediate dimerization. Furthermore, overexpression could titrate out co-transcriptional regulators leading to larger impacts. If we limit the overexpression of Tra^F and Fru^M to only fru P1 ∩ DIP-α neurons using an additional transgene (tub>GAL80>), we also see phenotypes that are less severe than observed when overexpression is in all DIP-α neurons (see Figure 9—figure supplement 1). Overall, quantitative and spatial changes in the expression of sex hierarchy genes alters the sexually dimorphic fru P1 ∩ DIP-α patterns. This demonstrates that sex differences in morphology are downstream of sex hierarchy regulation, through both cell autonomous and non-autonomous mechanisms.

To determine the functional roles of dprs/DIPs in fru P1 neurons, we conducted an RNAi and over-expressor screen. We use the DIP-α and DIP-δ drivers, given that they have the most restricted intersecting expression patterns, with the fewest neuronal cell bodies among the genotypes scored (Source data 1), which facilitates visually identifying altered patterns in fru P1 ∩ DIP neurons, as discussed above. Here, the DIP-Gal4 drives expression of an RNAi or over-expressor transgene of other dprs/DIPs. It should be noted that while these fru P1 ∩ DIP intersecting patterns are highly restricted, the DIP-Gal4 patterns that drive the perturbation are broader. Out of the 36 genotypes screened, only one perturbation robustly alters the fru P1 ∩ DIP expression pattern (Source data 5). Knocking down DIP-ε in all DIP-α neurons changes the fru P1 ∩ DIP-α pattern (Figure 10). Males show a significant loss of neuronal projections that have ‘U’ shaped arbors (see Figure 10C, subpanel I). Both males and females show a reduction of a set of descending neurons when compared to control flies expressing RFP RNAi (see Figure 10C, subpanel II). In addition, females show an enhancement of projections in the SMP region of the brain (see Figure 10C, subpanel III). These enhanced SMP projections have longer and more extensive projections that are not observed in males. Given that no other dpr or DIP RNAi perturbation shows these three phenotypes, this suggests that they are specific to the DIP-ε perturbation (Source data 5). No obvious morphological changes are observed in the ventral nerve cord.

We next examined the phenotypes when the DIP-ε RNAi knockdown is limited to only the fru P1 ∩ DIP-α co-expressing neurons, rather than all DIP-α neurons. We continue to use the genetic intersecting approach to visualize the neurons with GFP. To restrict expression of DIP-ε RNAi to fru P1 ∩ DIP-α neurons, we use an additional construct (tub>GAL80>), such that Gal4 is now only transcriptionally active in fru P1 ∩ DIP-α (Figure 10B). Males no longer show a significant reduction of the ‘U’ shaped projections, and neither sex shows a significant reduction of descending neurons (Figure 10). This suggests that these phenotypes are due to reduction of DIP-ε outside of fru P1 ∩ DIP-α neurons, in a non-cell-autonomous manner. Conversely, females still have the enhanced projections in the protocerebrum. This suggests that this phenotype is cell autonomous and driven by a reduction in DIP-ε expression inside the fru P1 ∩ DIP-α neurons. The results are consistent with the observation that both fru P1 ∩ DIP-α and DIP-ε are expressed in similar patterns in the SMP (Figure 3) and so it is not unexpected that expression of DIP-ε RNAi can have a functional impact in fru P1 ∩ DIP-α. Taken together, these results suggests that DIP-ε plays a critical role in establishing wildtype fru P1 neuronal patterns, in both a cell-autonomous and non-cell-autonomous manner.

In the optic lobe, antennal lobe and neuromuscular junction, genetic analyses have demonstrated that Dprs/DIPs have a role in synaptic specificity and connectivity, with Dpr-DIP interactome partner pairs mediating these critical functions (Carrillo et al., 2015; Tan et al., 2015; Barish et al., 2018; Xu et al., 2018; Ashley et al., 2019; Courgeon and Desplan, 2019; Menon et al., 2019; Venkatasubramanian et al., 2019; Xu et al., 2019). Our screen to identify morphological or synaptic changes in fru P1 ∩ DIP-α and DIP-δ neurons, using dpr/DIP RNAi or overexpression transgenes identified only one perturbation with an impact: reduction of DIP-ε by RNAi on the fru P1 ∩ DIP-α pattern, with both cell autonomous and non-autonomous roles (Figure 10). This suggests that there is sufficient redundancy in these neurons that removal or addition of one member of the Dpr/DIP interactome cannot change patterning robustly. DIP-ε interacts with a large number of Dprs, which may be one of the reasons a reduction of DIP-ε results in morphological changes. We found that some Dprs that interact with the same DIP are expressed in the same brain regions (Figure 7), and/or are detected in the same neurons, consistent with the idea of redundancy. This could also be due to other members of the IgSF that were identified by our genomic-scale screens as expressed in fru P1 neurons and/or are regulated by Fru^M, or other adhesion molecules. The enhanced set of projections in the superior medial protocerebrum region of the brain due to reduced DIP-ε is reminiscent of the synaptic targeting phenotypes seen in the optic lobe due to dpr/DIP perturbations (Carrillo et al., 2015; Tan et al., 2015; Courgeon and Desplan, 2019; Menon et al., 2019; Xu et al., 2019), which supports a role of dprs/DIPs in the development of fru P1 neuroanatomical projection patterns and/or synaptic targets.

Future studies that are performed with genetic tools that yield more penetrant phenotypes than RNAi, including CRISPR/Cas9 generated alleles, will likely reveal additional roles for Dprs/DIPs and can provide further confirmation of the RNAi results presented. Furthermore, CRISPR/Cas9 gene knock-out approaches can target multiple dprs/DIPs, allowing one to test for functional redundancy. Additional analyses to determine the subcellular localization of each Dpr/DIP will also be important to understand their roles in the nervous system, especially to determine if they are present in synaptic termini and dendrites, which would be consistent with a role of directing synaptic specificity. It is clear that higher resolution analyses of the fru P1 ∩ DIP-α pattern revealed more sexual dimorphism, so additional analysis of other genotypes at this resolution will be important, including determining developmental patterns to gain insight into mechanisms that underlie sexual dimorphism. Furthermore, our expression data reveal expression beyond development, well into adult stages. Adult roles of the dprs/DIPs may include mediating neuronal connectivity changes due to reproductive experiences.

We find that activating and silencing the subsets of neurons defined by each fru P1 ∩ dpr/DIP genotype differentially impacts male courtship behaviors, with the results highlighting that the activity of different combinations of neurons can generate a similar behavioral outcome (Figure 7). This analysis provides further insight into how similar behavioral outcomes can be generated in different ethological contexts, through the integration of information across many different neuronal subtypes. An examination of the similarities of fru P1 ∩ dpr/DIP expression patterns and behavioral outcomes suggests that interactions between the mushroom body and lateral protocerebral complex are critical to reach a certain threshold of activation for male courtship behaviors, in both the male-female and male-alone paradigm, given that the genotypes with expression in those two regions had the most consistent and robust behavioral phenotypes. Interactions between neurons in these two regions have previously been proposed to integrate disparate sensory information and behavioral experiences in order to direct courtship outcomes (Yu et al., 2010).

While there has been an impressive effort to map functions onto small subsets of neurons (Robie et al., 2017), our results suggest that it will also be important to understand the roles of different combinations of neurons to fully understand behavioral outcomes. This will facilitate understanding of how different sensory and courtship experiences impart physiological changes to direct behavior. Furthermore, these activation experiments may also reveal insights about evolution of behavior. In some Drosophila species, males perform a double wing extension during courtship (reviewed in Anholt et al., 2020). We observe double wing extension in several genotypes in the neuronal activation experiments, suggesting that changing levels of neuronal activity are a way to evolve a new behavior. Although this study focused on male reproductive behaviors, it will also be interesting to examine the role of fru P1 ∩ dpr/DIP neurons on female behavioral outcomes.

Over the last several years, genomic studies have pointed to a role of the dprs/DIPs in fru P1 neurons (Goldman and Arbeitman, 2007; Dalton et al., 2013; Neville et al., 2014; Vernes, 2015; Newell et al., 2016). Indeed, our early study showed that dpr1 had a role in courtship gating, or the timing of the steps that the male performs (Goldman and Arbeitman, 2007). Until recently, a systematic analysis of the role of fru P1 neurons that express dprs/DIPs was not possible. The use of the intersecting genetic strategy with the set of dpr/DIP Gal4 tools revealed their expression patterns and that there is an interplay between mushroom body and lateral protocerebral complex neurons, with respect to coordinating behavioral outcomes. In addition, focused experiments on the fru P1 ∩ DIP-α pattern revealed roles for the sex hierarchy and the functional redundancy of dprs/DIPs, based on the RNAi perturbation results. Future functional studies aimed at a systematic analysis of the Dpr/DIP interactome will further elucidate the role of these cell adhesion molecules in terms of specifying neuroanatomy and also as a powerful tool to gain insight into the functions of different sets of fru P1 neurons.

All flies were raised at 25°C on a 12:12 hr light-dark cycle. The flies were grown using standard cornmeal food media (33 L H₂O, 237 g Agar, 825 g dried deactivated yeast, 1560 g cornmeal, 3300 g dextrose, 52.5 g Tegosept in 270 ml 95% ethanol and 60 ml Propionic acid).

A list of Drosophila strains is provided (Key Resources Table).

Brain and ventral nerve cord (VNC) tissues were dissected from animals that were either 0–24 hr adults, or 4–7 day adults. Samples were dissected in 1x Phosphate Buffered Saline (PBS; 140 mM NaCl, 10 mM phosphate buffer, and 3 mM KCl, pH 7.4) and immediately transferred to fix (4% paraformaldehyde, 1x PBS) for 25 min at room temperature. Samples were washed for 5 min with 1x PBS, three times. The tissue was then permeabilized with TNT (0.1 M Tris-HCl [pH 7.4], 0.3 M NaCl, 0.5% Triton X-100), for 15 min, followed by two additional 5 min TNT washes. The tissue was rinsed in 1x PBS, and then Image-iT FX Signal Enhancer (Invitrogen) was applied for 25 min. Finally, the tissue was washed in TNT for two washes of 5 min each. Diluted primary antibody in TNT was applied, and samples were incubated overnight at 4°C. Next, the tissue was washed six times in TNT for 5 min each, and then secondary antibody diluted in TNT and applied. The samples were then incubated for 2 hr at room temperature or overnight at 4°C. Following this incubation, samples were washed six times in TNT for 5 min each and then mounted in Secureseal Image Spacers (Electron Microscopy Services), on glass slides with VectaShield Mounting Medium (Vector Laboratories; H-1000), and covered with #1.5 coverslips. Primary antibodies were used in the following dilutions, as indicated in the figure legends: mouse α-nc82 (1:20; Developmental Studies Hybridoma Bank, AB_2314866), rabbit α-Myc (1:6050; abcam, ab9106), rabbit α-GFP Alexa Fluor 488 (1:600; Invitrogen, A21311). Secondary antibodies were used in the following dilutions: goat α-rabbit Alexa Fluor 568 (1:500; Invitrogen, A11036), goat α-mouse Alexa Fluor 633 (1:500; Invitrogen A21052). For labeling of three MCFO markers (FLAG, V5, and HA), brains and VNCs samples were dissected and stained by following the method modified from Nern et al., 2015. The primary antibodies rabbit α-HA (1:300; Cell Signaling, 3724S), mouse α-FLAG (1:500; Sigma, F1804), and rabbit α-V5 DyLight 549 (Rockland, 600-442-378), and the secondary antibodies goat α-rabbit Alexa Fluor 633 (1:500; Invitrogen A21071) and goat α-mouse Alexa Fluor 488 (1:500; Invitrogen A11001) were used. All the antibodies were diluted in TNT.

Images were acquired on a Zeiss LSM 700 confocal microscope with a 20x objective and bidirectional scanning. The interval of each slice was set as 1.0 μm. Zeiss Zen software (Black edition, 2012) was used to make adjustments to laser power and detector gain to enhance the signal to noise ratio.

Brain and VNC confocal images of 4- to 7-day-old male or female adults were analyzed for the presence of certain morphological features and cell body numbers of select neurons. The images were scored blind, in randomized batches, by three independent people. The analysis was performed using Fiji-ImageJ 14.1, with the cell counter Janelia version 1.47 hr plugin. To determine which regions to analyze, the following criteria were used: (1) regions that had sexually dimorphic structures, (2) were present in many of the different genotypes, and/or (3) are known to be important for reproductive behaviors. A template of example images, with regions indicated, was used to ensure accurate and similar image analyses across all researchers (Source data 1). As a test to ensure accuracy of scoring across the three individuals, a round-robin scoring design was employed, with each image scored by three individuals, for a subset of 26 images, which showed high concordance. The raw cell count numbers and morphological observations were recorded in excel, compiled and then unblinded (Source data 1).

Heatmaps and correlation plots of the image analysis data were generated using Morpheus (Broad Institute; https://software.broadinstitute.org/morpheus). For features that were scored as present or absent, a value of 0 or one was calculated as the number of samples with the feature present divided by total number of samples. For the cell count data, the replicate data was averaged, and then all data was divided by the highest value for that cell count feature, so all data were between 0 and 1. The hierarchal cluster heatmap was made using the following parameters: one minus spearman rank correlation as the metric, average for linkage method, and clustering by the columns (data for each dpr/DIP). The correlation heatmaps were created using the Morpheus similarity matrix tools, using the following parameters: spearman rank as the metric, computed for the columns.

For all behavior, male flies were collected 0–6 hr post-eclosion, housed individually in small vials, and aged for 4–7 days. Canton S virgin females (white) were also collected 0–6 hr post-eclosion, and aged for 4–7 days in groups to be used as female targets for courtship with males containing the UAS > stop > TrpA1:myc transgene. Canton S virgin females were collected and kept in a similar manner to be used for courtship with male flies containing the UAS > stop > TNT/TNTQ transgenes. Flies were kept in a 25°C incubator on a 12:12 hr light:dark cycle, unless otherwise noted. Courtship chambers were placed on a temperature-controlled metal block at 25°C and videos were recorded between ZT 5–10, in a 10 mm chamber for 10 min, or until successful copulation occurred, whichever came first. For courtship using male flies harboring the UAS > stop > TrpA1:myc transgene, the male flies were reared and housed in a 19°C incubator, on a 12:12 light:dark cycle, so the Trp channel would not be activated. Courtship chambers were placed on a temperature-controlled metal block for ten minutes prior to the courtship assay, at either 20°C or 32°C.

The courtship video recordings were analyzed using The Observer XT (Noldus) (version 14.0), with an n = 14–16 for male-female behavior and n = 10 for male alone behavior. Coded behaviors included: following (a start-stop event defined as any time the male is oriented toward the female and is less than half a chamber distance away from the female), wing extension (a start-stop event defined as any time one wing is extended from the fly and is vibrating), double wing extension (a start-stop event when both wings are extended from the body and are vibrating), abdominal bending (a start-stop event when the abdomen is curled under and is not thrusting or is not in the correct position to copulate with the female), motor defect (a start-stop event when the male falls onto his back and is unable to right himself), attempted copulation (a point event when the male attempts to copulate with the female but is not successful), and successful copulation (a point event when the male is able to attach and successfully copulate with the female).

These data were graphed and analyzed using the JMP Pro 14.0.0 statistical software. A non-parametric Wilcoxon test was used to compare differences between the control and experimental temperature (for TrpA1 experiments) or between control and experimental strains (for TNT experiments), for the data for which an index is calculated. The unpaired t-test was used to determine significant differences between experimental and control conditions, with the same dpr/DIP-Gal4, to determine if the number of attempted copulations were different (test assumes equal variance).

Males were collected 0–6 hr post-eclosion and aged for 3 days in a 25°C incubator on a 12:12 hr light:dark cycle. On day three, they were individually loaded into 5 × 65 mm glass tubes (Trikinetics Inc), plugged on one end with standard cornmeal food media dipped in paraffin wax to seal. The non-food end was sealed with parafilm, with small air holes. The vials were loaded into Drosophila activity monitors (TriKinetics Inc), and placed in a 25°C incubator in 12:12 hr light:dark. Each condition was run for five days. The data from the first day of activity was not used in the analysis, as flies were recovering from CO₂ anesthesia. Activity was measured as the number of beam breaks and collected in 5-min bins. Beam crossings were summed over the 24 hr period from day 5 ZT0 (lights-on) to day 6 ZT0 per individual fly (Source data 2). These data were graphed and analyzed using the JMP Pro 14.0.0 statistical software. A non-parametric Wilcoxon test was used to compare differences between the control (TNTQ) and experimental (TNT) strains with the same dpr/DIP-Gal4.

Twenty freshly dissected male brains and ventral nerve cords, from 48 hr after puparium formation (APF) stage, were used. The flies had expression of membrane-bound GFP in fru P1 neurons and were the following genotype: w[*]; P{y[+t7.7] w[+mC]=10XUAS-IVS-mCD8::GFP}attP40/UAS-Gal4; fru P1-Gal4/+. The tissue was dissected in cold Schneider 2 Drosophila culture medium (S2 medium, Gibco) and transferred to a LoBind tube (Eppendorf), containing 200 μl of S2 medium. The tissue was centrifuged at 500 g for 5 min, and then was washed with 300 μl of EBSS (Earle’s Balanced Salt Solution), and centrifuged again at 500 g for 5 min. After centrifugation, the supernatant was replaced with 100 μl of papain for disassociation (50 units/ml, Worthington) diluted in EBSS. Brains were dissociated at 25°C in a LoBind tube for 30 min., with pipette mixing to reinforce dissociation every 3 min with a P200 tip during the first 15 min, and a P10 tip for the final 15 min. Cells were washed twice with 700 μl cold S2 medium containing 10% FBS (Gibco) and centrifuged at 700 g for 10 min to quench the papain. Cell suspensions were passed through a 30 μM pre-separation filter (Miltenyi Biotech). Cell viability and concentration were assessed by hemocytometer using Trypan blue.

Single-cell libraries were generated using Single Cell 3′ Library and Gel Bead Kit v2, Chip Kit, and the GemCode 10X Chromium instrument (10X Genomics, CA), according to the manufacturer’s protocol (Zheng et al., 2017). In brief, single cells were suspended in S2 medium with 10% FBS and the maximum volume of cells, 34 μl, was added to a single chip channel. After the generation of nanoliter-scale Gel bead-in-EMulsions (GEMs), the mRNA in GEMs underwent reverse transcription. Next, GEMs were broken, and the single-stranded cDNA was isolated, cleaned with Cleanup Mix containing DynaBeads MyOne Silane beads (Thermo Fisher Scientific). cDNA was then amplified with the following PCR machine settings: 98°C for 3 min, 9 cycles of (98°C for 15 s, 67°C for 20 s), 72°C for 1 min, held at 4°C. Subsequently, the amplified cDNA was cleaned up with SPRIslect Reagent kit (Beckman Coulter), fragmented, end-repaired, A-tailed, adaptor ligated, and cleaned with SPRIselect magnetic beads between steps. This product was PCR amplified with the following PCR machine settings: 98°C for 45 s, 12 cycles of (98°C for 20 s, 54°C for 30 s, 72°C for 20 s), 72°C for 1 min, and hold at 4°C. The library was cleaned and size-selected with SPRIselect beads, followed by Pippin size selection for a 350–450 bp library size range. Single cell libraries were sequenced on the Illumina NovaSeq with 150 bp paired-end reads on an S2 flowcell. This produced 1,870,220,065 reads. The sequence data are available through the GEO repository (GSE162098).

Raw reads were processed using the CellRanger software pipeline (v.2.1.1) ‘cellranger count’ command to align reads to the Drosophila melanogaster (BDGP6.92) STAR reference genome, customized to contain the sequence for the mCD8-GFP cDNA. The ‘force-cells’ command was used to call 25,000 single cells, based on the inflection point of the CellRanger barcode rank plot, a criterion for dividing single cells from empty GEM droplets (Source data 3). The recovered 25,000 single cells had a mean sequencing depth of 74,808 reads per cell. We detected a median of 2118 genes per cell. The obtained feature-barcode matrix was further processed and analyzed in the R package Seurat (v3.0) (Stuart et al., 2019). To filter the expression matrix for high quality cells we removed cells with >5% mitochondrial transcripts (dying cells),<200 genes (empty droplets), and/or expressing more than 6000 genes (potential doublets or triplets). This filtering produced a matrix of 24,902 high-quality cells which were computationally subset to the population of fru P1 neurons, based on mCD8-GFP expression, obtaining 5,621 cells. We next followed the Seurat ‘Guided clustering tutorial’ for default normalization and scaling steps (https://satijalab.org/seurat/v3.0/pbmc3k_tutorial.html). Expression was normalized using the ‘NormalizeData’ function where gene counts within each cell are divided by the total gene counts for that cell, multiplied by a scaling factor of 10000, and natural-log transformed (log1p). A linear transformation was applied to the normalized gene counts, to make genes more comparable to one another, using the default ‘ScaleData’ function to center the mean expression to 0 and set the variance at 1. We performed a principal component analysis (PCA) using only the data from 33 dpr/DIP genes. We used the top 20 principal components based on visual inspection of DimHeatmap outputs and the ElbowPlot. Selecting more than 20 PCs did not dramatically change our results. We then continued to follow Seurat’s standard workflow to reduce dimensionality and cluster cells using the default ‘FindNeighbors’, ‘FindClusters’, and ‘runUMAP’ functions (resolution = 1.3).

To evaluate expression combinations of the dpr/DIPs within our single cells we used an UpSet plot analysis (Conway et al., 2017). To do this, we transposed our matrix which contained normalized, log-transformed, and scaled expression data (Source data 3) for dpr/DIPs for each single cell barcode and binarized the data (any expression of a dpr or DIP >1 = 1, and >1 is considered as no expression = 0, Source data 3). All plots generated are ordered by the highest frequency of an expression combination occurring within single cells (order.by = ‘freq’). A single-cell expression hierarchical clustering dendrogram was produced using the normalized, log-transformed, and scaled expression data (Source data 3). A Pearson correlation distance measure was calculated using the factoextra (v. 1.0.7) ‘get_dist’ function and hierarchical cluster analysis was performed using the ‘hclust’ core R statistics function with the argument method = ‘average’.