Maintenance of Drosophila germline stem cell sexual identity in oogenesis and tumorigenesis

Homeostasis of adult tissues depends on a stable population of stem cells that have the capacity to give rise to both self-renewing and differentiating daughter cells. Many cancers exhibit stem cell-like phenotypes, suggesting a direct link between aberrant stem cell behavior and tumor formation (Friedmann-Morvinski and Verma, 2014). The mechanisms involved, however, remain largely unknown.

Drosophila oogenesis is a powerful model system for the study of adult stem cells and their connection to cancer stem cells (Hudson and Cooley, 2014; Spradling et al., 2011; Tipping and Perrimon, 2014). Adult ovaries are composed of individual strands of developing egg chambers called ovarioles. Each ovariole maintains a steady population of two to three germline stem cells (GSCs) located at the tip in a structure called the germarium. The GSCs divide asymmetrically to give rise to one daughter that remains a stem cell and another daughter, called a cystoblast (CB), that commits to differentiation. The CB then undergoes four mitotic divisions to form an interconnected 16-cell cyst. Only one of these 16 cells differentiates into an oocyte, while the remaining 15 cells develop as polyploid nurse cells. An egg chamber is formed as the somatic follicle cells surround the 16-cell cyst and bud off from the germarium (Fig. 1A).

The decision between stem cell maintenance and differentiation is controlled by both intrinsic and extrinsic mechanisms. The GSCs, which are located at the anterior end of the germarium, receive differentiation-inhibiting signals from their somatic neighbors. For example, somatic cells at the tip of the germarium activate BMP signaling in GSCs to directly repress transcription of the differentiation-promoting gene bag of marbles (bam) (Chen and McKearin, 2003a,,b,, 2005; Song et al., 2004). Cell-intrinsic programs, such as those carried out by the translational repressor Nanos, are also required for stem cell maintenance (Forbes and Lehmann, 1998; Gilboa and Lehmann, 2004; Harris et al., 2011; Wang and Lin, 2004). Following an oriented cell division, the GSC daughter that is specified as a CB displays an increase in Bam protein levels and a decrease in self-renewal factors, including Nanos protein (Chau et al., 2009,, 2012; Li et al., 2009).

We recently found that the switch to the CB fate requires the female-specific RNA-binding protein Sex lethal (Sxl) (Chau et al., 2009,, 2012). Germ cells without Sxl protein form tumors that comprise a few bona fide GSCs located at the tip of the germarium and cells that coexpress both Bam protein and a set of GSC markers, including Nanos protein. Studies, showing that nanos is an Sxl target gene and that Nanos downregulation in CB cells is controlled at the level of translation, indicate that Sxl enables the GSC-to-CB switch by directly inhibiting nanos translation (Chau et al., 2009,, 2012; Li et al., 2009). Although nanos is clearly necessary for tumor growth, both loss- and gain-of-function studies indicate that the failure to regulate nanos is not a trigger for tumorigenesis (Chau et al., 2012; Harris et al., 2011; Li et al., 2009). Consequently, the mechanism driving tumor formation in the absence of Sxl protein remains unknown.

Previous studies have shown that Sxl also functions in somatic cells, where its activation in early embryogenesis is the central female determining event (Salz, 2011; Salz and Erickson, 2010). Sxl, however, is not essential for establishing sexual identity in the female germline. In the absence of Sxl, germ cells develop normally and exhibit the appropriate female-specific behaviors and expression patterns through the end of the larval period (Casper and van Doren, 2009; Chau et al., 2009; Steinmann-Zwicky, 1994). Germ cells, unlike somatic cells, acquire their female identity by a non-cell-autonomous mechanism, as demonstrated by studies showing that the gene expression program and behavior of embryonic XX germ cells is masculinized if placed in a male somatic environment (Casper and van Doren, 2009; Horabin et al., 1995; Staab et al., 1996; Wawersik et al., 2005). Dictation by somatic signals continues through the larval stages, after which point XX germ cells control their own sexual development in a cell-autonomous manner.

In this study, we establish that, although Sxl does not play a cell-autonomous role in germ cell sex determination, it is necessary for the maintenance of female identity. We show that, when germ cells lack Sxl protein, tumor formation is accompanied by a global derepression of testis genes, including aberrant expression of the male germline sexual identity gene PHD finger protein 7 (Phf7) (Yang et al., 2012). We further ascertain that Phf7 drives tumor formation through a mechanism that includes the sex-inappropriate activation of Janus kinase/Signal transducer and activator of transcription (Jak/Stat) signaling in XX germ cells. Notably, the tumor phenotype depends on paracrine signals from neighboring somatic gonadal cells. Together with previous studies showing that male, but not female, germ cells are able to activate the Jak/Stat pathway in response to signals emanating from the somatic niche (Decotto and Spradling, 2005; Kiger et al., 2001; Leatherman and DiNardo, 2008,, 2010; López-Onieva et al., 2008; Tulina and Matunis, 2001; Wang et al., 2008), our work suggests that tumors in this model system form because mutant germ cells respond to their environment as if they were male germ cells. Remarkably, derepression of testis genes, including inappropriate Phf7 expression, is also observed in bam ovarian tumors. This work demonstrates that female GSCs must maintain their female sexual identity, through a mechanism that requires the concerted actions of Sxl and bam, as they differentiate or risk tumor formation.

Here, as in our earlier studies, we investigate Sxl function in germ cells by taking advantage of the viable sans fille (snf¹⁴⁸) allele, which interferes with Sxl regulation in germ cells without disrupting Sxl regulation and expression in the somatic cells of the ovary (Chau et al., 2009,, 2012). In wild-type females Sxl protein is present in all female somatic cells and accumulates to high levels in the cytoplasm of only a few germ cells located at the tip of the germarium (Chau et al., 2009) (Fig. 1B). By contrast, in snf¹⁴⁸ mutants the ovary is filled with proliferating undifferentiated germ cells that fail to accumulate Sxl protein even though Sxl protein is clearly detectable in the surrounding somatic cells (Chau et al., 2009) (Fig. 1C). Importantly, all aspects of the snf¹⁴⁸ germline tumor phenotype described to date can be restored by expression of P{otu::SxlcDNA}, a transgene that expresses Sxl cDNA under the control of a germline-specific promoter (Chau et al., 2009,, 2012; Nagengast et al., 2003).

To gain a better understanding of why the loss of Sxl protein in the germline leads to tumors, we used high-throughput sequencing (RNA-seq) to compare the transcriptomes of 3-day-old snf¹⁴⁸ tumorous ovaries with ovaries from 0- to 24-h-old virgin wild-type females (Fig. 1D,E). Because ovaries from young wild-type females lack late-stage egg chambers, we reasoned that this comparison would minimize the identification of gene expression changes unrelated to the mutant phenotype. Visualization of our RNA-seq data on the UCSC genome browser (UCSCdm3/FlyBase r5) showed that male-specific Sxl RNA products are detectable in tumors, but not in wild type (Fig. 1F). Sxl transcripts are sex-specifically spliced to produce mRNAs with different coding potentials (Salz, 2011; Salz and Erickson, 2010). In males, all transcripts include the translation-terminating third exon and encode truncated, inactive proteins. In females, the third exon is always skipped to generate protein-encoding mRNAs. Our finding that tumors contain male-specific transcripts is consistent with earlier studies documenting that loss of Sxl activity in the snf¹⁴⁸ germline is caused by a splicing defect that ultimately prevents Sxl protein production (Nagengast et al., 2003).

We identified 804 genes that change expression at least twofold in snf¹⁴⁸ tumorous ovaries relative to wild type (P<0.05). Of these, 483 are upregulated and 321 are downregulated (supplementary material Tables S1 and S2). Interestingly, Gene Ontology (GO) analysis identified a set of ‘male-meiosis’ (GO:0007140) genes that are significantly over-represented in the upregulated gene set (Fig. 2A). Among these genes are two testis-specific transcription factors, aly (a mip130 paralog) and rye (Taf12L – FlyBase; a Taf12 paralog), that are required for expression of the spermatocyte transcription program (White-Cooper, 2010). A visual survey of the most highly overexpressed genes in tumors revealed several more genes known to be specifically expressed in spermatocytes, such as the testis-specific transcription factor TfIIA-S2 (TFIIA gamma subunit) (Aoyagi and Wassarman, 2000), the bromodomain-containing proteins mitoshell (mtsh) (Bergner et al., 2010) and tbrd-1 (Leser et al., 2012), as well as the eIF4E-3 translation initiation factor (Hernández et al., 2012).

These observations suggested that the expression pattern of snf¹⁴⁸ tumors includes the anomalous activation of the testis gene expression program. Indeed, of the 483 upregulated genes, 42% (204/483) are expressed in tumors but not in wild type (FPKM<1; Fig. 2B; supplementary material Table S3). Moreover, an examination of the expression data provided by the modENCODE_mRNA-seq_tissues data sets (Brown et al., 2014) reveals that 68% (139/204) of these uniquely expressed tumor genes are expressed in testis (RPKM≥1). Together, these findings indicate that the tumor cells are trending towards a male fate.

Since male identity in germ cells is controlled by Phf7 (Yang et al., 2012), we asked whether it is inappropriately expressed in snf¹⁴⁸ tumors. Visualization of our RNA-seq data on the UCSC genome browser (UCSCdm3/FlyBase r5) reveals that the Phf7-RC transcript is expressed in snf¹⁴⁸ tumors but not in wild-type ovaries, whereas Phf7-RA is expressed in both samples (Fig. 3A). Published modENCODE RNA-seq analysis of adult tissues indicates that Phf7-RC is transcribed from an upstream testis-specific transcription start site (TSS) (Brown et al., 2014).

Using RT-qPCR, we confirmed that Phf7-RC RNA expression is sexually dimorphic, with higher levels in testis than in ovaries (supplementary material Fig. S1). Moreover, there is a significant increase in Phf7-RC levels in tumors as compared with wild-type ovaries, whereas the level of Phf7-RC is once more comparable to that of wild-type ovaries in a snf¹⁴⁸; P{otu::SxlcDNA} mutant background (Fig. 3B). Based on these data, we conclude that Phf7-RC misexpression is due to the lack of Sxl protein in the snf¹⁴⁸ mutant germline.

Although Phf7-RC has the same coding potential as Phf7-RA, recent studies have shown that Phf7 protein expression is limited to testis (Yang et al., 2012). There is no available antibody against Phf7. We therefore used a functional HA-tagged transgene (Yang et al., 2012) to follow Phf7 protein expression. This analysis confirmed that Phf7 protein expression is sex-specifically regulated, with expression detectable in testis but not in ovary (supplementary material Fig. S2). In contrast to wild-type ovaries, HA-Phf7 protein is detectable in snf¹⁴⁸ ovaries (Fig. 3C-F). In both testis and snf¹⁴⁸ mutant ovaries, expression of the tagged protein is detectable in the cytoplasm and in the nucleus, raising questions about whether the localization of the tagged protein accurately reflects the distribution of the endogenous protein. Nevertheless, our data strongly suggest that Phf7 protein is ectopically expressed in snf¹⁴⁸ tumors.

Next, we tested whether snf¹⁴⁸ tumors are dependent on the sex-inappropriate expression of Phf7, using RNA interference (RNAi) to knock down Phf7 expression. We achieved efficient depletion of Phf7 (>99% efficiency, assayed by RT-qPCR) by driving expression of a short hairpin RNA under the control of the upstream activating sequence (UAS) with the germ cell-specific nanos (nos)-Gal4 driver (van Doren et al., 1998). Remarkably, germline-specific knockdown of Phf7 in the snf¹⁴⁸ mutant background suppressed the tumor phenotype (Fig. 4). Control experiments, in which Phf7 was knocked down in the germline of wild-type females (nos>Phf7^RNAi) with comparable efficiency, had no effect on oogenesis (data not shown), confirming previously published data (Yang et al., 2012).

In contrast to snf¹⁴⁸ females, which are sterile with germline tumors, the majority of snf¹⁴⁸; nos>Phf7^RNAi females lay eggs and many, but not all, of their ovaries contain late stage egg chambers, indicating that inhibition of Phf7 expression suppresses the tumor phenotype. To quantitate the efficiency of suppression, we stained ovaries with an antibody against Orb, a marker for oocyte determination. In the wild-type ovary, Orb protein concentrates in the cytoplasm of the oocyte (Fig. 4A). In the majority of snf¹⁴⁸ mutant egg chambers, Orb staining remains diffuse, consistent with earlier findings that the tumor is filled with germ cells that are blocked prior to oocyte determination (Chau et al., 2009). Similarly, Orb staining is diffuse in the majority of control ovaries from snf¹⁴⁸ females carrying the driver alone (Fig. 4B). On occasion, however, we observe Orb protein accumulation in the cytoplasm of a mutant germ cell (2.5%, n=277), even though differentiation is clearly blocked and the ovariole remains tumorous (supplementary material Fig. S3A). In sharp contrast, 42% (n=196) of ovarioles obtained from snf¹⁴⁸; nos>Phf7^RNAi mutant females contained at least one egg chamber with localized Orb staining (Fig. 4C). Based on these data, we conclude that sex-inappropriate expression of Phf7 is a key cause of the snf¹⁴⁸ ovarian tumor phenotype.

Finding that snf¹⁴⁸ tumor formation is dependent on anomalous Phf7 expression suggests that forcing Phf7 expression in wild-type female germ cells might be sufficient for tumorigenesis. Previous studies have reported that forced Phf7 expression can be achieved by driving the expression of Phf7^EY03023, an EP insertion line that contains a UAS sequence inserted within the first intron of Phf7, with the germline-specific nos-Gal4 driver (Yang et al., 2012). However, it was reported that forced expression leads to an empty gonad phenotype due to germ cell death during the larval period (Yang et al., 2012). To bypass this early germ cell lethality, we induced ectopic expression in the adult using a temperature shift protocol in which tub-gal80^ts; nos>Phf7^EY03023 flies were reared at the restrictive temperature (18°C) and then shifted as adults to the permissive temperature (27°C) for 10 days to allow for significant expression of Phf7. We found that 18% (n=156) of the Phf7-expressing ovarioles showed evidence of tumor formation, based on the number of round spectrosome-like structures present in the germarium (Fig. 4F,G). The spectrosome is a spherical α-Spectrin-containing structure that is normally found only in GSCs and CB cells (∼5 per ovariole). During the subsequent divisions, the round spectrosome elongates and branches out to form a fusome that connects the synchronously dividing germ cells (Fig. 4D). Tumors in nos>Phf7^EY03023 ovaries are composed of cells that contain round spectrosomes or abnormal fusome-like structures, resembling snf¹⁴⁸ tumors (Fig. 4E,F). The low penetrance of the tumor phenotype could be due to the inefficient production of Phf7 protein or might indicate that other key downstream targets in addition to Phf7 are needed to foster tumorigenesis.

Interestingly, our RNA-seq data reveal that the levels of two Jak/Stat pathway activating cytokines, unpaired 2 (upd2) and unpaired 3 (upd3) (Agaisse et al., 2003; Gilbert et al., 2005; Hombría et al., 2005), and of a downstream effector of the Jak/Stat signaling pathway, chronologically inappropriate morphogenesis (chinmo) (Flaherty et al., 2010), are significantly increased in snf¹⁴⁸ ovarian tumors (supplementary material Table S2). Jak/Stat activity in adult germ cells is sexually dimorphic, being required in male but not female cells (Decotto and Spradling, 2005; Kiger et al., 2001; Leatherman and DiNardo, 2008,, 2010; Tulina and Matunis, 2001). Consistent with these earlier studies, we find that germline-specific knockdown of Stat92E or hopscotch (hop), the sole Jak kinase in the Drosophila genome, does not disrupt oogenesis (supplementary material Fig. S3B,C). Together, these data suggest the possibility that Jak/Stat is aberrantly activated in the snf¹⁴⁸ mutant germline.

Using RT-qPCR, we confirmed that there is a significant increase in upd2, upd3 and chinmo levels in tumors as compared with wild-type ovaries (Fig. 5A). The levels of these three transcripts are once more comparable to those of wild type in a snf¹⁴⁸; P{otu::SxlcDNA} mutant background, confirming that the effects that we are observing are due to the absence of Sxl protein in the mutant germline. Notably, we also observed a genetic dependence on inappropriate Phf7 expression, as the levels of these three transcripts were restored to near wild-type levels in a snf¹⁴⁸; nos>Phf7^RNAi mutant background. These data suggest a genetic pathway in which the lack of Sxl protein leads to anomalous Phf7 expression, which in turn results in increased upd2 and upd3 production, inappropriate Jak/Stat signaling, and chinmo expression (Fig. 5B).

In agreement with these findings, Jak/Stat activity, as assayed by staining with an antibody against Stat92E, is detectable in snf¹⁴⁸ tumors (Fig. 5C,D). Stat92E encodes the only Stat transcription factor in the Drosophila genome, and activation of the pathway is accompanied by increased levels or stability of Stat92E protein (Chen et al., 2003). In wild-type ovaries, Stat92E protein is highly expressed in the somatic cells of the germarium, but is not detectable in the germline of wild-type ovaries (Fig. 5C). By contrast, Stat92E is detectable throughout the snf¹⁴⁸ tumor (Fig. 5D).

These observations, together with studies that have connected both aberrant Jak/Stat signaling and ectopic chinmo expression to tumor formation in other contexts (Flaherty et al., 2010), suggest that anomalous activation of this pathway might play a role in the formation of snf¹⁴⁸ germline tumors. Consistent with this hypothesis, we find suppression of the tumor phenotype upon inactivation of key components of the Jak/Stat pathway (Fig. 6). When Stat92E was knocked down in a tumor background, 12% (n=254) of the snf¹⁴⁸; nos>Stat92E^RNAi mutant ovarioles contained at least one egg chamber with localized Orb staining (Fig. 6A). Similarly, when hop was knocked down, 16% (n=178) of the snf¹⁴⁸; nos>hop^RNAi mutant ovarioles contained at least one egg chamber (Fig. 6B).

In wild-type ovaries, the Upd2 and Upd3 cytokines are thought to be produced by the somatic terminal filament (TF) cells located at the tip of the germarium (López-Onieva et al., 2008). It is possible, therefore, that the snf¹⁴⁸ mutant germ cells have acquired a novel ability to receive signals from the surrounding somatic cells that normally produce these signals. On the other hand, the increased level of upd2 and upd3 observed in tumors is dependent on aberrant Phf7 expression in germ cells (Fig. 5A), raising the possibility that snf¹⁴⁸ germ cells produce these cytokines and activate Jak/Stat in an autocrine manner. These two possibilities are not mutually exclusive, as the tumor phenotype might reflect a novel response to paracrine signals as well as an additional production and response to autocrine signals.

These scenarios may be distinguished by determining the cellular source of the cytokines important for snf¹⁴⁸ tumor formation. We therefore performed tissue-specific knockdown of upd2 and upd3 in the snf¹⁴⁸ mutant germline (with the nos-Gal4 driver), TF cells [with the bab1-Gal4 driver (Cabrera et al., 2002)] and escort cells [with the c587-Gal4 driver (Song and Xie, 2003)]. Interestingly, we observed that knockdown in the germline and in TF cells, but not in escort cells, led to suppression of the tumor phenotype (Fig. 6C). In germ cells, knockdown of upd2 effectively suppressed the tumor phenotype, with 24% (n=333) of ovarioles obtained from snf¹⁴⁸; nos>upd2^RNAi females containing at least one egg chamber with localized Orb staining. Knockdown of upd3 in the germline was significantly weaker, with only 2% (n=410) of snf¹⁴⁸; nos>upd3^RNAi mutant ovarioles showing evidence of Orb protein localization. In TF cells, both upd2-RNAi and upd3-RNAi effectively suppressed the tumor phenotype; 41% (n=90) of snf¹⁴⁸; bab1>upd2^RNAi ovarioles and 46% (n=166) of snf¹⁴⁸; bab1>upd3^RNAi ovarioles showed evidence of Orb localization. By contrast, knockdown in escort cells had no effect on the tumor phenotype; no Orb localization was observed in snf¹⁴⁸; c587>upd2^RNAi ovarioles (n=210) or in snf¹⁴⁸; c587>upd3^RNAi ovarioles (n=212). Overall, these studies demonstrate the importance of both autocrine and paracrine signaling in establishing and/or maintaining the tumor phenotype.

bam mutant ovaries, like snf¹⁴⁸ ovaries, are highly enriched for proliferating germ cells that fail to differentiate (McKearin and Ohlstein, 1995; McKearin and Spradling, 1990). We, and others, have previously noted the anomalous expression of several testis transcripts in bam ovarian tumors (Chau et al., 2009; Kai et al., 2005; Wei et al., 1994). To determine whether there is a global depression of testis genes in bam ovarian tumors, we examined the previously published RNA-seq analysis in which the transcriptomes of bam tumorous and wild-type ovaries from 0- to 24-h-old females were compared (Gan et al., 2010). From this analysis we identified 2661 genes that are upregulated at least twofold in bam tumors. Of these, 48% (1278/2661) are expressed in tumors but not in wild type (RPKM<1; Fig. 7A). Furthermore, 54% (692/1278) of these uniquely expressed tumor genes are expressed in testis (RPKM≥1). Our finding that the bam mutant gene expression program is masculinized is further supported by RT-qPCR analysis showing aberrant expression of the male-specific transcript Phf7-RC in bam^Δ86 ovarian tumors (Fig. 7B).

Together, these expression studies suggest that the failure to maintain sexual identity, as indicated by the global depression of testis genes, is a common feature of both snf and bam tumors. Interestingly, although there is a strong overlap between the uniquely expressed genes in snf and bam tumors, the overlap is not complete (Fig. 7C; supplementary material Table S4). We do not yet know whether these differences are important, but our finding that the bam^Δ86 tumor phenotype is not suppressed by germline-specific knockdown of Phf7 (supplementary material Fig. S4) suggests that there are fundamental biological differences between bam and snf tumors.

When housed in a normal ovary, female germ cells lacking Sxl are unable to differentiate and instead form tumors. Although Sxl has a pivotal role in the cell fate switch from a self-renewing GSC to a differentiation-competent CB, tumor formation is not simply the result of the failure of this switch (Chau et al., 2009,, 2012). By taking advantage of a snf mutant allele that specifically eliminates Sxl protein expression in the germline, we have shown that it is the loss of female sexual identity and the acquisition of male characteristics that leads to tumor formation. Our finding that Sxl jointly controls exit from the stem cell state and the maintenance of sexual identity provides the basis for a model (Fig. 8A) in which Sxl safeguards the previously made sexual decision as the stem cell reprograms itself towards a differentiated fate.

Taking a genome-wide approach, we establish that Sxl is a potent repressor of testis genes. This global analysis supports and extends earlier studies that identified several inappropriately expressed testis genes in ovarian tumors (Chau et al., 2009; Wei et al., 1994). Although these mutant germ cells are not sex reversed, as they continue to express ovary-specific genes, the degree to which they are in fact masculinized is illustrated by our key observation that Phf7, a gene known to be required for male germline sexual identity (Yang et al., 2012), is expressed in the absence of Sxl. This suggests that Sxl maintains female identity by silencing Phf7. Sex- and tissue-specific regulation of Phf7 appears to be achieved by a mechanism that involves alternative TSSs and translational regulation. Although Sxl is known to inhibit translation in other contexts (Salz and Erickson, 2010), the lack of putative Sxl binding sites (multiple polyuridine runs of seven or more nucleotides) in Phf7 transcripts argues that Sxl is likely to regulate Phf7 expression indirectly.

Interestingly, Phf7 encodes a methyl-histone-binding protein that preferentially recognizes lysine 4-dimethylated histone H3 (H3K4me2) (Yang et al., 2012). We therefore propose that Sxl maintains germline sexual identity by impacting epigenetic regulation. In this model, the global depression of testis genes observed in the absence of Sxl would be due to unscheduled epigenetic changes. In this regard, we find it intriguing that disrupting the H3K4 methylation pathway in female germ cells by knocking down Set1, the H3K4 di- and trimethyltransferase, also leads to germ cell tumors (Yan et al., 2014). A future challenge will be to understand the molecular connection between Sxl, which is an RNA-binding protein, and sexually dimorphic epigenetic programming.

Through our pathway analysis (Fig. 8B) we establish that tumor formation is caused by the activation of a sex-inappropriate gene expression network that is unleashed by the loss of Sxl protein. Phf7 is a key effector of this pathway, as we have shown that it is both necessary and sufficient for tumor formation. Our findings indicate that Phf7 drives tumor formation through inappropriate activation of the Jak/Stat pathway. It is intriguing that activation of Jak/Stat signaling leads to tumor formation because in male GSCs activity is only needed for adhesion to the somatic hub cells (Leatherman and DiNardo, 2010). It is possible, therefore, that the tumors express a Stat-regulated pathway that is unrelated to the pathway expressed in the male germline. Our data also suggest that Phf7 drives tumor formation through a mechanism that leads to increased levels of two of the Upd cytokine family members known to activate the Jak/Stat pathway. How inappropriate Phf7 expression leads to upregulation of Upd cytokines remains an open question. One possibility, suggested by our transcriptional profiling experiments, is that Phf7 expression leads to the derepression of genes that include, but is not limited to, testis-specific genes. Irrespective of the mechanism, the finding that depletion of Upd2, and to a lesser extent Upd3, in germ cells reverts the tumor phenotype indicates a requirement for autocrine signaling. Furthermore, aberrant autocrine signaling is consistent with data emerging from numerous studies connecting hyperactive Jak/Stat signaling to other Drosophila tumor models and human cancers (Amoyel et al., 2014).

A remarkable and unexpected aspect of our analysis is that tumor formation also depends on Upd2 and Upd3 produced by somatic cells in the neighboring microenvironment. The somatic microenvironment at the tip of the germarium consists of three cooperating cell types: TF cells, cap cells and escort cells. In wild type, secretion of the Upd family of cytokines from TF cells activates Jak/Stat signaling in cap and escort cells, but not in the adjacent germ cells (Decotto and Spradling, 2005; López-Onieva et al., 2008; Wang et al., 2008). The situation is different in males, where somatic cytokine production activates Jak/Stat in neighboring somatic and germ cells (Kiger et al., 2001; Leatherman and DiNardo, 2008,, 2010; Tulina and Matunis, 2001). Because germ cells lacking Sxl protein are masculinized, we hypothesize that mutant germ cells interpret the information provided by the microenvironment as if they were male, leading to sex-inappropriate Jak/Stat activation. Although we do not understand how these mutant cells acquired the ability to receive activating signals from the surrounding somatic cells, these results highlight the importance of interactions between mutant cells and the surrounding normal cells that make up the tumor microenvironment in driving tumor formation.

Masculinization of the gene expression program is also observed in bam ovarian tumors, suggesting that maintenance of sexual identity requires both Sxl and bam. A Sxl-bam partnership is strongly supported by genetic epistasis experiments, which show that the two proteins jointly repress expression of the stem cell maintenance factor nos (Chau et al., 2009,, 2012; Li et al., 2013). Based on these observations we propose that Sxl and bam co-regulate at least two independent pathways, one of which leads to germ cell differentiation whereas the other maintains female identity (Fig. 8A).

Altogether, our studies establish that sexual identity must be maintained in an adult stem cell lineage, or risk tumor formation. Although the gene regulatory networks that specify sex vary between species, the need for a germ cell to remember its sexual identity is likely to extend beyond Drosophila. In humans, germ cell tumors occur most frequently in individuals with intersex disorders (Hersmus et al., 2012; Kraggerud et al., 2013; Pleskacova et al., 2010). There is also increasing evidence of a connection between testicular germ cell tumors and the disruption of sex-specific processes (Kanetsky et al., 2011; Koster et al., 2014; Matson et al., 2011; Turnbull et al., 2010). Non-reproductive cells and tissues also exhibit sexually dimorphic gene expression programs, suggesting that the failure to maintain this dimorphism could have a role in cancer and other diseases (Bellott et al., 2014; Nakada et al., 2014; Ronen and Benvenisty, 2014). It will be important, therefore, to determine whether there is a mechanism to maintain sexual identity in tissues other than gonads and whether the failure in this process leads to disease.

Flies were reared on standard media at 25°C unless otherwise indicated. Mutant alleles, transgenic, enhancer trap and Gal4 drivers used in this study include snf¹⁴⁸ [Bloomington Drosophila Stock Center (BDSC) #7398 (Nagengast et al., 2003)], P{HA-Phf7} (Yang et al., 2012), Phf7^EY03023 (BDSC #15894), nanos-Gal4 [BDSC #4937 (van Doren et al., 1998)], c587-Gal4 (Song and Xie, 2003), bab1-Gal4 [BDSC #6802 (Cabrera et al., 2002)], tub-Gal80^ts [BDSC #7019 (McGuire et al., 2003)] and bam^Δ86 [BDSC #5427 (McKearin and Ohlstein, 1995)]. Additional information about the Drosophila strains used in this study is available from FlyBase (http://flybase.org).

Knockdown studies with UAS-RNAi were carried out at 27°C with the following lines from the Drosophila Transgenic RNAi Project (TRiP; Ni et al., 2011): Phf7-P{TRiP.GL00455} (BDSC #35807), stat-P{TRiP.GL00437} (BDSC #35600), hop-P{TRiP.GL00305} (BDSC #35386), upd2-P{TRiP.HMS00948} (BDSC #33988) and upd3-P{TRiP.HMS00646} (BDSC #32859). The HMS RNAi lines are constructed in the VALIUM20 vector (upd2, upd3) and are expressed in both somatic and germline cells. The GL RNAi lines are constructed in the VALIUM22 vector [Phf7, stat (Stat92E), hop] and show strong expression in the female germline and weak expression in the soma. The Gal4/Gal80^ts system was used to force Phf7 expression in adult germ cells. Animals were raised at 18°C to maintain the Gal80-dependent repression of Gal4 until adulthood. To induce Gal4 activity, adult females were transferred to 27°C for 10 days before immunostaining.

Total RNA was isolated using the RNeasy Kit (Qiagen), including the on-column DNase I digestion. RNA purity was assessed with the Nano 6000 RNA chip on an Agilent 2100 bioanalyzer, and quantification was performed using a RiboGreen fluorescence assay (Invitrogen). RNA-seq libraries were constructed using the Illumina Truseq Total Stranded RNA Kit, with 800 ng of input total RNA and 14 cycles of PCR amplification. 100 bp paired-end mRNA sequencing was performed on biological duplicates from each genotype on an Illumina HiSeq 2500 by the CWRU Genomics Sequencing Core. After quality assessment, the sequenced reads were aligned to the D. melanogaster genome (UCSCdm3/FlyBase r5.23) using TopHat (2.0.9) (Trapnell et al., 2009) with the RefSeq annotated transcripts as a guide. The differential expression analysis was performed using CuffDiff (2.0.2) (Trapnell et al., 2010). Genes with P≤0.05 after adjusting to a false discovery rate (FDR) with the Benjamini and Hochberg method and a twofold or higher change were considered significant. GO term enrichment analysis was performed separately on the upregulated and downregulated gene sets using High-Throughput GoMiner (Zeeberg et al., 2005).

Total RNA was isolated using TRIzol (Invitrogen) and treated with DNase (Promega). RNA yield and quality were assessed with a NanoDrop spectrophotometer (NanoDrop Technology), followed by reverse transcription using random hexamers with the SuperScript First-Strand Synthesis System Kit (Invitrogen). Transcript levels were measured with Quanta PerfeCTa SYBR Fastmix (VWR) in an MJ Research PTC-200 gradient cycler. Initial activation was performed at 95°C for 30 s, followed by 39 cycles of: 95°C for 5 s, 60°C for 15 s, 70°C for 10 s. The melting curve was generated ranging from 50°C to 95°C with an increment of 0.5°C each 5 s. Primers used for RT-qPCR are listed in supplementary material Table S5. Measurements were performed on biological triplicates, with technical duplicates of each biological sample. chinmo, upd2 and upd3 RNA levels were normalized to rp49 (RpL32 – FlyBase). Phf7-RC levels were normalized to the total level of Phf7 using a primer set that detects both Phf7-RA and Phf7-RC. The relative transcript levels were calculated using the 2^−ΔΔC_T method (Livak and Schmittgen, 2001).

Ovaries and testes were fixed and stained by standard methods. The following primary antibodies were used: mouse Sxl-M18 [1:100, Developmental Studies Hybridoma Bank (DSHB)], mouse Orb-4H8 (1:50, DSHB), mouse Orb-6H4 (1:50, DSHB), α-Spectrin (1:100, DSHB), rat HA high affinity (1:500, Roche, 11867423001), rat Vasa (1:100, DSHB), rabbit Stat92E [1:1000 (Chen et al., 2002)] and mouse Fas3-7G10 (1:50, DSHB). Secondary antibodies conjugated to Alexa Fluor 555 (Life Technologies) or FITC (Jackson ImmunoResearch Labs) were used at 1:200. Images were acquired on a Leica TCS SP8 confocal microscope and assembled using Photoshop (Adobe) and PowerPoint (Microsoft). A full description of these data can be found at http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE65932

We thank M. Van Doren, J. McDonald, the TriP at Harvard Medical School, the Bloomington Drosophila Stock Center and the Iowa Developmental Studies Hybridoma Bank for fly stocks and antibodies; the CWRU GTSC sequencing core for RNA-seq library generation and sequencing; A. Miron and S. Bai for help with the bioinformatic analysis; and R. Conlon, I. Greenwald, J. McDonald, M. Van Doren and M. Wawersik for helpful discussions.