Sex- and Tissue-specific Functions of Drosophila Doublesex Transcription Factor Target Genes

Abstract

Primary sex determination “switches” evolve rapidly, but Doublesex (DSX) related transcription factors (DMRTs) act downstream of these switches to control sexual development in most animal species. Drosophila dsx encodes female- and male-specific isoforms (DSX^F and DSX^M), but little is known about how dsx controls sexual development, whether DSX^F and DSX^M bind different targets, or how DSX proteins direct different outcomes in diverse tissues. We undertook genome-wide analyses to identify DSX targets using in vivo occupancy, binding site prediction, and evolutionary conservation. We find that DSX^F and DSX^M bind thousands of the same targets in multiple tissues in both sexes, yet these targets have sex- and tissue-specific functions. Interestingly, DSX targets show considerable overlap with targets identified for mouse DMRT1. DSX targets include transcription factors and signaling pathway components providing for direct and indirect regulation of sex-biased expression.

Introduction

Genetically encoded sexual dimorphism allows males and females to differ in appearance, physiology, and behavior. Differences in gamete morphology and systems that ensure they meet are often obvious, but there are subtle aspects of sex differentiation impacting organs and physiology throughout the body. Controlling the sexual development of a broad range of cell types is a challenge since sex-biased gene expression advantageous in one tissue may be detrimental in another. Sex determination systems must therefore provide organism-level, sex-specific modulation of gene expression simultaneously compatible with a range of tissue-specific requirements. Sex-specific and tissue-specific gene expression must be tightly integrated, but how this occurs is not well understood.

Primary sex determination signals vary, but doublesex and mab-3 related transcription factors (DMRTs) control sex determination and differentiation in many species (Zarkower, 2013). For example, XY humans with deletions of 3 DMRT genes exhibit sex reversal (Raymond et al., 1999). In Drosophila melanogaster, doublesex (dsx) is required for sexually dimorphic morphology, physiology, and behavior. Transformer (TRA) and Transformer 2 (TRA2) regulate female-specific alternative splicing of dsx to encode DSX^F protein. Without TRA, male-specific splicing of dsx pre-mRNA occurs, and this transcript encodes DSX^M (Burtis and Baker, 1989; Nagoshi et al., 1988). The DSX^F and DSX^M isoforms have the same DNA-binding and dimerization domains but have different C-termini (Bayrer et al., 2005; Zhang et al., 2006). Intersex (IX) binds the C-terminus of DSX^F and is required for DSX^F function (Yang et al., 2008) suggesting that the sex-specific C-termini are effector domains interacting with co-factors to modulate gene expression. DSX^F and DSX^M are required for proper sexual development, and loss of dsx function results in an intersexual phenotype. DSX^F and DSX^M have opposing effects on gene expression (Coschigano and Wensink, 1993). Thus, expressing both isoforms in the same fly results in an intersexual phenotype similar to dsx loss of function (Nagoshi and Baker, 1990).

In addition to regulation by alternative splicing, dsx is expressed highly tissue-specifically indicating that cells are on a “need to know” basis for sex (Hempel and Oliver, 2007; Lee et al., 2002; Rideout et al., 2010; Robinett et al., 2010). dsx is expressed in subsets of neurons, gut cells, gonadal somatic cells, and in adipose and hepatic tissues. These cell types derive from all primary germ layers and have diverse roles in metabolism, gametogenesis, morphology, and behavior. While the transcriptional inputs to dsx expression are not fully understood, Drosophila HOX and other patterning genes regulate dsx in at least some tissues (Foronda et al., 2012; Tanaka et al., 2011; Wang et al., 2011; Wang and Yoder, 2012; Yoder, 2012).

Although DSX has been studied for 50 years, there are still few defined DSX targets and these cannot fully explain the sexually dimorphic morphologies and behaviors regulated by dsx. The known DSX target genes were identified on a case-by-case basis (Burtis et al., 1991; Shirangi et al., 2009; Williams et al., 2008). There have been large numbers of genome-wide expression studies on the sexes but few attempts to link this expression directly to DSX (Lebo et al., 2009). One study identified genes with sex-biased and dsx-dependent expression in genital discs, but did not address whether these were directly or indirectly regulated (Chatterjee et al., 2011). DSX^F occupancy was examined genome-wide and filtered using a precise 13-mer to predict 23 direct target genes (Luo et al., 2011), but did not capture known DSX targets and is therefore unlikely to be complete. We combined an extensive DSX occupancy study on both DSX^F and DSX^M isoforms in multiple tissues with comparative genomic analyses (20 Drosophila species and mouse), expression profiling of a tissue during an acute switch in DSX isoform, and an unbiased dsx genetic interaction screen. We also determined the roles of predicted DSX targets in dsx-expressing cells. Our analyses reveal that DSX is bound to many of the same targets in males and females and in different tissues indicating that DSX action is regulated downstream of DSX binding. Further, we find a striking conservation of DSX targets in the Drosophila genus including orthologs of mouse DMRT1 targets (Murphy et al., 2010) suggesting that control of sexual dimorphism may be similar in diverse animal species.

Results

DSX occupancy

To determine where DSX binds in the D. melanogaster genome, we performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) on S2 cells expressing tagged DSX^M or DSX^F. We also performed DSX^M or DSX^F DNA adenine methyltransferase identification (DamID) on adult ovary and adult female and male fat body in transgenic flies followed by either sequencing (DamID-seq) or hybridization to microarrays (DamID-chip). We chose adult fat body and ovary since dsx plays a role in maintaining sexually dimorphic gene expression in both organs. We confirmed nuclear expression of tagged DSX and unfused Dam control by immunohistochemistry (Figure S1). Further, expression of Dam-dsx^F in males using dsx-GAL4 feminized appropriate tissues (e.g. sex combs, reproductive tract, and gonads) indicating these constructs were functional (Figure S1, over expression of Dam-dsx^M was lethal). We conducted DamID experiments using low, basal expression in the absence of a GAL4 driver to avoid known toxicity associated with Dam expression and artifacts due to DSX over expression. For all samples, we explored the continuous distribution of DSX occupancy using background-subtracted values to control for general chromatin accessibility. We identified peaks of occupancy using a stringent 1% FDR cutoff (Supplemental Experimental Procedures).

The first occupancy analysis step was at the level of peaks. We expected DSX occupancy near known DSX targets (Figure 1A,B). Indeed, the (Yolk Protein 1) and 2 loci (Yp1 and Yp2) showed strong DSX occupancy in the fat body and ovary, where these genes are expressed at high levels, along with weak occupancy in S2 cells. In contrast, the bric-a-brac 1 (bab1) locus showed strong DSX occupancy in all samples. While we observed occupancy at the previously identified Yp1/2 and bab1 DSX response elements, we also found a strongly occupied region upstream of bab1 that may represent an additional DSX-dependent enhancer.

Figure 1. DSX occupancy and binding sites.

(A–B) Scaled read density plots (background subtracted, arbitrary scale) from five replicated occupancy experiments (as labeled) for (A) the Yp1, Yp2, and (B) bab1. FlyBase gene models showing transcription start sites (bent arrows), coding exons (thick rectangles), non-coding regions (thin rectangles), introns (lines), and known DSX response elements (DSX-RE). (C) Heatmap of k-means clustering of background-subtracted, ranked occupancy scores (color scale on the left) for all D. melanogaster genes (optimal k value k = 5). (D) Box plots of gene-level occupancy scores averaged from 6 occupancy data sets in each occupancy cluster.

We next associated DSX binding sites with nearby genes and generated a DSX occupancy score. Yp1/2 and bab1 were typical examples of DSX occupancy patterns with a strong preference for occupancy in the gene body +1 kb upstream of transcription start (Figure S2). Therefore, we assigned DSX peaks to genes either using peaks occurring within this region or by using a 2 kb window centered on the annotated transcription start. Both methods limit artificial contributions of nearby upstream genes; however, the fixed-range method uncouples gene length from occupancy yet misses binding at many intronic enhancers. The gene body +1 kb definition captures intronic enhancers, but biases towards longer genes. We elected to use gene body +1 kb as this captured genes with intronic enhancers, such as bab1; however, both methods produced similar results (Table S1).

We determined gene occupancy strength using the strongest peak (peakmax) or the sum of all peaks (peaksum). We elected not to normalize for gene length as it introduced bias against long genes, such as bab1, with discrete, strong DSX binding. The two occupancy strength methods produced similar results (Spearman’s r > 0.9), and we chose the peaksum method to favor genes with multiple regions of strong occupancy.

There are many ways to examine the gene/sample relationships in DSX occupancy patterns. Supervised clustering (k-means, where k = 5) of genes’ ranked occupancy scores (Figure 1C, Table S1) revealed clusters of DSX occupancy patterns among genes that exhibit very low (cluster 4), tissue non-specific (clusters 3 and 5), and tissue- and/or technique-specific occupancy (clusters 1 and 2). In this analysis, the bona fide DSX target bab1 was in cluster 5 while the Yp1/2 genes were in cluster 3 due to modest occupancy in S2 cells. Genes ranking in the top 10% of occupancy were almost exclusively in cluster 5. Genes outside of clusters 3 and 5 had low occupancy values, although there were a few with strong occupancy in each cluster (Figure 1D).

Interestingly, DSX^F and DSX^M proteins had similar occupancy patterns (Figure 1C), suggesting the sex-specific effector domains and sex-biased chromatin environments had little impact on where DSX binds. However, there are transcriptional “hotspots” that are known to bind a host of different factors (Negre et al., 2011). To determine if the tissue non-specific occupancy and common DSX^M/F patterns were due to non-specific binding at accessible chromatin or hotspots, we correlated our results with modENCODE occupancy experiments and found that DSX^F and DSX^M occupancy pattern similarity is not explained by either chromatin accessibility or hotspots (Figure S3). Additionally, removing peaks associated with hotspots prior to analysis did not influence the overall occupancy patterns (not shown).

We conclude that the strongest DSX binding occurs in a largely sex and tissue non-specific manner. This observation focused our attention on genes following this pattern, but there are genes with tissue-specific or isoform-specific occupancy patterns that may be extremely interesting for future work. Since DSX has diverse roles in different sexes and tissues, focusing on these genes allowed us to address a previously unexplored question of how DSX integrates with tissue-specific factors rather than regulation simply by where DSX binds.

Sequence analysis of DSX binding sites

We hypothesized that the observed occupancy pattern would be due to direct DSX binding while other contacts may be indirect due to 3D structures such as looping. One simple prediction is that DSX-occupied regions should contain a DSX binding site. DSX DNA binding specificity has been defined biochemically (Erdman et al., 1996; Murphy et al., 2007; Yi and Zarkower, 1999). We were able to identify a motif statistically similar to the DSX PWM (Tomtom E-value<0.01) by de novo motif finding under occupied ChIP- seq regions (MEME-ChiP E-value<0.01), as well as enrichment of sequences matching the DSX PWM under peaks (p<0.01; Fisher’s Exact Test).

A major problem with transcription factor studies is that binding sites are common in the genome and can be bound in both functional and non-functional contexts (Fisher et al., 2012). To enhance predictions of functional binding sites, we used comparative genomics to analyze DSX binding site conservation among 20 species of Drosophila (Adams et al., 2000; Chen et al., 2014; Drosophila 12 Genomes et al., 2007; Richards et al., 2005). While conservation is not always predictive of function (Villar et al., 2014), and some non-conserved sites may be interesting species-specific targets, evolutionarily conserved sites are likely to regulate the vast array of genes showing sex-biased expression in the genus (Chen et al., 2014; Zhang et al., 2007).

The dsx sex-specific splicing pattern and encoded DNA binding domain was highly conserved across ~34 million years of Drosophila evolution (Figure S2). Therefore, we used the same biochemically-defined DSX PWM and 100 position shuffled PWMs as controls to scan the D. melanogaster genome and 19 other species in the Drosophila genus. We extracted the D. melanogaster sites using the same gene body +1 kb definition above except that we excluded coding sequence to avoid confounding DSX site and codon conservation. We assigned each identified DSX or control site in D. melanogaster a conservation index (CI) score based on the evolutionary distance at which sites could still be identified in the homologous gene using a combination of sequence and distance from the first coding exon (Table S2). Gene-level CI scores were calculated by summing site-level CIs (Table S1). We also extracted the well defined and gene length corrected PHylogenetic Analysis with Space/Time (PhastCons) sequences (Siepel et al., 2005) and calculated the mean PhastCons score for DSX sites. Briefly, a high CI or PhastCons score indicates a conserved site or the presence of a de novo site with similar sequence at the same relative position.

As expected, sites more closely matching the PWM were more likely to have deeper evolutionary conservation (Figure 2A). We observed a clear increase in the correlation between PWM score and normalized site-level CI score with a prominent “break” above the 90^th percentile of PWM scores. The PhastCons scores also showed a break but at a lower PWM score. This indicates that strong scoring sites show high evolutionary conservation, and are therefore more likely to be functional. Interestingly, both methods showed poorest conservation in moderately strong D. melanogaster sites. The meaning of this distribution is unclear but may suggest selection against sites with modest affinity for DSX that could potentially result in deleterious sex-specific regulation. At the gene-level, DSX CI scores were significantly more conserved (KS test p < 2.2e-16) across evolutionary distance than shuffled PWM CI scores (Figure 2B). For this study, we chose to focus our attention on genes with conserved DSX sites rather than D. melanogaster-specific due to either species-specific function or chance.

Figure 2. DSX occupancy and binding-site evolution.

(A) Normalized site-level CI scores plotted against PWM percentile rank score (red line) and PhastCons scores for DSX motifs (blue line). (B) Histogram of gene-level conservation index scores for DSX (red line) and the median of 100 shuffled DSX motifs (black line). (C–E) For each occupancy cluster the distribution of gene level DSX PWM scores (C), gene-level CI scores (D), and % genes in each cluster that are orthologs of mouse DMRT1 targets (E) are shown. Significant (p < 0.01) enrichment using Kruskal-Wallis (B,C) or Fisher’s exact tests (D) is indicated (*).

Comparing in vivo occupancy with sequence analysis

We combined datasets to focus on genes most likely to be functional DSX targets. As described above, genes in clusters 3 and 5 had the highest occupancy scores (Figure 1D). Interestingly, genes in clusters 3 and 5 also had significantly higher gene-level PWM scores than other clusters (Figure 2C). Genes in cluster 5 also showed significantly higher gene-level CI scores (Figure 2D), indicating that genes strongly occupied in D. melanogaster had better conservation of DSX binding sites in the Drosophila phylogeny. To determine if occupancy is also conserved, we asked if genes occupied by DSX had orthologs occupied by the mouse DSX ortholog, DMRT1 (Murphy et al., 2010). Strikingly, these orthologs were enriched in DSX occupancy (Figure 2E). This is somewhat surprising given the tremendous differences in sexual dimorphism between species. Perhaps this reflects the fact that DSX/DMRT1 orthologs control sexual dimorphism across the animal kingdom and act primarily in gonads where sexually dimorphic development is more similar in different species. Overall, our occupancy and sequence analysis are strongly concordant. We therefore focused much of our attention on genes with strong occupancy, strong PWM scores, and strong conservation.

Finally, we examined enrichment of gene ontology terms (GO terms) among occupied genes and occupancy clusters (Table S5). We found strong enrichment for many different coherent groups of genes in ontologies supporting the idea that DSX controls a wide-range of pathways and functions.

DSX-regulated expression in fat body

The above indicates that many DSX targets exhibit widespread occupancy independent of sex or tissue. However, to control sex-specific functions of distinct tissues, we expect that DSX^F and DSX^M should act on a subset of bound genes in any given tissue. To test this hypothesis, we examined expression in the adult fat body where we directly assayed DSX occupancy and where DSX-dependent expression of the Yp1/2 genes occurs. We induced an acute switch in DSX isoform (DSX^F to DSX^M or vice versa) using temperature-sensitive alleles of tra2 or a heat inducible tra system (UAS-tra^F; tub-GAL4/tub-GAL80^ts) and performed expression profiling by sequencing (RNA-seq) following temperature shifts (Figure 3, Table S1). We reasoned that switching between DSX isoform states would provide a greater net change in expression than loss of DSX function since DSX^M and DSX^F are thought to have opposing roles in target gene regulation. We measured expression genome-wide and performed k-means clustering to illustrate the overall pattern of expression in fat body (Figure 3C). 25 genes showed the strongest sex-biased expression, but only Yp1/2/3, and Fad2 showed an increase in expression correlating with higher DSX^F relative to DSX^M (Figure 3D). The Yp1/2 response was expected based on known DSX regulation, thus confirming that we manipulated known DSX outputs. The Fad2 locus encodes a female-specific sterol desaturase involved in sex pheromone signaling (Chertemps et al., 2006) that is directly regulated by DSX in oenocytes (Shirangi et al., 2009). Our data indicate that DSX also regulates Fad2 in the fat body, although we observed poor DSX occupancy raising the possibility of indirect regulation. There were a few genes, such as CG10924, CG11425, CG43051-a, showing sex-biased expression and strong occupancy, whose expression did not change upon DSX isoform switch. Perhaps these genes are regulated by DSX during development or in another context but are dsx-independent in adult fat body. Despite DSX occupancy at thousands of genes in the adult fat body, astonishingly few were transcriptionally regulated by DSX in this tissue. This suggests that many genes are poised to respond to DSX but that additional cues (temporal, spatial, nutritional, and/or hormonal) are also required. We conclude that DSX regulatory specificity depends both on where DSX is bound and the ability to coordinate with other sex-, tissue-, or condition-specific factors.

Figure 3. Tissue-specific DSX function.

(A) Sex determination in female (XX) and male (XY) flies. Functional mRNAs (black) and non-functional mRNAs (grey) are indicated. (B) dsx^M (blue) and dsx^F (red) mRNA isoform usage in control and experimental (genotypes below) in adult fat body following temperature shifts (time post shift above). Genes are listed (left). Significant differences (p<0.001, Fisher’s Exact Test) are shown (*). (C,D) Heatmap of gene expression (sample order fixed as labeled in D) and genes (rows). (D) The top cluster from (C). Mean occupancy scores (Occ) from fat body DamID-seq and DamID-array samples (color-coded).

Dose-dependent genetic interactions with dsx

If DSX binds to many genes independent of sex or tissue, then we would expect only a subset of targets would be relevant in any given sex and tissue. As a test of this hypothesis, we conducted an unbiased genetic screen to identify genomic regions that interact with dsx (Figure 4A–E, Table S3). To compromise dsx function, we used the dsx^D allele which only produces DSX^M. Consequently, XX; dsx^D/+ animals produce both DSX^F (from dsx⁺) and DSX^M (from dsx^D) resulting in an intersexual phenotype similar to dsx^- (Figure 4A–C, Figure S4). We tested 101 heterozygous deletions of the 2^nd chromosome (~33% of the genome) in the XX; dsx^D/+ background for modifications of external sexual morphology (i.e. genitalia, abdomen and sex combs) to determine if genetic interactions showed tissue-specificity (Table S3).

Figure 4. Tissue-specific genetic interactions with dsx^D.

DSX isoform in wildtype: (A) XX females, (B) XY males, and (C) XX; dsx^D/+ intersexes. (D) Feminized XX; dsx^D/+ intersexes in Df(2R)BSC109/+. Scanning electron micrographs (SEMs) of genitalia (below) showing a major female feature (vaginal plate, red) and a major male feature (genital arch, blue) in false color. Scale bar = 100μm. (E) The 2^nd chromosome with tested regions feminizing (red), masculinizing (blue), feminizing and masculinizing (purple), or having no effect (grey) on the intersexual phenotype in genitalia, abdominal pigmentation (Abd Pig), or sex combs (rows).

These experiments revealed extensive tissue-specific genetic interactions (Figure 4E). For example, in XX; Df(2R)BSC109/+; dsx^D/+ flies, all male-like genital structures were missing and female genital structures were more pronounced including a larger, fully-closed vaginal plate replete with teeth (Figure 4D); however, there were no changes in sex comb morphology, tergite number, or abdominal pigmentation. These data suggest that a gene(s) in the Df(2R)BSC109 region is required, in conjunction with dsx, for male genital disc development but not in other tested tissues. Of 101 deletions tested, 19 deficiencies defining 17 unique genomic intervals modified the dsx^D/+ external phenotype (Figure 4E). We also observed an enrichment of predicted target genes within regions exhibiting genetic interaction with dsx compared to those that did not (p < 0.01, Fisher’s exact test).

Strikingly, only a single region affected sex differentiation in more than one tissue, and this region includes intersex, which encodes a DSX^F-binding protein important for all known aspects of DSX^F function (Garrett-Engele et al., 2002). The remaining 16 interacting regions modified the dsx^D/+ phenotype in a single tissue. Assuming that regions interacting with dsx^D are randomly distributed in the genome, ~50 such “large-effect” regions exist. However, additional loci with smaller effects and loci altering internal sexual morphology, physiology or behavior are likely, suggesting that this is an underestimate. We conclude that genes interacting with dsx do so in a highly tissue-specific manner despite tissue non-specific DSX binding at many genes.

Tissue-specific effects of predicted DSX targets

Since most loci interacting with dsx do so in a highly tissue-specific manner, we wanted to determine if this was true for specific DSX target genes. We selected 80 genes and examined their roles in dsx-expressing tissues using dsx-GAL4 (Rideout et al., 2010; Robinett et al., 2010) to drive UAS-RNAi (Table S4). Genes were selected primarily due to high occupancy, PWM scores, and conservation. We also biased the set to named genes with existing alleles and selected some genes based on other criteria such as localization to a region interacting with dsx^D (Table S4). This was not a random screen but still allows us to analyze tissue- and sex-specificity in likely targets.

As in the dsx^D interaction screen, we observed striking tissue-specific phenotypes in 16 sexually dimorphic tissues (Table S4). For example, thickveins knockdown (tkv^RNAi) or dsx^RNAi both resulted in increased male-like abdominal pigmentation in females (Figure 5A), but tkv^RNAi had no effect in any other tissue in either sex. In gonads, abdominal-A (abd-A^RNAi) females exhibited disorganized ovaries that did not attach to the rudimentary genital tract but no testis phenotype, while bunched (bun^RNAi) males had a bulbous testis with no ovary phenotype (Figure 5B). Another clear tissue-specific sex transformation occurred in neuralized (neur^RNAi) females (Figure 5C), which had the male-specific large central bristle. In males, chameau (chm^RNAi) resulted in pointed sex comb teeth, as observed in females, but sex combs showed male thickness, rotation, and pigmentation (Figure 5C) indicating that multiple pathways regulate the wildtype male sex comb phenotype. In addition to the ovary phenotype, abd-A^RNAi females displayed recessed vaginal plates with reduced teeth number. Similarly, bun^RNAi males were missing the penis apparatus and most clasper teeth (Figure 5D) in addition to the testis phenotype. Interestingly, we also observed female defects in one tissue and male defects in another. For example, longitudinals lacking (lola^RNAi) females were almost entirely lacking external genitalia, while males had wide, bulbous testes (Table S4). Thus, the RNAi results demonstrate that genes bound by DSX in multiple tissues can have a striking combination of sex- and tissue-specific functions in sex differentiation.

Figure 5. Tissue-specific functions of DSX target genes.

All images left to right. (A) Abdominal pigmentation in wildtype female, male, dsx^RNAi female, and tkv^RNAi female. (B) Gross anatomy of gonads from wildtype female, abd-A^RNAi female, wildtype male, and bun^RNAi male. Terminal filaments and hubs (anti-N-Cad in green), somatic gonadal cells (anti-traffic jam, TJ in blue), and germ cells (anti-Vasa in red) are shown. Scale bar = 50μm. (C) First leg tarsal segments from wildtype female, male, neur^RNAi female, dsx^RNAi male, and chm^RNAi male. The male-specific central bristle is indicated (arrowhead). (D) SEMs of genitalia from wildtype female, abd-A^RNAi female, wildtype male, and bun^RNAi male. Scale bar = 50μm.

DOT1 complex

We would expect that multiple genes in a complex co-regulated by DSX would exhibit similar loss of function phenotypes. Many genes encoding the Disruptor Of Telomeric silencing-1 (DOT1) complex(es) are positive transcriptional regulators that methylate histone H3 at lysine 79 (Nguyen and Zhang, 2011) (Figure 6A). In Drosophila, DOT1 is encoded by grappa (gpp) (Shanower et al., 2005) and members of the complexes are encoded by lilliputian (lilli), ENL/AF9-related (ear), Alhambra (Alh), and Supressor of triplolethal (Su(Tpl)) (Figure 6B). We observed strong DSX occupancy at gpp, Su(Tpl), lilli, and Alh, but not ear (Figure 6A, Table S1). DSX^F occupancy at Alh and lilli was previously reported (Luo et al., 2011). Furthermore, the DSX binding sites for these genes were well conserved in the Drosophila phylogeny, and the mouse orthologs of Su(Tpl) and lilli are occupied by mouse DMRT1 (Murphy et al., 2010) suggesting DOT1-containing complexes are evolutionarily conserved targets of DMRTs. Given that these proteins function together in a variety of complexes, mutations should result in similar sex-transformation phenotypes.

Figure 6. Function of DOT1 in sex differentiation.

(A) DSX occupancy (see Figure 1) for the gpp, Su(Tpl), and lilli loci (see Figure 1). Positions of high scoring (≥90th percentile PWM score) DSX binding sites conserved in at least one other Drosophila species (yellow circles). (B) Model of DOT1 and associated yeast proteins (capital letters) and Drosophila orthologs (italics) loaded onto elongating RNA polymerase. (C–G) Images left to right unless indicated: (C) SEMs of wildtype, gpp^RNAi, and lilli^RNAi female genitalia with vaginal plate and teeth highlighted (dotted). Scale bar = 100μm. (D) Wildtype, gpp^RNAi, and dsx^RNAi male ejaculatory ducts (arrowheads) stained with DAPI (light blue). Scale bar = 100μm. (E) Wildtype (left top) and gpp^RNAi XY sex combs (left bottom). SEMs (last four panels) of wildtype, dsx^RNAi, gpp^RNAi, and Su(Tpl)^RNAi XY sex combs (teeth false colored). (F) Wildtype and two examples of gpp^RNAi female germline niches (see Figure 5 for antibodies). Scale bar = 10μm. G) XX; dsx^D/+ control and XX; dsx^D/gpp^X genitalia, showing the incomplete axis of rotation (dotted). Scale bar = 100μm.

When we knocked down the DOT1 complex, we observed sex- and tissue-specific phenotypes (except for Alh^RNAi). We found reduced vaginal teeth numbers in gpp^RNAi, Su(Tpl)^RNAi, and lilli^RNAi females, while males were missing lateral lobes, claspers, and penis apparati (Figure 6C, Table S4). Additionally, male-specific genital disc rotation (Adam et al., 2003) was incomplete in gpp^RNAi, Su(Tpl)^RNAi, and lilli^RNAi males (Table S4). Parovaria and spermathecae were missing from gpp^RNAi female reproductive tracts, while gpp^RNAi resulted in a narrow ejaculatory duct in males, similar to dsx^RNAi (Figure 6D). The female and male internal reproductive structures derive from different segments of the genital disc (Estrada et al., 2003), suggesting that gpp has sex- and segment-specific roles in both internal and external genital development. In males, gpp^RNAi, Su(Tpl)^RNAi, lilli^RNAi, or ear^RNAi resulted in decreased sex comb pigmentation (Figure 6E, Table S4). Additionally, gpp^RNAi and Su(Tpl)^RNAi reduced sex comb bristle number and resulted in feminized (thinner and pointed) bristles (Figure 6E). Lastly, gpp^RNAi altered the morphology of the ovarian niche where we observed collapsed terminal filaments and excessive numbers of early stage germ cells (Figure 6F), while the male niche was unaffected. In summary, the defects observed in these RNAi experiments indicate that members of the DOT1 complex(es) have similar sex- and tissue-specific functions in dsx-expressing cells.

The fact that knockdown of DOT1 complex members results in sex- and tissue-specific phenotypes is consistent with them being DSX target genes. Alternatively, DOT1’s general role in gene regulation could result in phenotypes unrelated to dsx function. To address this, we examined genetic interactions between dsx and alleles of DOT1 complex members. In the dsx^D background, heterozygosity for gpp^X reduced male genitalia structures like in gpp^RNAi and lilli^RNAi (Figure 6G, Table S4). In addition, XX; dsx^D/+ gonads had either male (hub) or female (terminal filament) germline niche structures (14% hub, n=106, Figure S4). We observed increased hub frequency in XX; dsx^D/+ gonads when heterozygous gpp (gpp^X, 62% hub, n=37) or lilli (lilli^A17–2, 39% hub, n=36). Thus, both RNAi and genetic interaction experiments suggest that DOT1 is involved in sex-specific niche development regulated by DSX.

Discussion

Identifying genes directly regulated by a transcription factor is complicated because transcription factors recognize short sequences that can arise by chance. The use of multiple genome-wide techniques helps winnow potential targets. To understand how DSX contributes to sex- and tissue-specific development, we undertook a series of genome-wide experiments and analyses to determine: where DSX is bound in different cell types, which sites are evolutionarily conserved, the relationship between site strength and occupancy, which genes respond to acute changes to DSX^F/DSX^M isoform abundance, and how many genomic regions genetically interact with dsx. We then performed RNAi knockdown of candidate targets and found striking tissue- and sex-specific transformations of sexually dimorphic structures. This rich set of targets will be useful for understanding the sex differentiation network in the powerful Drosophila system, and enrichment for orthologs of mouse DMRT1 targets among DSX targets strongly suggests that some of this network will be conserved in mammals.

The logic of DSX regulation

The sex-specific developmental programs of the gonadal mesoderm, the leg or genital imaginal discs, the fat body, and the nervous system are all likely to be highly divergent, yet all depend on DSX. How is this achieved? DSX^F and DSX^M could be recruited to different loci. While we do find a few genes with sex-specific occupancy patterns, this model is not well supported. DSX could bind different genes in different tissues. While we found examples of tissue-biased DSX occupancy, most genes are occupied regardless of tissue or sex. Finally, DSX could always bind a given target gene, but regulation would depend on the combinatorial activity of other gene-specific factors. Our work provides strongest evidence for the last model. There is also support for this combinatorial model in the literature. The bab1 locus is regulated by an enhancer that bears both DSX and homeobox protein binding sites to control sex-specific expression along the anterior/posterior axis (Williams et al., 2008). We did not examine occupancy and expression changes throughout development, but our results predict that for a given target there are both positive and negative transcriptional responses to DSX^F/M that vary among tissues of the sexes throughout development.

A large number of target genes might suggest that DSX acts as a “micro-manager” of sexual development, regulating the expression of many or most terminal sex-differentiation genes. However, our unbiased screen to identify genes interacting with dsx predicts a smaller number (~50) of “major effect” loci acting in the dsx pathway. How can we reconcile the disparity between the large numbers of potential DSX target genes with many fewer predicted to have “major” effects? DSX may delegate regulatory function to pattern formation pathways that lead to sex-specific development of organ systems. This would explain the large number of transcriptional regulators that show DSX-dependent, sex-biased expression (Chatterjee et al., 2011) predicted to be DSX targets. In addition, many genes regulated by DSX might provide subtle, but evolutionarily significant, “minor” polygenic effects on development or physiology. DSX regulation of these minor effect loci could help explain the effects of genetic background on sex-related phenotypes. These major and minor effect genes would both be strongly selected for in the course of evolution. Among predicted targets, we found enrichment for GO terms for transcription (adjusted p-value=1.85E-7) and signaling (adjusted p-value=1.44E-53), suggesting that DSX regulates gene expression of terminal differentiation factors by direct and indirect mechanisms (Table S5).

Types of DSX targets

The types of target genes predicted by our analyses illustrate how DSX is able to exhibit such powerful effects on developmental pathways. One group of predicted target genes are involved in short-range (e.g. WNT, EGF, and DPP), and long-range (e.g. Insulin and Ecdysone) signaling. Thus, DSX expression could have far-reaching effects on the development of surrounding cells and beyond. Indeed, DSX modulates short-range signaling pathways in both the genital disc (Ahmad and Baker, 2002; Gorfinkiel et al., 2003; Keisman et al., 2001) and gonad (DeFalco et al., 2008; Oliver et al., 1993; Wawersik et al., 2005). Also, YP expression requires hormonal communication in addition to DSX (Bownes et al., 1996), and titers of the steroid ecdysone are highly female-biased and germline-dependent in adults (Parisi et al., 2010), consistent with a physiological loop in which DSX as a direct transcriptional regulator of hormonal signaling pathways. This also provides a mechanism for cells to “consult” on sex-specific developmental paths and allows for reinforced and maintained sexual decisions. Such signaling mechanisms are common in sex determination. In C. elegans, the secreted factor HER-1 is a component of the primary sex determination cascade (Zarkower, 2006), and WNT and FGF signaling reinforces/specifies sexual differentiation in mammals (Eggers and Sinclair, 2012) and flies (Ahmad and Baker, 2002; DeFalco et al., 2008). These overlapping modules of gene interactions suggest significant commonalities between Drosophila and mammals.

Another major class of potential DSX targets encode transcriptional regulators, many of which have sex-specific expression patterns (Barmina et al., 2005; Chatterjee et al., 2011; Williams et al., 2008). By activating/repressing transcription factors, DSX could delegate regulation to activate pathways that proceed largely without further input by DSX. There is clear evidence for this mode of action. In the absence of dsx function, both male and female reproductive structures are found. This is opposed to the absence of all sexual structures expected for a “micro-manager” of their development. Further, in dsx^- (N. Camara, CW and MVD, in revision) or XX; dsx^D/+ gonads, either male or female stem cell niches form stochastically. We would expect these structures to be absent if dsx was required for their formation. These data are consistent with DSX being a regulator of other regulators that control female- and male-specific development.

We also found epigenetic transcriptional regulators among potential DSX target genes suggesting a function in fine-tuning and/or memory. For example, the DOT1 epigenetic machine mediates H3K79 methylation (Steger et al., 2008). While DSX control of DOT1 could act as cellular memory system and/or generally boost expression of many genes, it may also function to regulate gene expression of a few genes that contribute to sexually dimorphic phenotypes. For example, gpp^- males have a partial female-specific abdominal segment 7, reduced segment 5/6 pigmentation, and genital rotation defects (in Abd-B interactions), all of which are consistent with either segment identity change (Shanower et al., 2005) or feminization.

Feedback systems and cross-regulation also affect the output and stability of genetic pathways. Strikingly, members of the sex determination hierarchy also appear to be DSX targets. dsx and fru are bound both by DSX (this study) and FRU (Neville et al., 2014). Sex lethal (Sxl), which regulates tra, also has conserved DSX binding sites. There is precedent for feedback in sex determination as TRA is a feedback regulator of Sxl (Siera and Cline, 2008). Similarly, predicted targets of DSX such as Sex combs reduced, Abdominal-B (Abd-B), and others encode transcription factors known to regulate dsx expression (Chatterjee et al., 2011; Devi and Shyamala, 2013; Tanaka et al., 2011; Wang and Yoder, 2012). Recently, micro-RNAs have been shown to modulate sex determination (Weng et al., 2013) suggesting we are far from understanding even the basic framework for sex determination and differentiation. Even providing sexual information is more complicated than we anticipated as gpp is required for development of vaginal teeth in females and sex-combs in males. This suggests that the sexual directionality of DSX^F and DSX^M regulation of gpp depends on tissue-specific co-factors. We suggest that sex differentiation occurs via a set of context-dependent networks -- replete with rich auto-regulation, cross-regulation, and feedback -- not a hierarchy.

In summary, the wiring diagram surrounding dsx may be quite complex as DSX directly or indirectly regulates a broad set of transcription factor genes including some that regulate dsx expression. If DSX is regulated by, and a regulator of, a broad array of transcription factors widely deployed during development, then inappropriate expression of DSX could be deleterious. Indeed, ectopic expression of dsx results in widespread changes in morphology and lethality (Jursnich and Burtis, 1993), suggesting that dsx must be tightly regulated. The dsx gene is expressed in a highly tissue-specific manner consistent with the idea that only those tissues with sex-specific developmental programs express dsx suggesting that dsx expression must be tightly regulated. Understanding the logic by which DSX acts to control dimorphic developmental outcomes in different tissues in the context of multiple highly integrated networks is a key question in sex determination.

Conclusions

In over 100 years of studying sex determination and differentiation, only a few key genes have been identified. We provide a rich set of DSX targets for future studies and broadly outline the DSX mechanism of action. We conclude that DSX binding confers the possibility of sex-specific regulation where context-specific factors determine the consequences of binding. Resulting complex and context-dependent expression patterns mean that DSX^F can act as a positive regulator of a gene in one tissue, and DSX^M can act a positive regulator of the same locus in another. DSX acts by a combination of delegating control to transcription factors and by directly micromanaging terminal differentiation genes in a tightly integrated dance of regulatory inputs.

Experimental Procedures

Fly Stocks

Fly stocks were obtained from the Bloomington Drosophila Stock Center (Cook et al., 2010), the Transgenic RNAi Project (Ni et al., 2011), and from the B.S. Baker lab and other generous members of the Drosophila community. See FlyBase for gene and allele descriptions (Marygold et al., 2013) and Tables S3-S4 further information.

ChIP-seq, DamID-seq, DamID-array, RNA-seq

Transgenic DamID flies were made using sex-specific dsx cDNAs (gift of Gyunghee Lee) in pUASt-att-NDamMyc integrated into attP2 on chromosome 3L using φC31 site-directed integration (Bischof et al., 2007). S2 cells were transfected with pMT5.1-DSXM-V5-HisB or pMT5.1-DSXF-V5-HisB (Garrett-Engele et al., 2002) and pCoBlast (Invitrogen, Carlsbad, CA, USA) as the selection plasmid using Effectene (Qiagen, Valencia, CA, USA). Chromatin immunoprecipitation was performed with anti-V5 tag monoclonal antibody (Inivitrogen, Carlsbad, CA, USA) on Protein G coupled Dynabeads (Invitrogen, Carlsbad, CA, USA).

ChIP-seq libraries were constructed with the Genomic DNA sample preparation kit and were sequenced on a GA1 (Illumina, San Diego, CA, USA). RNA-seq libraries were made with the TruSeq RNA Sample Preparation Kit v2 and were sequenced on the HiSeq 2000 (Illumina, San Diego, CA, USA). Reads were mapped to FlyBase 5.46 using Tophat 1.4.1(Trapnell et al., 2009) and/or Bowtie 0.12.7 (Langmead et al., 2009). HTSeq 0.5.1p2 and DESeq 1.12.0 were used to count DNA-seq reads in 500 base pair bins (Anders and Huber, 2010; Anders et al., 2014). Transcript abundance was determined using Cufflinks 2.1.1 (Trapnell et al., 2012). k-means clustering of FPKM values was performed using the kmeans package in R (Gentleman et al., 2004). Splice junction counts were obtained using Spanki 0.4.2 (Sturgill et al., 2013). The WTD method of peak calling from the ChIP-seq analysis program SPP (version 1.11) was used to call peaks with an FDR of 0.01 (Kharchenko et al., 2008). Nimblegen performed DNA labeling and array hybridization (Roche NimbleGen, Madison WI, USA). Adjacent selected bins or probes were combined into features to produce peaks using BEDTools v2.16.2 (Quinlan and Hall, 2010). Gene level occupancy scores were calculated by summing signal in 500 bp bins under all called peak regions within the gene body + 1 kb upstream of the transcription start site. All experiments were performed in replicate. DSX occupancy data (ChIP-seq, DamID-seq, DamID-array) and RNA-seq data are available under GEO (Barrett et al., 2013) series accession GSE49480.

Sequence Analysis

We found motifs de novo using MEME-ChIP (Machanick and Bailey, 2011) and compared de novo identified motifs to the DSX PWM in TOMTOM (Gupta et al., 2007). We used the PWM for DSX in JASPAR format (Mathelier et al., 2013) to search genomes (Adams et al., 2000; Chen et al., 2014; Drosophila 12 Genomes et al., 2007; Richards et al., 2005) with the Bio.Site.search_pwm method in BioPython (Cock et al., 2009). Non-melanogaster sites were aligned to D. melanogaster using liftover chain file (Chen et al., 2014) and we summed log odds by position. Site-level conservation scores were computed by summing the substitution/site distance of each species for which a conserved sequence exists (Chen et al., 2014). Orthologs of mouse DMRT1 targets (Murphy et al., 2010) were identified using Ensembl biomart (Flicek et al., 2013). k-means clustering of FPKM values was performed using the kmeans package in R (Gentleman et al., 2004). See Supplemental Experimental Procedures for further details.

Supplementary Material

1

Figure S1: Expression of Dam-DSX fusion protein and resultant phenotypes, related to Figure 1. S2 cells carrying the pMT5.1-DSXM-V5-His construct before (A, A′) and after (B, B′) 60 hour induction. S2 cells carrying the pMT5.1-DSXF-V5-His construct before (C, C′) and after (D, D′) 60 hour induction. Scale bar = 10μm. A, B, C, and D are the V5 channel (white), and A′, B′, C′, and D′ are the DAPI channel (blue). Third instar larval fatbody from dsx-GAL4/+ (E, E′) dsx-GAL4/UAS-Dam-myc (F, F′) dsx-GAL4/UAS-Dam-myc-dsx^M (G, G′) and dsx-GAL4/UAS-Dam-myc-dsx^F (H, H′). E,F,G, and H are merged images of anti-myc (green) and DAPI (blue). E′, F′, G′ and H′ is a split of only the anti-myc signal (white). Testes (I,J) and ejaculatory ducts (K,L) were dissected and stained with DAPI. Light microscopy images of sex combs from control dsx-GAL4/TM6 (M) and dsx-GAL4/UAS-Dam-myc-dsx^F (N). Scale bar = 50μm. The Dam fusion proteins include the myc epitope incorporated at the C-terminus of the Dam coding sequence such that Dam fusion proteins can be detected with anti-myc antibodies.

Figure S2: Conservation of the DSX DNA binding domain and sex-specific splicing, related to Figures 1, 2, and 3.

(A) Diagram of the DSX^M and DSX^F proteins (above) and amino acid sequence alignment of the DSX DNA binding domain from 20 Drosophila species (below). Cysteine and histidine residues in the Zn-binding site are highlighted in tan. The evolutionary distance from D. melanogaster is indicated in substitutions/site (ss) (Chen et al., 2014). Color-coding of D. melanogaster amino acids represent mutations that do not affect DSX activity (green), partially affect activity (orange), or impair activity (red) (adapted from (Zhang et al., 2006)). (B) Bar graphs representing the percentage of dsx splicing events resulting in production of female (red) or male (blue) isoform from RNA-seq data obtained from adult females (F) or males (M) from 7 Drosophila species (Chen et al., 2014). (C) The normalized (% of maximum average occupancy) distribution of DSX occupancy values along a generic gene model using +1.5Kb upstream of transcription start (bent arrow), the gene body (rectangle), where the first 0.5Kb and last 0.5Kb are at base level, and the middle 0.5Kb is scaled from all gene models, and the -1.5Kb downstream region are shown.

Figure S3. DSX^F- and DSX^M- occupied regions are not correlated with other transcription factors, related to Figure 1 and 2. Hierarchically clustered heatmap of pairwise similarity metrics between all 255 available ChIP-chip and ChIP-seq experiments and DSX ChIP-seq (highlighted in red) as well as DSX DamID-seq/chip (highlighted in blue). Brighter colors indicate higher similarity (higher fraction of sites with p < 0.05); DSX^F and DSX^M (highlighted in red (ChIP-seq) and blue (DamID)) are more similar to each other than they are to any other assayed factor. Self-self comparisons along the diagonal are indicated in gray. Colored blocks along the left side indicate broad clusters. See methods for details. Table S6 contains the source and description of all occupancy data sets tested for correlation with DSX occupancy data.

Figure S4: XX; d s x^D/+ gonad phenotypes, related to Figure 6. Representative images of gonads dissected from XX; d s x^D/+ adults having either female-like terminal filaments (A) or male-like hubs (B). Terminal filaments and hubs are marked with anti-N-Cad (green), hubs are marked with anti-FasIII (blue) and anti-N-cad (green), and germ cells with anti-Vasa (red). Scale bar = 50μm.

2. Table S1: A summary of gene-level DSX occupancy, PWM score, CI score, orthology, and expression data (TableS1.xlsx), related to Figures 1, 2, and 3.

An .xlsx workbook file containing all gene-level occupancy scores, binary scores for occupied versus unoccupied, gene-level conservation index (CI) scores, gene-level DSX position weight matrix scores, k-means cluster ID for occupancy score clustering, binary scores for mouse DMRT1 target orthology, and RNA-seq FPKM values for each RNA-seq sample. See sheet titled “README” for more information and definition of column headings.

3. Table S2: Site-level position weight matrix scores, conservation index scores, and PhastCons scores, all position weight matrices (TableS2.xlsx), related to Figure 1.

An .xlsx workbook file containing DSX position weight matrix (PWM) scores for all 566,628 DSX motif-related sequences (sheet “PWM_scores”), conservation index scores, and PhastCons scores for all 173,775 sites with a positive PWM scores located within a D. melanogaster gene body plus 1 kb upstream excepting those in coding sequence (sheet “CI_scores”), conservation index (CI) scores for the 17,380 sites in the top 10% of all sites with any relationship to the DSX PWM that occur within a gene excluding coding sequence or 1Kb upstream of a gene (sheet “top_10_percent_PWM_CI_scores”), and the DSX position weight matrix along with all shuffled position weight matricies (“PWMs”). See sheet titled “README” for more information and definition of column headings.

4. Table S3: 2^nd chromosome deficiencies tested for genetic interaction with dsx^D (TableS3.xlsx), related to Figure 4.

A .xlsx workbook file containing results from the unbiased genetic interaction tests between dsx^D and 101 deficiencies of the 2^nd chromosome, 17 genomic intervals interacting with dsx^D, and the tested genomic intervals that did not interact with dsx^D. For each deficiency tested for genetic interaction with dsx^D, a description of the genetic interaction phenotype is provided (“Phenotype Description”) along with the FlyBase deficiency name and aberration ID (“FlyBase ID”). The known or estimated cytological location (“Deleted Segment”) as well as chromosome base position (FlyBase release 5) for the left and right breakpoints are also provided.

5. Table S4: A summary of RNAi data for putative DSX target genes (TableS4.xlsx), related to Figures 5 and 6.

For each gene tested by RNAi knockdown, the FlyBase gene name and ID (“FlyBase Identifier”) are provided along with a text description summarizing phenotypic findings after RNAi knockdown (“Phenotype Descriptions”). There is also a text summary of DSX occupancy, occupancy clustering, gene-level PWM score, gene-level conservation index, and other criteria used to select the gene for RNAi test (“Criteria for Selection”).

6. Table S5: Gene ontology term enrichment in occupied genes (TableS5.xlsx), related to Figure 1 and 2.

An .xlsx workbook file containing enriched gene ontology terms in the 3,717 occupied genes identified in this work. The ontology domains molecular function and biological process were examined, with output from each domain occupying a separate sheet in this file. See sheet titled “README” for more information and definition of column headings.

7. Table S6: Occupancy data sets used for comparison to DSX occupancy data, related to Figure 1 and Figure S3.

An .xlsx workbook file containing a source and description for all occupancy data sets tested for correlation with DSX occupancy data. See sheet titled “README” for more information and definition of columns headings.