Germline sexual fate is determined by the antagonistic action of dmrt1 and foxl3/foxl2 in tilapia

In vertebrates, the sexual fate of the undifferentiated gonads is controlled by genetic factors that determine entry into the male or female developmental pathway. It is well accepted that the sexual fate of germ cells is determined by factors derived from somatic cells (Capel, 2017). Sex determination of somatic cells precedes that of germ cells, as demonstrated by the identification of several master sex determining genes, which are specifically expressed in somatic cells, including SRY/Sry in human (Homo sapiens) and mouse (Mus musculus) (Sinclair et al., 1990; Koopman et al., 1991), DMY (dmrt1bY) in medaka (Oryzias latipes) (Matsuda et al., 2002; Nanda et al., 2002) and amhy in Patagonian pejerrey (Odontesthes hatcheri) and Nile tilapia (Oreochromis niloticus) (Hattori et al., 2012; Li et al., 2015). In addition, sex reversal in somatic cells leads to sex reversal in germ cells (Bowles and Koopman, 2010; Capel, 2017). However, we still know very little about how signals from somatic cells determine the sexual fate of germ cells.

It is well documented that Foxl2 is involved in ovarian differentiation in vertebrates (Bertho et al., 2016). FOXL2 is an ovarian determining gene in goats because mutation of FOXL2 induced XX sex reversal (Boulanger et al., 2014). Loss of Foxl2 in adult mouse granulosa cells activates Dmrt1 expression and reprograms granulosa cells into Sertoli cells (Uhlenhaut et al., 2009). In lower vertebrates, mutation of foxl2 in tilapia and zebrafish (Danio rerio) also triggers the expression of male pathway genes and sex reversal (Wang et al., 2007; Li et al., 2013; Zhang et al., 2017; Yang et al., 2017). A recent study in medaka revealed the crucial function of foxl3, a paralog of foxl2 found in most vertebrates, except placental mammals (Crespo et al., 2013; Geraldo et al., 2013; Bertho et al., 2016), in the germline sexual fate decision. Disruption of foxl3 in medaka, which is specifically expressed in germ cells, initiates spermatogenesis in a female gonadal environment, thus identifying a new mechanism for germline sexual fate decision in vertebrates (Nishimura et al., 2015). Further study has shown that foxl3 initiates oogenesis by directly regulating rec8a and fbxo47 in medaka (Kikuchi et al., 2019, 2020). However, it is unclear whether foxl3 is a conserved factor for germline sexual fate decision in other fish species. In addition, it is important to know which gene is activated to determine male germ cell fate in the foxl3 mutant ovary. In other words, which gene represses foxl3 expression in male germ cells?

Dmrt1 is a functionally conserved factor that promotes testicular differentiation across vertebrates. It plays crucial roles in vertebrate sex determination/differentiation by antagonizing Foxl2 (Smith et al., 2009; Matson and Zarkower, 2012; Li et al., 2013; Lindeman et al., 2015; Lin et al., 2017; Webster et al., 2017). Dmrt1 is expressed in both spermatogonia and Sertoli cells in mouse testes (Raymond et al., 2000; Krentz et al., 2009; Matson et al., 2010; Zhang et al., 2016; Stévant et al., 2019). Although Dmrt1 is not required for fetal sex determination in mammals (Raymond et al., 2000), loss of Dmrt1 in adult Sertoli cells activates Foxl2 and reprograms Sertoli cells into granulosa cells, indicating its role in maintaining male sexual fate of somatic cells (Bowles et al., 2006; Matson et al., 2011). Expression of Dmrt1 in germ cells is required for maintenance and replenishment of mouse spermatogonial stem cells (Zhang et al., 2016). Dmrt1 inhibits meiosis by suppressing retinoic acid signaling, which is essential for germ cell fate decision in mammals (Minkina et al., 2014; Bowles et al., 2018). Expression of dmrt1 was also detected in the Sertoli cells and spermatogonia in a number of fish species (Guo et al., 2005; Raghuveer and Senthilkumaran, 2009; Webster et al., 2017; Jeng et al., 2019). Recent knockout studies in zebrafish and medaka have verified that loss of dmrt1 upregulates foxl2 and cyp19a1a to induce ovarian development (Wang et al., 2010; Masuyama et al., 2012; Lin et al., 2017; Webster et al., 2017). Our previous studies have shown that overexpression of Dmrt1 in XX tilapia decreased Cyp19a1a and estrogen levels to induce testicular development (Wang et al., 2010). Conversely, knockdown of Dmrt1 in F0 XY tilapia feminized the somatic cells, as demonstrated by upregulation of the expression of foxl2 and cyp19a1a. However, no male-to-female sex reversal was observed in the dmrt1 F0 mutants (Li et al., 2013). Whether Dmrt1 represses female germ cell fate through inhibition of foxl3/foxl2 is unknown.

The Nile tilapia, a gonochoristic teleost fish with an XX/XY sex-determining system, provides a good model for studying gonadal sex determination/differentiation because genetic all-females and all-males are available (Kobayashi and Nagahama, 2009). During gonadal development, the differential expression of genes in XX and XY gonads at 5-6 days after hatching (dah) is crucial for the differentiation of gonads into either the ovary or testis (Ijiri et al., 2008; Tao et al., 2013). The first signs of gonadal sex differentiation appear in tilapia fry at 23-26 dah, with the formation of the ovarian cavity in the XX gonad or the efferent duct in the XY gonad. Germ cell meiosis begins in XX gonads at 25-30 dah, but does not begin in XY gonads until 60-85 dah (Kobayashi et al., 2000). In this study, we examined the expression and transcriptional regulation of dmrt1 and foxl3, and generated single and double mutants of these two genes in Nile tilapia. Our data demonstrate that Dmrt1 and Foxl3 play antagonistic roles in germline sexual fate decision via mutual transcriptional regulation, and highlight the role of the somatic environment in germ cell sex determination.

The foxl3 gene was isolated from the tilapia genome and confirmed by phylogenetic analysis (Fig. S1A,B). RT-PCR revealed that foxl3 mRNA was expressed in the adult tilapia ovary but not in the testis (Fig. S1C). Gonadal transcriptome analyses showed that foxl3 mRNA was expressed exclusively in XX tilapia gonads, with the highest expression level at 30 dah (Fig. S1D). Fluorescence in situ hybridization (FISH) using a foxl3 anti-sense probe and immunofluorescence (IF) using the Vasa antibody revealed that foxl3 mRNA was specifically expressed in the cystic germ cells in the ovary at 60 dah. A very weak signal of foxl3 was also detected in the stage I oocyte but not in the testis (Fig. S1E).

We used CRISPR/Cas9 to generate two foxl3 mutant lines, with 5 and 7 bp frameshift deletions before the forkhead domain, which created truncated Foxl3 protein (Fig. S2A-E; Table S1). The loss of intact foxl3 mRNA was also confirmed by RT-PCR using a specific primer located on the target site (Fig. S2F). Histological analyses of the wild-type (WT) XX ovary showed it filled with a number of oocytes. In contrast, mutation of foxl3 in XX tilapia resulted in 100% of the gonads being filled with different stages of spermatogenic cells, including spermatogonia, primary and secondary spermatocytes, and spermatids, at 120 dah (n=20: 5 for foxl3^Δ5/Δ5, 15 for foxl3^Δ7/Δ7). The ovarian cavity, which depends on estrogen for its formation (Suzuki et al., 2004), was also observed in the gonads of foxl3^Δ5/Δ5 and foxl3^Δ7/Δ7 XX tilapia (Fig. 1A; Fig. S2G). The germ cell type in the foxl3^Δ7/Δ7 XX tilapia was further demonstrated by the presence of spermatogenesis markers dmrt6 (Zhang et al., 2014), kif17, protamine and Sox30, and the absence of oocyte markers bmp15 and 42sp50 (Fig. 1B,C). Real-time PCR and IF showed that female-specific markers foxl2 and Cyp19a1a were expressed in the foxl3^Δ7/Δ7 ovaries but at significantly lower levels than those of the WT XX ovaries, whereas Sertoli cell markers dmrt1, Amh and Gsdf, were significantly upregulated compared with the WT XX ovaries but significantly lower than those of the WT XY testes (Fig. 1D,E; Fig. S3A,B). The results demonstrated the partial masculinization of somatic cells, and spermatogenesis of germ cells, in foxl3^Δ7/Δ7 XX mutants. No expression of Cyp11c1, a Leydig cell marker, was observed in the foxl3^Δ7/Δ7 ovaries (Fig. 1D). Consistently, the serum 11-ketotestosterone (11-KT) level in the foxl3^Δ7/Δ7 XX mutants was comparable with that of the WT XX fish. The serum estradiol-17β (E2) level in the foxl3^Δ7/Δ7 XX mutants was significantly lower than that of the WT XX fish, but significantly higher than that of the WT XY fish (Fig. S3C). However, a few oocytes were observed in the foxl3^Δ7/Δ7 XX mutants (n=5) at 210 dah, and various stages of follicles and degenerated testis tissue were present in the foxl3^Δ7/Δ7 XX mutants (n=6) at 360 dah (Fig. S4A-D), indicating oogenesis independent of foxl3 in the later developmental stages, as reported in medaka (Nishimura et al., 2015). In contrast, foxl3^Δ7/Δ7 XY mutants developed normal testes with functional sperm, like the WT XY fish (n=6) (Fig. S4E-H; Table S2), indicating that foxl3 is dispensable for testicular development.

FISH analysis showed that dmrt1 mRNA was expressed in both spermatogonia and Sertoli cells (Fig. S5A,B,G-J), but with no detectable expression in ovary in tilapia (Fig. S5C,D). To investigate the role of dmrt1 in tilapia sex determination, a new gRNA target was designed in exon1 to mutate the dmrt1 gene with high efficiency by CRISPR/Cas9. Two dmrt1 mutant lines were generated, with 2 and 5 bp deletions, which resulted in truncation of the Dmrt1 protein before the DM domain (Fig. S6A-E; Table S1). Western blot (WB) analysis did not detect any Dmrt1 protein in the gonads of dmrt1^Δ5/Δ5 XY mutants, confirming Dmrt1 loss-of-function (Fig. S6F). Histological analysis of the gonads showed that mutation of dmrt1 in XY fish resulted in 100% male-to-female sex reversal and normal oogenesis like the WT XX ovary at 60 dah (n=13: 5 for dmrt1^Δ2/Δ2, 8 for dmrt1^Δ5/Δ5) (Fig. 2A-C; Fig. S6G), demonstrating that dmrt1 is essential for tilapia male sex determination. In agreement with this, the female markers foxl2, Cyp19a1a and 42sp50 were detected, whereas the male markers Amh, Cyp11c1, Gsdf and dmrt6 in the dmrt1^Δ5/Δ5 XY gonads were as low as those in the WT XX ovaries at 60 dah (Fig. 2Aa-Cd; Fig. S7A-E). Cyp19a1a was upregulated, whereas Amh and Gsdf were downregulated in the gonads of dmrt1^Δ5/Δ5 XY mutants, which was similar to the WT XX at 15 dah (Fig. 2D-F). In contrast, histological examination of the dmrt1^Δ5/Δ5 XX and XY gonads (n=10) showed that all fish exhibited normal oogenesis, with no observable difference at adulthood compared with the WT XX fish (Fig. S7F-K), indicating that Dmrt1 is not required for female ovarian development.

IF analysis showed that Amhy, which is encoded by the sex determining gene of Nile tilapia (Li et al., 2015), was expressed in the gonads of dmrt1^Δ5/Δ5 XY mutants, similar to the WT XY fish at 4 dah (Fig. 2G-I). However, Gsdf, which was demonstrated to act downstream of dmrt1 (Jiang et al., 2016), was not expressed in the gonads of dmrt1^Δ5/Δ5 XY mutants compared with the WT XY fish at 4 dah (Fig. 2Ga-Ia). Our data revealed that the amhy-dmrt1-gsdf gene cascade constitutes the male sex determining pathway in tilapia.

It is unclear how the germ cell fate will be determined if both dmrt1 and foxl3 are lost. To answer this question, we generated the dmrt1 and foxl3 double mutants using dmrt1^+/Δ5;foxl3^+/Δ7 double heterozygous males and females. Different from the phenotype of the foxl3^Δ7/Δ7 XX mutants (n=10), the gonads of the dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XX mutants displayed ovarian structure with numerous oocytes at 60 and 90 dah (n=6) (Fig. 3A-E; Fig. S8A-E). This was further supported by significant upregulation of the oocyte-specific marker 42sp50 in the dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XX ovaries compared with the foxl3^Δ7/Δ7 ovaries (Fig. 3Aa-Ea). In addition, a few spermatocyte-like cells were observed in the dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XX mutant ovaries (Fig. S8E). Therefore, initiation of spermatogenesis in the foxl3^Δ7/Δ7 XX mutants is caused by activation of dmrt1 expression.

Histological examination of the dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XY mutants showed that the gonads developed as ovotestes, with spermatogenesis occurring at 60 and 90 dah (n=6) (Fig. 3F; Fig. S8F), indicating that disruption of foxl3 partially rescued the germ cell sex reversal in the dmrt1^Δ5/Δ5 XY fish. The somatic cells were feminized in the dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XY mutants, as indicated by strong expression of female markers Cyp19a1a and foxl2 (Fig. 3Ab-Fb; Fig. S8G), and weak expression of male somatic cell markers Amh and Gsdf, as seen in WT ovaries (Fig. 3Ac-Fc,Ad-Fd; Fig. S8H,I). E2 levels in the dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XY mutants were comparable with the WT XX fish (Fig. S8J). The spermatogenic cell markers Creb1b and dmrt6 (Wang et al., 2019), were upregulated in the dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XY mutants compared with those of the dmrt1^Δ5/Δ5 XY mutants (Fig. 3Ae-Fe; Fig. S8G). These results imply that Dmrt1 determines the male germline sexual fate by downregulation of foxl3.

The rescue of germ cell sex reversal by disruption of dmrt1 in foxl3^Δ7/Δ7 XX mutants led us to examine the expression profile of dmrt1 in foxl3^Δ7/Δ7 ovaries. RT-PCR and WB analyses showed that dmrt1/Dmrt1 was significantly upregulated in the foxl3^Δ7/Δ7 ovaries compared with the WT XX ovaries from fish at 60 and 120 dah (Fig. 4A,B). FISH showed that dmrt1 mRNA was mainly expressed in germ cells in the foxl3^Δ7/Δ7 ovaries (n=6) but not in the WT XX ovaries (n=6) (Fig. 4C). We conclude that upregulation of Dmrt1 in germ cells after loss of foxl3 is responsible for inducing spermatogenesis in the female gonadal environment.

Luciferase analysis showed that Foxl3 directly repressed dmrt1 transcription in a dose-dependent manner in human embryonic kidney 293 (HEK293) cells (Fig. 4D). Five potential Foxl3 binding sites were predicted, and a series of dmrt1 promoter deletion constructs were generated (Fig. S9A). Foxl3 repressed the transcriptional activity of the −2532 bp deletion construct, whereas it had no repressive effect on the −2000 bp and −1500 bp deletion constructs, indicating that the binding site located in the −2532 to −2000 bp region from the start codon ATG was responsible for Foxl3-mediated repression. A potential Foxl3 binding site GGAAACA (reverse complement sequence of TGTTTCC), similar to the 7 bp forkhead factor binding motif, 5′-(G/A)(T/C)(C/A)AA(C/T)A-3′ (Gómez-Ferrería and Rey-Campos, 2003), was found in this region. Mutation of the potential Foxl3 binding site from TGTTTCCATC to GGGGGCCCCC (Fig. S9B) significantly decreased the transcriptional repression efficiency (Fig. 4E). The Foxl3 binding site was further confirmed by electrophoretic mobility shift assay (EMSA) using probe and mutated probe designed according to the sequences between −2523 to −2491 bp of the dmrt1 promoter. Foxl3 protein bound to the biotin-labeled probe, producing a specific band of protein/DNA complex. Cold competitor (10×-100×) displaced the band in a dose-dependent manner, whereas the mutated cold probe could not (Fig. 4F). In addition, compared with IgG antibody as a negative control, co-incubation of HA-antibody with nucleoproteins extracted from HEK293 cells overexpressing HA-Foxl3 caused a loss of Foxl3 binding to the biotinylated DNA probes (Fig. S9C). Furthermore, chromatin immunoprecipitation (ChIP) analysis using a Nile tilapia Leydig stem cell line revealed that a PCR band covering the binding site for Foxl3 was detected in the chromatin precipitated with the antibody against HA. In contrast, no band was detected in the chromatin precipitated with the nonspecific IgG as a negative control (Fig. 4G). These results indicate that Foxl3 negatively regulates transcription of dmrt1 by direct binding to its promoter.

Real-time PCR and FISH analyses showed that foxl3 mRNA was significantly upregulated in the dmrt1 mutant XY gonads compared with the WT XY gonads from fish at 30 dah (Fig. 5A,B). Luciferase assays revealed that Dmrt1 repressed foxl3 transcription in a dose-dependent manner in HEK293 cells (Fig. 5C). Two potential Dmrt1 binding sites were predicted and a series of foxl3 promoter deletion constructs were generated (Fig. S10A). Deletion of the −1450 to −700 bp from the start codon ATG had no effect on repression mediated by Dmrt1. In contrast, the repression was considerably decreased by deletion of the −700 to −600 bp. A potential Dmrt1 binding site ACAATGT (−695 to −688 bp), similar to the 7 bp DM domain protein recognition motif, 5′-ACA(A/T)TGT-3′ (Murphy et al., 2007), was found in this region. As expected, mutation of this motif to GGGGCCCC considerably decreased the transcription repression efficiency of Dmrt1 (Fig. 5D; Fig. S10B).

To determine whether the DM domain is important for its suppressive effect, we used a Dmrt1 mutant without the DM domain (DM6) (Wang et al., 2010) to perform the luciferase assay. Co-transfection of DM6 with the −1450 bp foxl3 promoter-reporter construct in HEK293 cells revealed that the DM domain was essential for the repression (Fig. 5D). EMSAs were performed using probe and mutated probe designed according to the sequences between −708 to −675 bp of foxl3 promoter. The results showed that Dmrt1 protein bound the biotin-labeled probe, producing a specific band of protein/DNA complex. Cold competitor (10×-100×) displaced the band in a dose-dependent manner, whereas the mutated cold probe could not (Fig. 5E). Compared with IgG antibody as a negative control, co-incubating nucleoproteins extracted from HEK293 cells overexpressing Flag-Dmrt1 with Flag-antibody caused a loss of Dmrt1 binding to the biotinylated probes (Fig. S10C). In addition, ChIP analysis using cultured Nile tilapia Leydig stem cells revealed that a positive DNA fragment was detected in the chromatins precipitated with the antibodies against Flag or Dmrt1. In contrast, this DNA fragment was not detected in the chromatin precipitated with the nonspecific IgG as a negative control (Fig. 5F). These results indicated that Dmrt1 directly represses transcription of foxl3 by binding to its promoter.

Estrogen is crucial for female sex determination and differentiation in fish (Guiguen et al., 2010). It would be interesting to know whether blockage of estrogen synthesis could prevent the male-to-female sex reversal caused by dmrt1 mutation, in which Cyp19a1a expression was significantly upregulated during sex determination. Unexpectedly, administration of fadrozole, an AI, from 5 to 30 dah failed to masculinize dmrt1^Δ5/Δ5 XY and XX mutants, but successfully masculinized the dmrt1^+/Δ5 XX fish (Fig. 6A-D; Fig. S11). This supports the idea that female-to-male sex reversal induced by blockage of estrogen action requires upregulation of dmrt1. As described above, mutation of foxl3 in dmrt1^Δ5/Δ5 XY mutants resulted in ovotestes with Cyp19a1a expression. However, administration of AI to dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XY or XX mutants from 5 to 30 dah induced testicular development as demonstrated by presence of spermatogenic cells and increased expression of somatic cell marker Amh (Fig. 6E-H; Fig. S12A-E), indicating that mutation of foxl3 and blockage of estrogen synthesis rescued the germ cell sex reversal caused by loss of Dmrt1.

Administration of E2 from 5 to 30 dah failed to induce oogenesis in the foxl3^Δ7/Δ7 XX and XY mutants, but successfully induced oogenesis in the foxl3^+/Δ7 XY fish (Fig. 6I-L; Fig. S12K-R). This indicates that foxl3 is important for female germ cell fate decision. Real-time PCR analysis showed that the expression of dmrt1 in the ovary of foxl3^Δ7/Δ7 XX fish treated with E2 was upregulated rather than downregulated, which was significantly different from that of WT XY fish treated with E2. In the latter, the expression of dmrt1 was significantly downregulated, whereas the expression of foxl3 was significantly upregulated to the same level as WT XX fish (Fig. 6M). In contrast, foxl3 expression remained high in the AI-treated ovaries of dmrt1^Δ5/Δ5 XY and XX fish, which was different from that of the AI-treated WT XX fish, in which foxl3 was significantly downregulated and dmrt1 was significantly upregulated (Fig. 6N). These data further demonstrate the mutual repression between dmrt1 and foxl3.

Subsequently, we carried out AI treatment to masculinize the somatic cells of the foxl3^Δ7/Δ7 XX mutants from 5 to 30 dah. AI-treated foxl3^Δ7/Δ7 XX fish developed testes filled with spermatogonia and spermatocytes like the WT XY fish at 60 dah. Cyp19a1a was not expressed, whereas Amh and Gsdf were abundantly expressed in the gonads of AI-treated foxl3^Δ7/Δ7 XX fish, as seen in the WT XY fish (Fig. 7A-Db; Fig. S12W-Z). Consistently, knockdown of foxl2 by TALENs in foxl3^Δ7/Δ7 XX fish resulted in testis development, with spermatogonia and spermatocytes at 60 dah, as demonstrated by masculinized somatic cells with increased expression of Gsdf and the absence of Cyp19a1a (Fig. 7E-Gb; Fig. S13). Knockdown of foxl2 in WT XX fish also resulted in complete sex reversal (Fig. 7I), as reported by our previous study (Li et al., 2013). Therefore, disruption of estrogen synthesis induced testicular development in foxl3^Δ7/Δ7 XX tilapia, suggesting that foxl3 functions downstream of foxl2/estrogen.

We injected foxl2 TALENs mRNA into the dmrt1^Δ5/Δ5 embryos to knock down its expression (Table S3). At 60 dah, the dmrt1/foxl2 double mutant XX and XY fish developed testes, as seen in the WT XY fish. Knockdown of foxl2 in dmrt1^Δ5/Δ5 XX and XY fish resulted in masculinization of somatic cells, as demonstrated by the high expression of somatic cell marker Amh (Fig. 7Hd-Kd) and absence of the female markers 42sp50 and Cyp19a1a (Fig. 7Ha-Ka, Fig. 7Hc-Kc). However, in the dmrt1/foxl2 double mutant testes the germ cell number was significantly reduced compared with that of the foxl2 knockdown XX testes, as shown by decreased Vasa signals (Fig. 7Hb-Kb). Unlike the foxl2 knockdown XX testes, germ cells in the double mutant testes failed to enter spermatogenesis, indicating that Dmrt1 is essential not only for male sex determination but also for spermatogenesis.

The germ cell is the only cell type that can transmit the genetic information to the next generation, and therefore the germ cell fate decision attracts the attention of reproductive biologists (Nishimura and Tanaka, 2016; Capel, 2017). A striking finding of our study is that mutation of dmrt1 rescued the germ cell sex reversal in foxl3^Δ7/Δ7 XX mutants, and mutation of foxl3 partially rescued the germ cell sex reversal in dmrt1^Δ5/Δ5 XY mutants. A series of in vivo and in vitro experiments revealed an antagonistic role of Dmrt1 and Foxl3 in the germline sexual fate decision. Therefore, we have provided an answer to the question of why spermatogenesis can occur in female environment in the absence of Foxl3. Our data highlight the role of Foxl3 and Foxl2 in germ cell and somatic cell fate decision, as they were expressed in germ cell and somatic cell, respectively. Likewise, Dmrt1 is involved in both germ cell and somatic cell fate decision as it was expressed in both cell types.

It is well accepted that antagonistic actions of female and male pathway genes contributes to gonadal cell fate decision in vertebrates (Capel, 2017; Li et al., 2013). In line with this, Foxl2 and Dmrt1 play antagonistic roles in both the mammalian gonadal somatic cell fate decision and in fish sex determination (Uhlenhaut et al., 2009; Matson et al., 2011; Lindeman et al., 2015; Li et al., 2013). In XX tilapia, we found that mutation of foxl3 also induced spermatogenesis, as reported in medaka (Nishimura et al., 2015). Expression of dmrt1 was highly induced in germ cells but not in somatic cells of foxl3^Δ7/Δ7 XX ovaries at an early stage of development, raising the possibility that dmrt1 might be a repressive target of Foxl3. Evidence from our genetic studies supports the hypothesis that activation of dmrt1 in germ cells of foxl3^Δ7/Δ7 XX fish is responsible for inducing spermatogenesis because disruption of dmrt1 prevented the germ cell sex reversal. Therefore, future studies need to investigate whether conditional overexpression of dmrt1 in germ cells of XX fish could induce spermatogenesis within a feminized gonadal environment. A conditional knockout of dmrt1 in XY fish germ cells could clarify the role of Dmrt1 as a germline male fate determiner. A study in mammals has shown that Foxl2 directly antagonizes the transcriptional activity of the dmrt1 promoter (Lei et al., 2009). Both dmrt1 and foxl3 were found to be expressed in germ cells, and Foxl3, a paralog of Foxl2 (Bertho et al., 2016), repressed dmrt1 transcription, thus identifying dmrt1 as a target of foxl3. Therefore, significantly increased dmrt1 expression in germ cells caused by loss of foxl3 antagonized the feminization of estrogen and contributed to the male germ cell fate and spermatogenesis in the ovarian environment.

Although our previous study demonstrated that Dmrt1 directly represses cyp19a1a expression in somatic cells (Wang et al., 2010), the target of Dmrt1 in fish germ cells remains unclear. Expression of foxl3 was found to be significantly upregulated and contributed to female germ cell fate in the dmrt1^Δ5/Δ5 XY tilapia. Indeed, this is the main reason for, rather than the consequence of sex reversal, because mutation of foxl3 in dmrt1^Δ5/Δ5 XY tilapia partially rescued germ cell sex reversal, as demonstrated by the spermatogenesis observed. There might be another factor that induces oocyte formation in the foxl3^Δ7/Δ7;dmrt1^Δ5/Δ5 double mutant XY tilapia. Double knockout of foxl3 together with this unknown factor might be able to completely rescue the germ cell sex reversal in dmrt1^Δ5/Δ5 XY fish. In addition, in vitro studies showed that Dmrt1 directly repressed the expression of foxl3. In vivo evidence also supported the mutual repression between Dmrt1 and Foxl3 in germline sexual fate decision as well. For example, dmrt1 was significantly upregulated in foxl3^Δ7/Δ7 XX mutants, but significantly downregulated in the WT XY fish after E2 treatment. foxl3 expression in the WT XY fish was significantly upregulated after E2 treatment. These data indicate that repression of dmrt1 expression in germ cells by E2 is dependent on the presence of Foxl3. Notably, the foxl3 gene has been identified in fishes, reptiles, birds and marsupials in which sex determination is tightly regulated by E2 (Crews et al., 1996; Guiguen et al., 2010; Ayers et al., 2013; Chew and Renfree, 2016). On the other hand, inhibition of estrogen synthesis by AI treatment downregulated foxl3 expression in the WT XX, but not in the dmrt1^Δ5/Δ5 XX and XY mutants. This result indicates that repression of foxl3 is dependent on the presence of Dmrt1. Taken together, we concluded that Dmrt1 and Foxl3 play an antagonistic role in the germline sexual fate decision in fish.

It has been reported in several fish species that loss of dmrt1 resulted in upregulation of aromatase expression and male-to-female sex reversal (Masuyama et al., 2012; Lin et al., 2017; Webster et al., 2017), whereas inhibition of estrogen synthesis by AI treatment, or mutation of cyp19a1a, resulted in upregulation of dmrt1 expression and female-to-male sex reversal (Sun et al., 2014a; Lau et al., 2016; Zhang et al., 2017; Tang et al., 2017; Nakamoto et al., 2018). However, AI treatment failed to rescue the male-to-female sex reversal in dmrt1^Δ5/Δ5 XY tilapia. Consistently, in zebrafish, mutation of cyp19a1a failed to rescue the male-to-female sex reversal in the early stages of dmrt1 mutants (Wu et al., 2020; Romano et al., 2020). These results further confirmed that Dmrt1 is essential for male germ cell fate decision. Female-to-male sex reversal induced by AI treatment relies on upregulation of dmrt1 expression.

It is well accepted that germ cell fate is determined by the environment, that is the surrounding somatic cells in vertebrates (Bowles and Koopman, 2010; Spiller and Bowles, 2015). This dogma is supported by the following evidence in fish. First, treatment with exogenous hormone or its inhibitor can successfully induce sex reversal in both sexes if applied during the critical period for sex determination and differentiation (Guiguen et al., 2010). Second, in males, mutation of the sex determining genes expressed in somatic cells, such as amhy and gsdf, resulted in complete sex reversal of both somatic and germ cells. Similarly, in females, mutation of female pathway genes expressed in somatic cells, such as foxl2 and cyp19a1a, resulted in complete sex reversal of both somatic and germ cells (Li et al., 2013; Lau et al., 2016; Tang et al., 2017; Zhang et al., 2017; Yang et al., 2017; Nakamoto et al., 2018), irrespective of the presence or absence of Foxl3, the female germ cell fate determiner. Third, transplantation of rainbow trout spermatogonia to germ cell-free triploid salmon produced sperm in male, but produced eggs in female (Okutsu et al., 2007). However, the finding obtained from foxl3 mutants in medaka and tilapia seem to challenge this dogma, as spermatogenesis occurred in a female environment, suggesting that germ cell fate is determined cell autonomously instead of by its surrounding somatic cell environment. Further analysis of the tilapia mutants overturned this conclusion, as inhibition of estrogen synthesis or knockdown of foxl2 in foxl3^Δ7/Δ7 XX mutants activated male pathway gene expression in somatic cells and resulted in complete sex reversal of both germ cells and somatic cells. This is different from the results in medaka that inhibition of estrogen synthesis did not block oocyte formation in foxl3 XX mutants (Nishimura et al., 2015). The results in tilapia suggested that Foxl3 acts downstream of Foxl2 and estrogen, both of which repress dmrt1 expression in somatic cells of the ovary. This is consistent with the results reported in a number of fish species that disruption of foxl2/cyp19a1a activated Dmrt1 expression and resulted in complete sex reversal (Lau et al., 2016; Yang et al., 2017; Zhang et al., 2017; Tang et al., 2017; Nakamoto et al., 2018). On the other hand, the somatic cells in dmrt1^Δ5/Δ5 or dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XY mutants were feminized, which could be explained by the induction of foxl2 and cyp19a1a expression in somatic cells. Inhibition of estrogen synthesis in somatic cells of dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 XY and XX mutants, or disruption of the upstream gene foxl2 in dmrt1^Δ5/Δ5 XY mutants, induced testicular development, indicating that foxl2 is epistatic to dmrt1. Therefore, Dmrt1 functions to repress foxl2 and estrogen synthesis in somatic cells and to repress foxl3 in germ cells to determine germline sexual fate. Our study clearly demonstrates a gonadal sub-functionalization scenario for the two paralogous genes foxl2 and foxl3 in vertebrate sex determination as previously predicted (Bertho et al., 2016). These data support the accepted dogma that sexual fate of germ cells is determined by factors derived from somatic cells.

In the normal condition, Dmrt1 functions as a switch between the male and female gonadal cell fate decision (Herpin and Schartl, 2011; Matson and Zarkower, 2012). In the context of foxl2 loss, Dmrt1 is dispensable to induce female-to-male sex reversal in tilapia. As discussed above, the presence of Dmrt1 is the prerequisite for occurrence of spermatogenesis in a female environment after loss of foxl3. A recent study in zebrafish showed that in the context of dmrt1 loss, aromatase/Cyp19a1a is dispensable for early oocyte development (Wu et al., 2020; Romano et al., 2020). Different from the tilapia dmrt1^Δ5/Δ5 XY and foxl3^Δ7/Δ7 XX single mutants, in which the sex of germ cells was not rescued by AI and estrogen treatment, respectively, the germline sexual fate was dependent on the somatic cell environment in the dmrt1 and foxl3 double mutants. Here, we conclude that the role of a specific gene in gonadal sex fate decision is dependent on its genetic background. Mutation of either dmrt1 in XY or foxl3 in XX fish resulted in loss of sexual plasticity of germ cells. In other words, sexual plasticity of germ cells relies on the presence of dmrt1 and foxl3 in fish. The existence and normal expression of dmrt1 and foxl3 in vertebrates avoids an autonomous germ cell sexual fate decision, allowing environmental factors to influence the sex fate decision.

In conclusion, in XY males, Dmrt1 in germ cells represses foxl3 transcription, and Dmrt1 in somatic cells represses foxl2 and cyp19a1a transcription, to ensure testicular development and spermatogenesis. In XX females, Foxl2-activated cyp19a1a expression produces estrogen which is required to repress dmrt1 expression in somatic cells. High expression of Foxl3 in germ cells represses dmrt1 expression to ensure ovarian development and oogenesis (Fig. 7L; Fig. S14). Our work highlights the role of foxl3 and foxl2, two paralogs, in germ cell and somatic cell fate decision as they were expressed in germ cell and somatic cell, respectively, and dmrt1 in both germ cell and somatic cell fate decision as it was expressed in both germ cell and somatic cell.

Nile tilapia was kept in recirculating freshwater tanks at 26°C before use. All-XX progenies were obtained by crossing a pseudomale (XX male, producing sperm after hormonal sex reversal) with a normal XX female. All-XY progenies were obtained by crossing a YY supermale with a normal XX female. In our experimental conditions, these all-XX and all-XY fish develop as all-female and all-male populations, respectively. All animal experiments were conducted in accordance with the regulations of the Guide for Care and Use of Laboratory Animals and were approved by the Committee of Laboratory Animal Experimentation at Southwest University, China.

Tilapia foxl3 was isolated from tilapia genome data (http://asia.ensembl.org/Oreochromis_niloticus/Info/Index). For more information about the foxl3 sequence, see the Multiple sequence alignment and phylogenetic analysis section of the supplementary Materials and Methods. The tilapia dmrt1 and foxl3 mutant lines were generated by CRISPR/Cas9 as previously described (Li et al., 2014). Briefly, the gRNA target containing a restriction enzyme site was selected by identifying sequences corresponding to GGN₁₈NGG on the sense or antisense strand of dmrt1 and foxl3 using the online tool ZIFIT Targeter (http://zifit.partners.org/zifit/Introduction.aspx). The gRNA and Cas9 mRNA components were produced as previously described (Li et al., 2014). Artificially synthesized gRNA and Cas9 mRNA were co-injected into one-cell stage embryos at a final concentration of 250 ng/μl and 500 ng/μl, respectively. The genomic DNA (gDNA) was extracted from pooled control and injected embryos at 72 h after injection using PureLink™ Genomic DNA Mini Kit (Invitrogen). DNA fragments spanning the target site were amplified using the primers listed in Table S4. The mutations were analyzed by restriction enzyme digestion and Sanger sequencing. F0 fish were screened by restriction enzyme digestion. Heterozygous F1 offspring were obtained by mating F0 XY male founders with WT XX females. The F1 fish were genotyped by the fin-clip assay. XY male and XX female siblings of the F1 generation that carried the same mutation were mated to generate homozygous F2 mutants. The genetic sex of the mutants was identified using tilapia sex specific marker-5, as described in the Sex genotyping section of the supplementary Materials and Methods (Sun et al., 2014b). For generation of dmrt1 and foxl3 double mutants, dmrt1^+/Δ5;foxl3^+/Δ7 double heterozygous mutants were obtained by mating dmrt1^+/Δ5 XY males with foxl3^+/Δ7 XX females. The dmrt1^Δ5/Δ5;foxl3^Δ7/Δ7 double mutants were further obtained by mating dmrt1^+/Δ5;foxl3^+/Δ7 XY males with dmrt1^+/Δ5;foxl3^+/Δ7 XX females.

Our previous study showed that TALEN-mediated knockdown of foxl2 in F0 XX fish induced female-to-male sex reversal (Li et al., 2013). In this study, in vitro synthesis of mRNA for foxl2 TALENs was carried out using a Sp6 mMESSAGE mMACHINE Kit (Ambion). About 200-600 pg mRNA was injected into progeny of foxl3^+/Δ7 XX females crossed with foxl3^Δ7/Δ7 XY males or dmrt1^+/Δ5 XX females crossed with dmrt1^+/Δ5 XY males at the one-cell stage. The injected embryos were hatched at 26°C and collected at 60 dah to extract gDNA for mutation analysis.

In the treatment group, the newly hatched fry were fed with commercial diet sprayed with 95% ethanol containing fadrozole (200 µg/g feed) or 17β-estradiol (E2) (200 µg/g feed) (Sigma-Aldrich) from 5 to 30 dah. Later on, all fish were fed with normal commercial diet. The control group fish were fed with normal commercial diet sprayed with 95% ethanol only. The treatments were repeated three times. The gonad phenotype (by histological examination) and the gene expression of gonads were determined at 60 dah.

The gonad samples were fixed in Bouin's solution for 24 h at room temperature. They were then dehydrated and embedded in paraffin. Tissue blocks were sectioned at 5 μm and stained with Hematoxylin and Eosin (H&E) as previously described (Wang et al., 2007).

The gonads of WT and mutant fish were sampled at the time indicated. The samples were fixed in 4% paraformaldehyde (Sigma-Aldrich) in 0.85× PBS at 4°C overnight. Specimens were embedded in paraffin and sectioned at 5 µm. FISH was performed to examine the gene expression as previously described (Li et al., 2020). Fragments of partial foxl3, dmrt1 and bmp15 open reading frames (ORF) were amplified and cloned into the pGEM-T easy vector via TA cloning. Probes for foxl3, dmrt1 and bmp15 antisense or sense digoxigenin-labeled RNA strands were transcribed in vitro from a linearized pGEM-T easy-foxl3, dmrt1 and bmp15 plasmid using the T7 RNA labeling kit (Roche). For more sensitive FISH detection, the tyramide signal amplification (TSA^TM) Plus Fluorescence Systems (PerkinElmer Life Science) were used according to the manufacturer's instructions. In addition, hybridization signal was also detected using alkaline phosphatase-conjugated anti-DIG antibody, and NBT/BCIP as the chromogen as previously described (Wang et al., 2007). Primers for preparing DIG-labeled probes are listed in Table S4.

For IF, the gonads were fixed in Bouin's solution at room temperature overnight with gentle shaking. Specimens were embedded in paraffin and sectioned at 5 µm. The samples were permeabilized with 1% Triton X-100 in PBS for 10 min and then blocked in 5% bovine serum albumin (BSA)/PBS for 30 min at room temperature. The sections were then incubated with polyclonal antibodies in 5% BSA/PBS overnight at 4°C. The following rabbit polyclonal antibodies were prepared by our laboratory: Amh (1:500), Gsdf (1:1000), Sox30 (1:1000), Vasa (1:1000), 42sp50 (1:500), Cyp19a1a (1:2000), Creb1b (1:1000) and Cyp11c1 (1:1000) (Table S5). The dilution and specificity of these antibodies have been analyzed previously (Jiang et al., 2016; Li et al., 2020; Zheng et al., 2020; Wang et al., 2019). Goat anti-rabbit antibody Alexa Fluor 488- and 594-conjugated secondary antibodies (Thermo Fisher Scientific) were diluted to 1:500 in blocking solution and incubated with tissue sections overnight at 4°C to detect the primary antibodies. The nuclei were stained by 4′,6-diamidino-2-phenylindole (DAPI; 1:1000; Sigma-Aldrich) staining. At least three independent biological replicates for each sex and genotype were analyzed in FISH and IF experiments. Fluorescence signals were captured by confocal microscopy (Olympus FV3000).

Total RNAs were extracted from all samples using a column-based RNA extraction kit (Qiagen) specialized for small quantities of RNA. DNase I (RNase free) treatment and cDNA preparation were carried out according to the manufacturer's instructions. For single exon genes such as foxl3, total RNA was directly used as template for PCR as negative control to exclude the genomic DNA contamination. RT-PCR was carried out to check the gene expression according to methods previously described (Wang et al., 2005). For more details about foxl3 mRNA tissue distribution by RT-PCR, please see the Tissue distribution section of the supplementary Materials and Methods. A 342 bp fragment of β-actin was amplified (as internal control) to test the quality of the cDNAs used in the PCR. The PCR conditions were as follows: after an initial denaturation at 95°C for 3 min, a 35-cycle reaction was carried out at 95°C for 30 s, 60°C for 30 s and 72°C for 30 s. This was followed by a final step at 72°C for 10 min. The PCR products were subjected to agarose gel (1.5%) electrophoresis. Real-time PCR was performed with Fast SYBR Green Master Mix (Takara) on a 7500 Fast real-Time PCR system (Applied Biosystems). β-actin was used as the internal control. The relative abundance of target gene mRNA transcripts was evaluated using the formula: R=2^−ΔΔCt as previously described (Livak and Schmittgen, 2001). At least three samples for each genotype were analyzed. Primer sequences used for real-time PCR are listed in Table S4.

Four pairs of RNA preparations from gonads of XX and XY tilapia at 5, 30, 90 and 180 dah were sequenced using Illumina 2000 HiSeq technology as in our previous study (Tao et al., 2013). A normalized measure of reads per kb per million reads (RPKM) was used to normalize the expression profile of foxl3. Clean reads from each library were aligned to the reference genome [O_niloticus_UMD_NMBU (GCA_001858045.3), http://asia.ensembl.org/Oreochromis_niloticus/Info/Index] using Tophat with the default parameters. The RPKM method was used to calculate the gene expression level. The assembled transcripts were merged using the reference annotation (Oreochromis niloticus: Orenil1.0.78.gtf, downloaded from Ensembl) with Cuffmerge. Short-read data of the eight sequenced transcriptomes were deposited in NCBI’s Short Read Archive under the accession number SRA055700.

Western blot was performed following the protocol previously reported (Zhang et al., 2017). Total protein was extracted from mutants and WT XX and XY gonads at the time indicated. The protein lysates were resolved by SDS/PAGE on 12% Tris·glycine gels followed by transfer to nitrocellulose membranes. Unspecific binding was blocked with 5% BSA in Tris buffered saline with Tween-20 (TBST) [10 mM Tris (pH 7.9); 150 mM NaCl; 0.1% Tween] for 1 h at room temperature. Incubation with Dmrt1, HA and Flag primary antibodies at a dilution of 1:1000 (Table S5) was performed overnight at 4°C. After washing with TBST three times, the membranes were incubated with HRP-conjugated secondary antibody (Thermo Fisher Scientific, 1:1000) in blocking solution for 1 h. The abundance of α-Tubulin was examined as a loading control using rabbit anti-α-Tubulin (Cell Signaling Technology) at a dilution of 1:1000. Signal was detected with Pierce™ ECL Western Blotting Substrate (Thermo Fisher Scientific) and was visualized on a Fusion FX7 (Vilber Lourmat).

Tilapia dmrt1 promoter (−2523, −2000, −1500 bp from ATG) and foxl3 promoter (−1450, −700, −600 bp from ATG) were amplified from the gDNA by PCR and then subcloned into the pGL3-basic vector (Promega). Tilapia dmrt1 and foxl3 ORF were subcloned into the pcDNA3.1 vector (Invitrogen). The dmrt1 mutant without DM domain (DM6 in pcDNA3.1) was prepared as described previously (Wang et al., 2010). Potential binding sites for transcription factors within the dmrt1 and foxl3 promoters were predicted using the MatInspector program (http://www.genomatix.de). Moreover, site-directed mutagenesis of the dmrt1 and foxl3 promoters was performed using a MutanBEST Kit (Takara). Plasmids were purified using QIAfilter Plasmid Midi Kit (Qiagen). All primer sequences used for constructing recombinant vectors are listed in Table S4. Cell culture, transient transfections and luciferase assays were performed as previously reported (Wang et al., 2007). Briefly, HEK293 cells (ATCC number: CRL-1573) were maintained in DMEM supplemented with 10% fetal bovine serum (FBS), penicillin (100 IU/ml) and streptomycin (100 μg/ml) at 37°C in a humidified atmosphere containing 5% CO₂. All cells were checked for microbial contamination before use. pRL-TK was adopted to monitor the transfection efficiency and as the internal control. Firefly luciferase and Renilla luciferase readings were obtained using the Dual-Luciferase Reporter Assay System (Promega) and Luminoskan™ ascent luminometer (Thermo Fisher Scientific). Relative luciferase activity was calculated by dividing the firefly luciferase activity with the Renilla luciferase activity.

EMSA experiments were performed as previously described (Wang et al., 2007). Briefly, biotin-labeled and unlabeled double-stranded DNA probes containing intact potential binding sites for Dmrt1 and Foxl3 were prepared according to the manufacturer's protocol for the Chemiluminescent EMSA Kit (Beyotime). The mutated probes of Dmrt1 and Foxl3 binding sites were also prepared. The related primers are listed in Table S4. Nuclear proteins were extracted from HEK293 cells overexpressing the Nile tilapia dmrt1 or foxl3 gene with NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Fisher Scientific). Then, 3-5 µg nuclear protein was incubated with 500 nM biotin-labeled probes in 1× binding buffer for 10 min at room temperature. For the competition analysis, a 10, 50, 100-fold molar excess of the unlabeled or mutated probe was added to the nuclear extracts before the addition of the labeled probe. Protein-DNA complexes were separated on 4% polyacrylamide gels in 0.5× TBE buffer via electrophoresis at 100 V for 90 min and transferred to nylon membranes. The biotin-labeled DNA on the membrane was detected using the Chemiluminescent EMSA Kit (Beyotime).

The Nile tilapia stem Leydig cells (Huang et al., 2020) were maintained in conditioned complete medium, which is DMEM medium supplemented with 20 mM Hepes, penicillin (100 IU//mL) and streptomycin (100 μg/ml), 15% FBS, 2 mM L-glutamine, 1 mM Napyruvate, 2 µM Na-selenite, 1 mM non-essential amino acids, 55 µM β-mercaptoethanol, 10 ng/ml human recombinant basic fibroblast growth factor, 10 ng/ml Nile tilapia leukemia inhibitory factor, Nile tilapia embryo extract (0.4 embryo/ml) and 0.2% fish serum (FS), incubated at 28°C in a humidified atmosphere without CO₂. The stem Leydig cells were cultured for ChIP analysis, with the cells being checked for microbial contamination before use. Simple ChIP Plus Enzymatic Chromatin IP Kit (Cell Signaling Technology) was used for ChIP assay according to the manufacturer's instructions. Briefly, cells overexpressing Flag-tagged Dmrt1 or HA-tagged Foxl3 were fixed with 37% formaldehyde to cross-link chromatin, and then sonicated to shear into DNA fragments of 200-1000 bp in length. Complexes were immunoprecipitated and separately enriched with 1 μg antibody against IgG, Flag, HA and Dmrt1 as previously described (Wei et al., 2019). The purified DNA from the immunoprecipitated chromatin was used for PCR analysis. The PCR products were electrophoresed in 1.5% agarose gels. The primers amplifying the specific region covering the potential binding site for Dmrt1 and Foxl3 are listed in Table S4.

Blood samples were collected from the caudal veins of WT XX (n=6), foxl3^−/− XX (n=6), dmrt1^−/−;foxl3^−/− XX or XY (n=3) and WT XY fish (n=6), and kept at 4°C overnight. Serum was collected after centrifugation and stored at 0°C until use. Serum 11-KT and E2 levels were measured using the enzyme-linked immunosorbent assay (EIA) Kit (Cayman) according to the manufacturer's instructions. Absorbance was measured at a wavelength of 412 nm using a Multiskan™GO microplate reader (Thermo Fisher Scientific).

Data are expressed as mean±s.d. of at least three independent biological replicates. The statistical significance of differences between data means was determined using a unpaired two-tailed Student's t-test or one-way ANOVA, followed by Tukey test for multiple comparisons using GraphPad Prism 5. P<0.05 was statistically significant (*P<0.05, **P<0.01 and ***P<0.001).

We thank Prof. Thomas D Kocher (Maryland University, USA) for his critical editing of the manuscript. We thank Prof. Jing Wei (Southwest University, People's Republic of China) for supplying the Nile tilapia stem Leydig cell line. We are grateful to Deqiang Wang and Pingyuan Luo for fish maintenance.