Sexually dimorphic, Fru^M-independent peaks reflect dosage compensation

Sexually dimorphic splicing of fruP1 transcripts represents a late, tissue-specific event in the sex differentiation hierarchy [11,12]. In order to test the validity of our data, we first assessed whether we could detect hallmarks of X chromosome dosage compensation in adult neurons, which is expected to be sexually dimorphic but Fru^M independent (Fig 2A). In Drosophila melanogaster, the dosage compensation complex (DCC) binds to ~700 regions on the male X to upregulate gene expression chromosome-wide (Fig 2A) [38,39]. Two X-linked lncRNAs that help to target the DCC to the male X, roX1 and roX2, were expressed only in our male samples and expressed similarly in fru⁺ and fru^- cells, as expected (Fig 2B) [40]. We also observe male-specific ATAC-seq signals at the roX1 and roX2 loci (Fig 2B). Finally, we observe expected sexually-dimorphic splicing of transformer (tra), sex lethal (sxl), and doublesex (dsx) (S2C–S2E Fig). Just as we find both functional (“male”) and non-functional (“female”) fru transcripts in the male, but only non-functional transcripts in the female (Fig 1D), we observe both non-functional (“male”) and functional (“female”) sxl and tra isoforms in female, and only non-functional isoforms in male (S2C–S2E Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Sexually dimorphic, Fru^M-independent peaks reflect dosage compensation.

A. Schematic of dosage compensation in D. mel. B. UCSC genome browser screenshot of ATAC-seq and RNA-seq signal at canonical dosage compensation genes lncRNA:roX1 and lncRNA:roX2. roX2 CHART (black bars) shows known tethering sites of lncRNA:roX2 to chromatin [39]. C. Manhattan plot of log₂(female/male) chromatin accessibility in fru⁺ and fru^- neurons across chromosomes. Points represent genomic regions. Vertical axes show the relative change in accessibility of a region from DiffBind output. Horizontal axes represent scaled chromosomal locations. Regions with no change in accessibility between samples (FDR > 0.05) are plotted in gray. Regions with male-biased accessibility (chartreuse) appear as negative fold change, and regions with female-biased accessibility (purple) appear as positive fold change. D. Summary of X versus autosome distribution of differentially accessible genomic regions (DiffBind, FDR<0.05) versus all peaks (MACS2, FDR<0.001). E. Relative distance of regions specifically accessible in female versus male fru^- neurons to lncRNA:roX2 binding sites identified by CHART [39]. Sex-biased regions which directly overlap a roX2 tethering site are highlighted in magenta. F. Signal heatmap of log2 fold change (log2FC) in ATAC-seq coverage between female fru^- and male fru^- neurons. Line plot shows mean log2FC in signal across reference-anchored regions. Signal heatmaps below are split into regions with female-biased (purple) or male-biased (chartreuse) accessibility. G. I-cisTarget analysis of male-biased regions shows enrichment of a GAGA motif matching the known CES motif for the dosage compensation machinery. The normalized enrichment score of the motif is 12.3.

https://doi.org/10.1371/journal.pgen.1009338.g002

To define differentially accessible peaks across our sex-specific datasets, we first developed methods to correct for the different numbers of X chromosomes in male and female. When the X chromosome and autosomes were analyzed together using common pipelines, most differential peaks genome-wide were found to be female-biased and from the X chromosome, presumably due to the two-fold difference in genetic material from female versus male X. As the X chromosome comprises a large proportion of the fly genome, the bias this induced in our analysis was strong: the X chromosome constituted ~20% of our total ATAC-seq reads in female flies, and only 14% in male flies. To correct for this disparity, we separated X chromosome and autosome reads in each dataset (S2A Fig). We then used DiffBind to call differential peaks on autosomes alone, and on the X chromosome alone [41,42]. This method increased our sensitivity to detect differential peaks on the autosomes between the two sexes and allowed us to predict true accessibility differences on the X chromosome (S2B Fig).

For all MACS2-called peaks genome-wide, we calculated the change in chromatin accessibility between fru^- male and female samples, and between fru⁺ male and female samples (Fig 2C and 2D and S2 Table). The genomic distribution of differential peaks was strikingly different between the fru^- samples and between the fru⁺ samples. In fru^- samples, 93 of 95 differential peaks were located on the X chromosome, suggesting that the main differences in chromatin accessibility between these samples derived from the process of dosage compensation itself. In contrast, differentially accessible regions between male and female fru⁺ neurons were evenly distributed across the genome (Fig 2C and 2D). Together, these patterns validate our ability to identify Fru^M-dependent regulatory events in ATAC-seq data and allow us to subtract signatures of dosage compensation from our analysis.

On the X chromosome, we observed ~44 regions that were strongly male-biased, similar to signals in rox1 and rox2 (Fig 2B, 2C, 2E and 2F). These peaks, which we predicted represented DCC binding sites, were common to both fru⁺ and fru^- samples. To test this, we took advantage of known male-specific DCC binding profiles mapped by analysis of rox2 chromatin binding by CHART (capture hybridization analysis of RNA targets) [39]. Indeed, male-biased peaks were relatively closer to DCC binding sites identified by roX2 CHART than were female-biased peaks, and >50% overlapped DCC binding sites (Fig 2E). Using i-cisTarget, we found sharp enrichment of the MSL recognition element (MRE) in these sequences (Fig 2F–2G) [43].

Transcriptional regulation of fruP1

We next characterized accessibility differences between the fru⁺ versus fru^- datasets. We found just 22 regions genome-wide whose accessibility was common to the two fru⁺ or two fru^- datasets, consistent with a model where fruitless neurons are heterogeneous in the absence of Fru^M protein production (S3A and S3B Fig). To ask whether these morphologically and lineally distinct neurons activate fruitless transcription through common or distinct mechanisms, we next turned our attention to the fru locus itself. fruitless has a high intron:exon ratio, which has been suggested to correlate with diversification of cis-regulatory elements and expression patterns [44,45]. The fruP1 promoter was accessible regardless of transcriptional status (Fig 1D); hierarchical control of promoter opening versus transcription has been observed in other neural systems [46]. There was an enrichment of called peaks in the locus in the two fru⁺ datasets, particularly in the first two introns downstream of fruP1 (Fig 3A). There also appeared to be pervasive enrichment of reads across the locus in the fru⁺ datasets. We reasoned that if each subpopulation of fru neurons uses a different enhancer within the fru locus to activate transcription from fruP1, averaging of these accessibility signals across a variety of fru⁺ cell types could lead to the observed pervasive opening. To quantify accessibility across the 120kb locus, we measured coverage, i.e. number of reads per base pair. We observed up to 4-fold enrichment in reads across the locus as a whole in fru⁺ samples compared to fru^- samples (Fig 3B). We re-mapped histone-mark ChIP from adult brain neurons [47] and found that this region was also enriched for H3K27Ac, a mark of active enhancers (Fig 3A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Diverse regions of the fruitless locus act as enhancers in subsets of fru⁺ neurons.

A. UCSC genome browser screenshot of the fruitless locus (dm6 assembly). Blue signal tracks represent INTACT histone ChIP data for R57C01(Nsyb)-labeled neurons in the adult head [47]. The P1 promoter of fru shows enrichment of H3K4me3, a promoter mark, while the whole gene body shows signal for H3K27ac, an enhancer mark, compared to low signal upstream of fruP1. ATAC-seq signal tracks show numerous accessible regions across the gene body. Regions whose enhancer activity is imaged in Fig 3C are highlighted. B. Aggregate reads across the fru locus in fru⁺ versus fru^- male neurons plotted in 5kb windows. Pervasive opening is observed across much of the >100kb gene, with up to four-fold more reads in the fru⁺ condition. C. Peak landscapes and conservation across four genomic fragments in the Vienna Tiles collection, as indicated. Corresponding genomic locations are highlighted in A. D. Maximum intensity 2-photon stacks of enhancer activity of the four tiles in Fig 3C within fruitless neurons in the male overlaid with the fru^P1-LexA expression pattern (top) and as GFP signal alone (bottom). Each tile drives expression in distinct fruitless neurons. Remarkably, VT043701, which has higher ATAC signal in fru^- cells, drives expression in γ Kenyon cells, which we sorted into the fru^- population.

https://doi.org/10.1371/journal.pgen.1009338.g003

To ask whether this enhanced gene-wide coverage was unique to fruitless, we used featureCounts to count reads across a gene locus and used DESeq2 to measure changes in coverage between sample conditions using a cutoff of p.adj <0.05 (S3C and S3D Fig). fruitless was near the top of the list of genes with differential coverage in this analysis, and other high-ranked genes were not like fru: They were short genes dominated by localized, robust peaks (like those found in rox1 and rox2). We also made use of this gene-scale ATAC coverage measurement to compare accessibility and RNA expression levels in the male fru⁺ condition; we found mild correlation (r² = 0.21, S3E Fig).

To ask if diverse regions of the fru locus act as enhancers in different subpopulations of fru⁺ cells, we first examined male brain expression patterns of 69 reporter lines carrying pieces of DNA from the fru locus. These alleles, from the “Vienna Tiles” collection, contain 1-3kb pieces of genomic DNA placed upstream of a minimal promoter and the coding sequence of the Gal4 transcription factor; they are recombined into a standard genomic location [48]. We viewed these expression patterns in silico, using the Brainbase database (https://implegacy.brainbase.at). Using reference anatomies of fru neurons, we found that three swathes of the fru locus had many tiles labeling fru neurons: The P1 promoter, the intron upstream of the sexually dimorphic exon, and the second intron (S3 Table) [2,3]. Remarkably, each of the 14 tiles that labeled fru neurons labeled only 1–3 fru classes, suggesting that individual regions of the fru locus act as fru enhancers in small subsets of fru neurons. To test this further, we obtained four tiles from the first and second introns and used a genetic intersection approach to identify fru neurons in which these fragments of the fru locus could act as enhancers (Fig 3C and 3D). Indeed, each region we examined drove expression in distinct fruitless subpopulations, and none of the regions we examined was able to drive gene expression across all fruitless neurons. We therefore conclude that the fruitless locus is densely packed with enhancer elements that each drive fru transcription in a subset of fru⁺ cells, i.e. that distinct gene regulatory mechanisms control fru expression across the diverse neurons that express it. This is consistent with analyses of the transcriptional control of neurotransmitter systems across ontogenetically diverse neuronal populations in C. elegans [49].

Identification of candidate Fru^M-regulated genomic elements

We sought to identify genomic regions whose accessibility robustly correlated with Fru^M status: Peaks present in male fru⁺ cells and absent from all other datasets represent genomic regions opened in the presence of Fru^M, while peaks absent in male fru⁺ cells and present in all other datasets represent regions closed in the presence of Fru^M. To identify these regions, we used DiffBind to call differentially accessible peaks between male and female samples, and between fru⁺ and fru^- samples (Figs 4A–4C and S4A–S4C and S2 Table). We observed 1037 differential peaks between male fru⁺ and male fru^-, 861 between male fru⁺ and female fru⁺, and only 58 between the two female datasets (Fig 4A). The depletion of differential peaks between the two female datasets, both lacking Fru^M, supports our hypothesis that the female fru⁺ cells, lacking Fru^M, are a heterogeneous population whose constituents are no more similar to one another than those in the fru^- population. Further, we attributed a small number of chromatin accessibility changes which depend on sex or fruitless transcriptional status alone, suggesting that the large number of accessibility changes depend on the activity of Fru^M itself (Fig 4A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Chromatin changes downstream of Fru^M are near genes involved in neuronal projection and synaptic matching.

A. Number of differentially accessible peaks between different sample comparisons at FDR<0.05. A “biased” region indicates a region which is relatively more open in a given comparison. Labels at right describe how compared conditions differ. B. Examples of peaks with increased (left) and decreased (right) accessibility specific to Fru^M neurons. C. UpSet plot showing intersection of DiffBind results at FDR <0.05 from male versus female fru⁺ neurons (861 sites) and male fru⁺ versus fru^- neurons (1037 sites). D. Distribution of fold change between the binary comparisons shown in (C). Each point is a peak. Points are colored according to their status in (C). All peaks that are differentially accessible in both comparisons vary in the same direction in both comparisons. Tail of aqua points at left represent dosage compensation signals from the X chromosome. E. Distribution of peak locations for all peaks versus Fru^M-specific peaks. F. 436 Fru^M-specific peaks are located near 303 unique genes, which are enriched for the gene categories noted. G. GO analysis using gProfiler of the 303 genes shown in Fig 4F. We include only terms with <1000 members. Genes with Fru^M-specific annotated peaks were used as input gene lists, and all genes with peaks were used as background. Green terms relate to axonogenesis and teal to adhesion.

https://doi.org/10.1371/journal.pgen.1009338.g004

To identify changes in the activity of gene regulatory elements downstream of Fru^M, we took the intersection of (1) the 1037 peaks specific to male fru⁺ versus male fru^- and (2) the 861 peaks specific to male fru⁺ versus female fru⁺ (Fig 4C and 4D). This resulted in 436 high-confidence peaks (FDR<0.05 in both comparisons) genome-wide (S4 Table). Comparing between the two male samples allows us to filter out sex-specific, Fru^M-independent elements, while comparing between male and female fru⁺ cells allows us to compare populations of cells with roughly matched identities and thus filter out peaks associated with cell type distribution. In line with our biological expectations, all peaks that satisfied both conditions were biased in the same direction in both comparisons i.e. present in the male fru⁺ dataset and absent in the other three (255 peaks), or absent in the male fru⁺ dataset and present in the other three (181 peaks) (Figs 4D and S4D). Most differential peaks were intronic or intergenic, while promoters were sharply underrepresented among differential peaks (Fig 4E). We consider these 436 genomic elements to be candidate enhancers or repressors whose activity is regulated directly or indirectly by Fru^M.

The fly genome is compact relative to common model vertebrates, and regulatory elements are often found in or near the genes they regulate [48]. We therefore hypothesize that our 436 candidate elements regulate the genes closest to them. Our Fru^M-dependent peaks were in or near 303 unique genes, which were particularly enriched for membrane proteins (122 genes), transcription factors (22) and immunoglobulin superfamily (IgSF) members (27) (Fig 4F). GO terms for cell adhesion and axonogenesis were also highly enriched compared to all the genes containing peaks across our dataset (Fig 4G). Of course, these are attractive candidate factors for determination of neuronal identity or connectivity, and IgSF proteins have been implicated previously in the fru circuit [50].

We manually inspected a subset of the 436 candidate regulatory elements; they were clearly differentially accessible by both qualitative inspection and statistical thresholds, were reproducible between replicates, and were found in genes plausibly involved in the biological processes under study. However, these differential peaks were much smaller than the peaks we observed in promoters, which were of common height across samples (Figs 4E and 1D). Because “accessibility” as measured by ATAC is essentially quantized—each region of the chromosome is tagged 0, 1, or 2 times per cell—we reasoned that peak height provides a rough measure of the proportion of cells in the analyzed population in which a locus is open, or unbound by nucleosomes. If this is the case, we would predict that each regulatory element we identified is being used by only a small portion of the fru⁺ cell population, suggesting that Fru^M might induce different gene regulatory programs in different ontogenetic classes of fru⁺ neurons.

Genomic elements specifically accessible in male fruitless neurons act as enhancers in subsets of male fruitless cells

To test the ability of candidate gene regulatory elements to drive gene expression, we identified reporter Gal4 alleles matched to their genomic loci, as in the fru locus shown in Fig 3C–3E [48,51]. Such reporter alleles are available for many of our 436 Fru^M-dependent elements, and we selected the top seven ATAC-seq peaks specifically open in Fru^M cells for which reporter alleles were available (Fig 5A, 5D and 5G). In order to visualize neurons with recent transcription from the fruP1 promoter, we constructed animals in which fruP1-LexA [52] drove expression of a red fluorophore, and reporter Gal4 constructs drove GFP expression. These animals allowed us to examine overlap between the fruitless population and neurons in which the candidate genomic region was capable of acting as a transcriptional enhancer (Fig 5B, 5C, 5E and 5F). To simplify observation of enhancer activity within the fruitless population, we also used an intersectional genetic approach to identify neurons positive for enhancer activity and current or past expression from fruP1 (S5A and S5B Fig). Enhancer activity of all seven fragments was tested with both genetic strategies, with consistent results.

In six of seven cases, we observed male-specific reporter expression in one or a few anatomic classes of fruitless neurons, validating these differential peaks as bona fide enhancers with sex-specific activity and confirming that our ATAC-seq peaks demarcated accessible chromatin from subpopulations of fru⁺ neurons (Figs 5B, 5C and 5E–5G and S5B). Reporter expression driven by the seventh construct was mutually exclusive with fru-LexA expression (84D12 tile, S5B Fig). We also note that three classes of fruitless neurons, aSP2, aSP6, and aDT6, were repeatedly labeled by these six fragments. These fruitless classes are particularly numerous, and we assume their over-representation reflects the fact that we analyzed reporters matched to the strongest differential peaks. Reporters never labeled all the aDT6, aSP2, or aSP6 neurons, suggesting that these anatomic groups contain multiple transcriptional subtypes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Regulatory elements specifically accessible in FruM neurons have sexually dimorphic and cell-type-specific activity in vivo.

A. UCSC genome browser screenshot of an intergenic region with accessibility specific to Fru^M neurons. The 80B02 fragment encompasses this peak. B. Confocal maximum intensity projections of reporter element 80B02 driving 10XUAS-IVS-myr::GFP in male (left) and female (right). 80B02 activity is sexually dimorphic in cells with somata located at the dorsal midline (arrowhead), and common to the sexes in dorsolateral cells that innervate the central complex. Representative of 3–4 images per sex. Imaging conditions were matched between sexes. C. Overlap of 80B02 signal with fru^P1-LexA-driven myr::TdTomato expression in the brains shown in (B). Insets show double-positive somata at the male dorsal midline and fibers in the lateral protocerebral complex. D. UCSC genome browser screenshot of a second intergenic region with accessibility specific to Fru^M neurons. The 91A09 fragment encompasses this peak. E. Maximum intensity projection of two-photon stack of 91A09-Gal4 driving 10XUAS-IVS-myr::GFP in male and female. 91A09 expression is shared between the sexes in lateral and subesophageal regions, but male-specific dorsally (arrowhead). Representative of 2–3 images per sex. Imaging conditions were matched between sexes. F. Overlap of 91A09 signal with fru^P1-LexA-driven myr::TdTomato expression in the brains shown in (B). Insets show double-positive somata at the male dorsal midline and fibers in the lateral protocerebral complex. G. Summary of fru⁺ cell types in which analyzed genomic fragments drive reporter expression.

https://doi.org/10.1371/journal.pgen.1009338.g005

In addition to sex-specific labeling of fru⁺ neurons, each allele also drove non-dimorphic reporter expression in some fru^- neurons (e.g. as seen in Fig 5C and 5F). Because reporter tiles were much larger than our peaks (2–3 kb versus 500bp), they could comprise multiple enhancer elements which act in fru⁺ versus fru^- neurons. To ask whether our identified peaks played a causal role in driving gene expression in fru neurons, we examined Brainbase images of overlapping tiles when possible (S5D Fig). For enhancer reporter 91A09, only tiles containing our ATAC peak labeled fru⁺ neurons, while overlapping tiles that did not cover the peak labeled fru^- neurons.

Finally, while most reporter constructs drove sex-biased expression in subpopulations of male fruitless neurons, none of the reporters drove expression across the whole fruitless population. Together with the scale of the differential peaks we observe, these data lead us to conclude that Fru^M has different direct and/or indirect genetic targets in different ontogenetic/anatomic subpopulations of fruitless neurons. Moreover, individual Fru^M-regulated enhancers are activated downstream of Fru^M only in specific subpopulations of fruitless neurons. Future enhancer-bashing experiments will be required to delineate minimal regulatory elements and to characterize whether individual elements act independently to drive the activity we observe, or in concert with nearby enhancers or repressors.

Transcriptional effectors of Fru^M are neither universal across fru⁺ cells nor dedicated to fru⁺ cells

Our ATAC-seq and enhancer activity assays suggest that Fru^M genomic targets vary across distinct anatomic populations of fruitless neurons. To ask whether there are any genes broadly regulated by Fru^M status across the fruitless population, we used DESeq2 on our RNA-seq data to call differentially expressed genes between male and female fru⁺ neurons, between male fru⁺ and fru^- neurons, and between female fru⁺ and fru^- neurons (Figs 6A and S6A and S6B and S1 Table) [53]. As shown in Fig 1, fruitless transcript quantity and splice isoform tracked the cell type and sex of the library. Aside from fruitless, we observed <300 genes differentially expressed between male fru⁺ and male fru^- cells, and most of these were also differentially expressed, in the same direction, between female fru⁺ and female fru^- cells (Figs 6A and S6A and S6B). We interpret these to be signatures of the particular populations of cells we analyzed, and that these differences in expression at the population level are independent of Fru^M. For example, eyeless is a marker of Kenyon cells, which we sorted into the fru^- population in both sexes; eyeless transcripts are enriched in both fru^- datasets, as expected. Many other genes common to fru⁺ or fru^- datasets were involved in neurotransmitter or neuropeptide production or reception suggesting the potential for distinct distributions of transmitter usage between sexually dimorphic (fru⁺) versus sex-shared (fru^-) cells (S6C Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Fru^M-regulated genes are neither distinct to nor universal across male fruitless neurons.

A. UpSet plot showing intersection of differential gene expression (DESeq2 results at FDR <0.05) from male versus female fru⁺ neurons (33 genes) and male fru⁺ versus fru^- neurons (272 genes). Only 7 genes are differential across both comparisons. B. Distribution of fold change between the binary comparisons shown in (A). Points are colored according to their status in (A). Only 5 genes have differential expression specific to Fru^M cells (i.e. fold changes in the same direction in both comparisons). C. Adult female brain expression pattern of R19B09 (green) driving expression in aDT6/median bundle neurons with nc82 counterstain (magenta). Image from Janelia FlyLight database. D. UCSC genome browser screenshot of RNA-seq and ATAC-seq signal at Lgr3, a gene that is expressed in female but not male median bundle neurons [25]. Highlighted is R19B09.3A, which Meissner et al. found to be a minimal regulatory element bound by Fru^MB.

https://doi.org/10.1371/journal.pgen.1009338.g006

We observed only 33 statistically significant differences (DESeq p. adj <0.05) in gene expression between male and female fru⁺ neurons, most of which also differed between male versus female fru^- neurons (S6A Fig). We interpret these as being sex-specific but Fru^M independent.

If Fru^M alters transcription of distinct genes across the dozens of classes of fruitless neurons, we would expect these differences to average out when the whole pool of fruitless neurons is analyzed en masse. In contrast, if Fru^M regulated the same effectors across all fruitless classes, we would expect to observe a strong transcriptional signature specific to male fru⁺ cells, containing Fru^M. We identified only seven transcripts that are uniquely active or inactive in cells containing Fru^M (Fig 6A and 6B). The implication of this analysis is that aside from fruitless itself, there are unlikely to be strong differences in gene expression that are dedicated to or shared across the fru⁺ cell population in the adult. These findings are consistent with our results that Fru^M regulates different genomic elements in different populations of fru⁺ cells and with prior research suggesting that the same neuronal specificity factors are used throughout the brain in different combinations [54–56].

While cell-type-specific gene expression changes are difficult to detect in pooled cells, genome-wide expression trends may still be apparent. To ask how Fru^M-specific regulatory elements influence transcriptional regulation genome-wide, we analyzed expression of the 303 genes near Fru^M-specific ATAC peaks in male versus female fru⁺ neurons (S6D Fig). While genes near Fru^M-open peaks were neither activated nor repressed on average, we found a small and statistically significant female bias in expression of genes near Fru^M-closed elements.

Fru^MB was shown previously to bind to a minimal regulatory element in Lgr3 and to repress Lgr3 expression in a subset of male Fru^M median bundle/aDT6 neurons (Fig 6C) [25]. Remarkably, we observe increased transcription of lgr3 in pooled female fru⁺ cells, which we presume emanates from the aDT6 subpopulation; aDT6 cells comprise 5–10% of our fru⁺ populations (Fig 6D). Fru^MB could decrease Lgr3 expression in the male by decommissioning an Lgr3 enhancer or commissioning an Lgr3 repressor. To discriminate between these, we visualized this minimal region in our ATAC-seq data and found it to be accessible in female fru⁺ cells and not in male fru⁺ cells (Fig 6D). This suggests that the regulatory element in Lgr3 identified by Meissner et al. is an enhancer that is decommissioned by Fru^M, resulting in downregulation of Lgr3 expression in male aDT6 cells.

Enrichment of Fru^M binding sites in peaks specifically closed in male fru⁺ neurons

To ask whether Fru^M acts to decommission regulatory elements more generally, we first needed to separate direct targets (i.e. those bound by Fru^M), from indirect targets (i.e. those regulated by direct Fru^M targets). To explore this using our ATAC data, we searched for Fru motifs across the 436 Fru^M-specific peaks. Fru^M has three DNA-binding domains, termed Fru^A, Fru^B, and Fru^C. The three isoforms are expressed in largely overlapping neuronal populations, with Fru^B and Fru^C expressed more broadly than Fru^A [31,57,58]. The Fru^B and Fru^C isoforms of Fru^M are each independently required in the male for courtship behavior and male aggression, while loss of Fru^A has little effect on courtship [22,31,57,58]. Loss of Fru^C causes feminization of neuronal anatomy, while loss of Fru^A and Fru^B have little anatomic effect [22,31,58]. The three isoforms have been shown by SELEX to bind distinct DNA motifs (Figs 7A and S7A) [59].

We used FIMO to search our peak datasets for Fru^A,^B, and ^C motifs. In each peak, we quantified the strongest match to each of these motifs [60]. We were surprised to find that motifs for all three DNA binding domains were more common among peaks specifically closed in the presence of Fru^M than in those specifically open in the presence of Fru^M (Figs 7A and 7B and S7A–S7C). This pattern was particularly apparent for Fru^B motifs, which were strongly enriched in peaks closed in the presence of Fru^M (Fig 7A and 7B). Together, these results suggest that, in the adult, Fru^M decommissions (or closes) the regulatory elements to which it directly binds. Moreover, these results suggest that the 181 regions inaccessible in male fru⁺ cells are those most likely to be direct Fru^M targets, while regions specifically opened in male fru⁺ cells are likely to be downstream effectors of primary targets. Analysis of additional Fru^A motifs shown in S7A–S7C Fig.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Strong Fru motifs are enriched among peaks closed in the presence of Fru^M.

A. SELEX motif for Fru [59] and cumulative frequency distribution of Fru^B motif strengths identified using FIMO across peaks open (orange), closed (magenta), and unchanged (grey) in the presence of Fru^M. All three motifs are depleted from regions specifically open in Fru^M neurons, and Fru^B motifs are enriched among peaks specifically closed in the presence of Fru^M. The p-value threshold for a well-matched motif (p <0.001) is marked as a dashed line. B. Volcano plots showing the strength of the best match to each motif across our 436 Fru^M-specific peaks. Regions opened in the presence of Fru^M have negative values. C. Fru^COM binding profiles across our Fru^M-closed and Fru^M-opened peaks. Fru^COM data is from whole L3 larvae [62]. Fru^COM signal is enriched at Fru^M-specific peaks, especially those closed in the presence of Fru^M. D. UCSC genome browser screenshot of ATAC-seq signal across a FruM-closed region covered by the enhancer reporter element R64C09. E. Genetic intersection of R64C09 enhancer activity with fruitless expression. R64C09 labels a distinct neuron population in female that is not labeled in male fruitless neurons, consistent with it acting as an enhancer that is decommissioned in the presence of Fru^M. 64C09 also drives sex-shared expression in fru⁺ olfactory sensory neurons.

https://doi.org/10.1371/journal.pgen.1009338.g007

Fru^M and Fru^COM share most of their protein coding regions and their DNA binding domains, and Fru^COM can rescue loss of Fru^M in behavioral experiments [61]. We therefore asked whether our putative Fru^M-regulated sites overlapped with Fru^COM-bound sites identified by ChIP from larvae through the modERN project (ENCODE ENCGM860XOW) [62]. We re-analyzed raw data from the modERN dataset and compared ChIP enrichments across our peaks closed in the presence of Fru^M versus open in the presence of Fru^M. We found strong enrichment of Fru^COM binding centered over peaks closed in the presence of Fru^M, and weaker Fru^COM binding enrichment at peaks opened in the presence of Fru^M (Fig 7C). This analysis suggests that despite distinct cellular contexts, Fru^M and Fru^COM may have targets in common, and supports our hypothesis that Fru decommissions regulatory elements to which it directly binds.

To test whether additional regions closed in Fru^M cells function as enhancers in female fruitless neurons, we selected one such region, an intergenic peak located near cry, vib, and CG31475 (Fig 7D). We used our intersectional genetic strategy to analyze expression of 64C09, a Janelia fragment encompassing this peak (Fig 7E). Remarkably, we found that this element drove robust, female-specific reporter expression in a subset of fru neurons, while overlapping fragments that do not encompass the peak drove expression in fru^- cells (S7E and S7F Fig). Together, these results suggest that Fru^M acts cell-type-specifically to decommission enhancer elements in male fruitless neurons.

Fru^M as a decommissioner of regulatory elements

Despite decades of work on the cellular, circuit, and behavioral functions of Fru^M, we know very little about what Fru^M does as a transcription factor to masculinize neurons. Our findings suggest that Fru^M directs genome-wide changes in chromatin accessibility and decommissions its direct targets. This mechanism is strikingly consistent with previous work: The two genes previously identified as direct Fru^M targets, robo1 and Lgr3, are both downregulated in the male [25,26]. A third gene, teiresias, was recently shown to be downregulated by Fru^M in the male and predicted to contain Fru^B binding motifs [28]. Moreover, Fru^M can recruit the cofactor Bonus, followed by the chromatin modifiers HDAC1 or HP1 [63]; HDAC1 and HP1 both convert euchromatin to heterochromatin.

This decommissioning activity raises an important question of whether Fru^M remains bound to such loci. Currently, few transcription factors show ability to bind nucleosomes (“pioneer activity”). And while currently no high-resolution binding profiles exist for Fru^M, ChIP enrichment of Fru^COM in whole third instar larvae correlates with accessible chromatin in third instar larval brains. This suggests Fru proteins bind nucleosome-displaced and not nucleosome-occupied regions. FruM may then use a “hit and run” mechanism–binding a locus, recruiting chromatin modifiers, and leaving when the region is no longer accessible. If so, ChIP-based methods may miss direct targets of Fru^M which can nevertheless be detected by looking at resultant chromatin changes with ATAC-seq.

Beyond a putative molecular mechanism for the action of Fru^M on gene expression, this finding suggests that neuronal masculinization in the insect proceeds through cell-type-specific dismantling of female or sex-shared gene expression programs. Maleness is typically conceived as an addition to female programming, as exemplified by the striking male-specific courtship routines performed in birds and insects. However, Fru^M is not required for courtship actions per se, but rather for their regulation [11,64]. Moreover, fru-expressing neurons in the female control egg-laying, and ectopic Fru^M expression in female flies dismantles egg-laying behavior [15,65]. Together, these findings are consistent with masculinization as a process of loss of female programming.

Fru^M masculinizes transcription differently in different classes of fruitless neurons

Fruitless masculinizes dozens of classes of neurons in the central brain alone, and does so by altering cell genesis, apoptosis, arbor anatomy, connectivity, and neurophysiology [2,3,6,20,21,32,33]. Consistent with the variable nature of these masculinizing mechanisms, we find that Fru^M status alters the activity of distinct gene regulatory elements across distinct classes of fruitless neurons. Indeed, we find that each relevant genomic region is likely Fru^M-regulated in just one or a few fruitless populations.

The specificity of Fru^M action across distinct fruitless neurons could result from distinct availability of cofactors with different DNA-binding domains, or from differences in the pre-existing chromatin landscape that Fru^M encounters in each cell type. Fruitless is one of 40 members of the BTB/POZ family of transcription factors in Drosophila. The BTB/POZ domain is a dimerization domain, and family members have been shown to both homo- and hetero-dimerize. The presence of different Fru^M heterodimers across neuron types that bind different composite motifs is thus a feasible mechanism. Another interesting possibility is that the distinct factors that activate fruitless transcription in different populations of fruitless neurons could themselves cooperate with Fru^M in regulating cell-type-specific effectors.

Comparison with previous genome-wide datasets

Two previous studies attempted to identify Fru^M-bound sites genome-wide through ectopic expression of tagged isoforms that enzymatically label DNA. Each of these experiments is conceptually difficult to interpret because Fru^M is studied outside of the typical cellular context and transcriptional millieu. A DamID study [57] analyzed central nervous systems in which Fru^M-Dam expression is driven by leak from an uninduced UAS-driven construct, allowing expression in any cell at any time of development; a BirA/BLRP study [66] induced Fru^M expression in the S2 cell line, which is not neural [67]. Both studies reported broad peaks (e.g. median length 10kb in the BirA study). The low resolution, likely a result of the spatial resolution of enzymatic tagging, the accretion of covalent DNA labeling over long time scale labeling, and the use of microarrays for sequence detection, preclude a crisp comparison with our nucleosome-scale peak landscapes. Gene-level intersection of putative Fru binding from the BirA/BLRP study [66] versus genes with differential peaks in our analysis is shown in S7D Fig.

We argue here that Fru^M alters expression of distinct genes across fru subpopulations, and that these transcriptional differences are masked when the whole fru populations are analyzed by RNA-seq. However, a previous study identified 772 genes enriched in male fru+ neurons in adult by TRAP [68]. We were somewhat surprised to find that we did not re-discover these genes, especially as our analysis is more sensitive than TRAP (i.e. we observe 4.8-fold enrichment of fru versus 1.5-fold enrichment by TRAP). One possibility is that differences in TRAP were dominated by fru⁺ Kenyon cells, which we filtered out in our analysis (Fig 1A). The TRAP approach also relied on comparing polysome-bound transcripts from fru⁺ neurons with input from the whole head, which would be expected to enrich neural transcripts generally.

Action of Fru^M across the life cycle

fruitless expression begins in late larvae, peaks in mid-pupal stages, and continues robustly in the adult; new populations of cells turn on fruitless expression across the life cycle, with Kenyon cell expression arising only in late pupae [13]. Fru^M exerts masculinizing effects that could arise at each of these stages. We have begun our analysis here with the adult stage, and observe gene regulatory alterations of synaptic matching molecules, transcription factors, and ion channels that are consistent with neural specificity functions required in the adult. We expect that some of these differences, especially in synaptic matching molecules, would also be observed at earlier developmental stages—synaptic matching molecules are used to guide synaptogenesis and often continue to be expressed to maintain the synapse [56,69,70]. At earlier stages, we might begin to observe Fru^M-dependent regulation of genes required for axon guidance or the induction or suppression of apoptosis. Alternatively, as differentiation is thought to proceed through loss of gene regulatory potential over time, the adult state of these cells as measured by ATAC-seq could represent a summation of all the gene regulatory alterations that occurred during earlier stages of cellular development; this is particularly likely if Fru^M decommissions regulatory elements. Finally, we observe in adults that Fru^M-dependent regulatory elements are cell-type-specific. We cannot rule out the possibility that these or other regulatory elements are used more broadly across fruitless neuron types at earlier developmental stages.

Effects of Fru^M on connectivity are unlikely to occur through convergent gene expression

The construction of neural circuits from diverse cellular parts is an extraordinarily complex problem, and the idea that circuit construction could be simplified by expression of the same factors across cells of a circuit has arisen repeatedly as a potentially simplifying mechanism [71–73]. Sexually dimorphic transcription factors are particularly compelling examples in which expression of a particular gene appears to “paint” multi-layered circuits dedicated to particular behaviors [1,29,74]. Our results here suggest that even when the same transcription factor does label lineally diverse, connected cells, this pattern is unlikely to make the process of establishing circuit connectivity any simpler. First, we suggest that lineally diverse neurons use distinct gene expression programs to activate the shared transcription factor, i.e. that expression of fruitless occurs through multiple convergent mechanisms rather than a single mechanism. The diverse paths to fruitless expression suggest that sexual behavior could be evolutionarily diversified by adding and removing cells from the fruitless-expressing network [58]. The extreme length and low exon/intron ratio of many genes that function as neural specificity factors suggest that, like the fruitless locus, they are packed with regulatory elements that modularly govern their expression across distinct neuronal populations.

Second, once Fru^M is produced, it does not homogenize the expression profiles of these diverse cells, but rather alters expression of distinct gene batteries in each population of cells. While Fruitless is thus shared across these cells, the gene regulatory elements upstream and downstream of it are not. If anything, maintaining Fruitless as a shared node across these cells imposes an additional layer of complexity and constraint on the gene regulatory events that construct the brain. We expect that common expression of transcription factors across layers of sexually dimorphic neural circuits is evolutionarily retained to allow categorization of a set of neuronal transformations as sex-related—these are “switch genes,” not “terminal selectors” [29,75]. While examples of shared expression of transcription factors or homophilic adhesion molecules across connected cells will certainly occur from time to time, we do not expect this to be a general model for the construction of circuits.

Modular control of gene expression

Our findings suggest that aside from fruitless itself, there are not genes whose expression is specific to or universal across the circuit. In this view, Fru^M does not transcriptionally unify cells of the circuit or simplify circuit specification. Rather, Fru^M flags these cells as “male,” and induces male-specific circuit function by tweaking expression of the same effector genes as are used in other combinations in other circuits. The data presented here are consistent with a model where Fru^M acts in concert with the distinct transcriptional milieu of each subpopulation of fruitless neurons to enact distinct gene regulatory programs and thus alters each class of neurons in unique ways. Gene regulatory programs therefore diverge downstream of Fru^M. We propose that this modular organization allows evolutionary diversification, as mutations to regulatory elements would be expected to alter gene expression only in individual populations of cells, rather than across the circuit.

Flies

Flies were maintained on cornmeal-molasses food, or on cornmeal food with a yeast sprinkle (‘R’ or ‘B’ recipes, Lab Express, Ann Arbor, MI) in a humidified incubator at 25C on a 12:12 light:dark cycle. Flies analyzed in all experiments were 2–7 day old adults who were housed in mixed-sex groups.

Genotypes:

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

https://doi.org/10.1371/journal.pgen.1009338.t002

Flow cytometry

Brains were dissected for up to 90 minutes in Schneider’s medium supplemented with 1% BSA and placed on ice. Optic lobes were removed during dissection. Brain dissections were interspersed such that both male and female brains were dissected throughout the dissection period. About 20 brains were obtained for each sex. After all dissections were completed, collagenase was added to a final concentration of 2mg/mL and samples were incubated at 37C for 20 minutes, without agitation.

Samples were dissociated by trituration and spun down at 300g, 4C, for 5 minutes. Collagenase solution was removed and replaced with PBS+0.1% BSA, and cells were passed through a cell strainer cap and supplemented with 50ng/mL DAPI before being subjected to flow cytometry on an a FACS Aria II. Plasticware for cell dissociation and collection was pre-treated by rinsing with PBS+1% BSA to prevent cells from sticking to bare plastic.

During flow cytometry, dead and dying cells were excluded using DAPI signal, and forward scatter and side scatter measurements were used to gate single cells. Using our dissociation methods, 50–80% of singlets appeared viable (DAPI-low), and 2–5% of viable singlets were GFP⁺. We collected 6,000–10,000 GFP⁺ cells for each fru⁺ sample and analyzed matched numbers of GFP^-/fru^- cells. For each replicate, we sorted male and female cells during the same session and performed transposition or RNA extraction in parallel. During sorting, we made two adjustments to protect the fly primary cells, which were very delicate—we disabled agitation of the sample tube, and sorted using the “large nozzle,” e.g. 100μm, i.e. using larger droplet size and lower pressure. For ATAC-seq, we sorted cells into PBS supplemented with 0.1% BSA. For RNA-seq, we sorted cells directly into Trizol-LS.

Brain dissections, staining, and imaging

Brains were dissected in external saline (108 mM NaCl, 5 mM KCl, 2 mM CaCl2, 8.2 mM MgCl2, 4 mM NaHCO3, 1 mM NaH2PO4, 5 mM trehalose, 10 mM sucrose, 5 mM HEPES pH7.5, osmolarity adjusted to 265 mOsm). For two-photon imaging, brains were then transferred fresh to 35mm imaging dishes and pinned to sylgard squares with tungsten wire. Imaging was performed on a Bruker Investigator using a 1.0 NA, 20x, water-dipping objective. Stacks were collected along the anterior-posterior axis with 1 micrometer spacing in Z and ~350nm axial pixel size.

For immunostaining and confocal imaging, brains were dissected for up to twenty minutes before being transferred to 1% paraformaldehyde in PBS, on ice. All steps were performed in cell strainer baskets (caps of FACS tubes) in 24 well plates, with the brains in the baskets lifted from well to well to change solutions. Brains were fixed overnight at 4C in 1% PFA in PBS. On day 2, brains were washed 3x10’ in PBS supplemented with 0.1% triton-x-100 on a shaker at room temperature, blocked 1 hour in PBS, 0.1% triton, 4% Normal Goat Serum, and then incubated for at least two overnights in primary antibody solution, diluted in PBS, 0.1% triton, 4% Normal Goat Serum. Primary antibody was washed 3x10’ in PBS supplemented with 0.1% triton-x-100 on a shaker at room temperature, then brains were incubated in secondary antibodies for at least two overnights, diluted in PBS, 0.1% triton, 4% Normal Goat Serum. DAPI (1 microgram/mL) was included in secondary antibody mixes. Antibodies and concentrations can be found in the resources table.

Brains were mounted in 1x PBS, 90% glycerol supplemented with propyl gallate in binder reinforcement stickers sandwiched between two coverslips. Samples were stored at 4C in the dark prior to imaging. The coverslip sandwiches were taped to slides, allowing us to perform confocal imaging on one side of the brain and then flip over the sandwich to allow a clear view of the other side of the brain. Scanning confocal stacks were collected along the anterior-posterior axis on a Leica SP8 with 1 micrometer spacing in Z and ~150nm axial pixel size.

Assay for transposase-accessible chromatin

After FAC-sorting, nuclear isolation and Tn5 transposition were performed as in [30] with modifications made for small numbers of cells as described in [36]. Transposed DNA was isolated and stored at -20C until library preparation. Transposed DNA was amplified with barcoded primers in NEBNext High Fidelity 2X PCR Master Mix (NEB) and purified with Ampure XP beads (Beckman Coulter) at a ratio of 1.6uL beads per 1uL library. Purified library was eluted in 30uL of 10mM Tris-HCl pH8, 0.1mM EDTA.

The quality of prepared libraries was verified with a Bioanalyzer 2100 using a high sensitivity DNA kit (Agilent). Libraries were quantified using KAPA qPCR assay (KAPA Biosystems), and multiplexed and sequenced on a NextSeq Illumina machine with 75 base pair, paired-end reads. Libraries were sequenced to a depth of 35–70 million reads.

RNA sequencing

Additional genomic analyses