The transcription factor SOX30 is a key regulator of mouse spermiogenesis

Spermatogenesis is a life-long cyclic process by which sperm are generated from spermatogonial stem cells (SSCs) via meiosis of spermatocytes (SCs). The postmeiotic development of spermatids (STs), also known as spermiogenesis, is unique in many aspects compared with other cellular developmental processes (Sassone-Corsi, 2002). For example, chromatin undergoes global remodeling via the concerted action of histone variants, transition proteins and many other epigenetic modifiers (Bao and Bedford, 2016). As a result, the majority of histones are replaced by protamines and a highly condensed nucleus of the sperm is formed. Moreover, the majority of the cytoplasm containing various common cellular organelles is discarded, and specialized structures such as the flagellum and the acrosome, which are important for fertilization, are formed. It is therefore not surprising that a large number of testis-specific genes are expressed during and/or after meiosis. Several studies have shown that about 4-8% of mRNAs and 30% of long non-coding RNAs (lncRNAs) are specifically expressed in SCs and/or STs (Cabili et al., 2011; Lin et al., 2016; Schultz et al., 2003; Soumillon et al., 2013).

An intriguing question remains regarding how the unique gene expression in spermatogenesis is regulated. Gene regulation in germ cells, as in somatic cells, occurs at transcription and post-transcriptional levels. As transcription in STs and sperm with condensed nuclei are silenced, the transcription of genes involved in fertilization must be finished in round STs (rSTs) that have not started the histone replacement process. However, the translation of these transcripts has to be inhibited, otherwise the normal processes occurring in rSTs can be disrupted by these proteins. Specific examples of such regulatory mechanisms have been described and several global patterns of gene regulation have been identified by integrated analyses of proteomic and transcriptomic data (Gan et al., 2013a).

Several key regulators of gene expression in spermatogenesis, particularly in spermiogenesis, have been identified. For example, CREM-τ is a transcription activator specifically expressed in rSTs due to alternative splicing of the Crem transcripts in pachytene SCs (pacSCs) coupled with delayed translation in rSTs (Delmas et al., 1993; Foulkes et al., 1992). Crem gene knockout (KO) mice arrest spermatogenesis at the early rST stage (Blendy et al., 1996; Nantel et al., 1996). CREM-τ regulates the expression of a large number of genes involved in diverse biological processes either directly or indirectly (Kosir et al., 2012). Owing to these properties, CREM-τ has been regarded as a key regulator of spermiogenesis. Similar regulators include transcription factors (TFs) such as TRF2, TAF7L, RFX2 (Kistler et al., 2015; Martianov et al., 2001; Shawlot et al., 2015; Wu et al., 2016; Zhang et al., 2001; Zhou et al., 2013) and RNA-binding proteins (RBPs) such as TPAP, MIWI, RNF17, BOULE/BOLL, GRTH/DDX25 (Deng and Lin, 2002; Kashiwabara et al., 2002; Pan et al., 2005; Tsai-Morris et al., 2004; VanGompel and Xu, 2010; Wasik et al., 2015). Generally, the transcription of these key regulators starts as early as in SCs under the control of other key TFs such as MYBL1 (Bolcun-Filas et al., 2011; Li et al., 2013).

Several proteins belonging to the SOX family of TFs play important roles in spermatogenesis (Denny et al., 1992; Osaki et al., 1999). SRY, the founding member of this family, is expressed in Sertoli but not germ cells and is believed to be the master TF of downstream genes involved in testis formation (Gubbay et al., 1990; Koopman et al., 1991; Sinclair et al., 1990). Sox9, a direct target of SRY, is sufficient for driving testis development in the absence of Sry (Qin and Bishop, 2005; Sekido and Lovell-Badge, 2008). Upon Sox9 activation, Sox8 and Sox10, the other two SOX E group members, are also induced, and either gene can replace Sox9 for its function in testis formation and development (Lavery et al., 2012; Polanco et al., 2010). Although these SOX proteins execute functions in somatic cells, other members expressed and functioning in germ cells have also been identified. For example, SOX3 is expressed in undifferentiated spermatogonia (SG), and is required for their entry into a more differentiated state (Raverot et al., 2005). Interestingly, the infertile phenotype of Sox3 knockout (KO) mice can be rescued by the ectopic expression of SOX2, another member of the SOX A subfamily (Adikusuma et al., 2017). SOX17, the primary role of which is in endoderm induction but not in the specification of primordial germ cells (PGCs) in mouse, has been shown to be crucial in the in vitro induction of human PGCs from pluripotent stem cells (Irie et al., 2015; Viotti et al., 2014).

To date, 20 SOX genes have been discovered in mice and humans (Schepers et al., 2002). SOX30, which is the most divergent member of this family, has been reported to be specifically expressed in the testis and to act as a sequence specific TF (Osaki et al., 1999). SOX30 is highly conserved during evolution, and the protein sequence identity between mice and humans is 76% (Han et al., 2010). We have previously shown that Sox30 transcripts are upregulated in cultured SSCs, which are also known as germline stem cells (GSCs), by retinoic acids (RA) (Wang et al., 2016). In the present study, we have confirmed this observation and show that Sox30 is a direct target of MYBL1. We also find that SOX30 is preferentially expressed in rSTs and that the spermatogenesis of Sox30 KO mice arrests at step 3 rSTs with abnormal chromocenters. Moreover, we identify a large number of genes regulated by SOX30 by analyzing differentially expressed genes (DEGs) between wild-type and KO mice. Based on these results, we propose SOX30 as a novel key regulator of transcription in mouse spermiogenesis.

We have previously reported that mRNAs of Sox30 are upregulated in response to RA treatment but that the gene is not involved in meiosis initiation (Wang et al., 2016). Here, we confirm the RA-induction of Sox30 using different GSC samples treated with 100 nM RA for 24 h (Fig. 1A). The RA-mediated induction of Stra8, a gene essential for meiosis initiation, was used as a positive control (Griswold, 2016). The addition of RA receptor inhibitor BMS493 simultaneously with RA treatment significantly reduced the fold-change values of both Sox30 and Stra8, confirming the specific action of RA.

Sox30 was expressed only in the testis among several organs tested, as shown by both RT-PCR and western blotting assays (Fig. 1B,C). We also conducted real-time RT-PCR assays to examine the expression of Sox30 in fetal ovaries of E13.5 and E14.5 mice when meiosis is initiated and compared this with its expression in fetal and adult testis. We found that Sox30 expression levels in fetal and adult ovaries were only 10% of that in adult testis, although they were about two- to fourfold higher than in fetal testis (Fig. 1D). We then examined its expression in GSCs, various isolated male germ cells and Sertoli cells, and found that Sox30 was expressed only in germ cells (Fig. 1E). Its expression started to increase sharply from preleptotene SCs (plpSCs), further increasing in pachytene SCs (pacSCs) and maintained at high levels in rSTs. Based on the expression comparison among these cell types, Sox30 mRNA is not normally expressed in mitotic germ cells, including GSCs and SG-A. The twofold upregulation by RA in GSCs may only be an indirect effect of spermatogonial differentiation induced by RA.

As the antibody used in western blotting did not work for immunostaining of testis sections, we knocked one copy of the HA tag-coding sequence into the N-terminal-coding regions of the Sox30 gene using the CRISPR-Cas9 gene-editing technique (Fig. 1F). Using western blotting with antibodies against SOX30 and the HA peptide, we validated the transgene expression in the testis as the SOX30 antibody revealed the expression of SOX30 in both wild-type and transgenic mice, while the HA antibody recognized HA-SOX30 only in the transgenic mice (Fig. 1G). We next examined SOX30 expression in the testis in view of seminiferous stages. SOX30 expression was not detected in somatic cells, SG, plpSCs, leptotene SCs (lepSCs) or zygotene SCs (zygSCs), but was detected in pacSCs of stage III-VIII. Its expression increased slightly in diplotene SCs (dipSCs) at stage IX-XI and in metaphase SCs (metSCs) or secondary spermatocytes (SC2) at stage XII. SOX30 was the most abundant in rSTs and early elongating STs (eSTs) until step 11 (Fig. 1H). Based on these data, SOX30, unlike STRA8, was not responsive to the cyclic pulses of RA, which peak at stages VII-VIII (Endo et al., 2015) (Fig. 1I). However, the mRNA level of Sox30 might be regulated by the RA pulse, as suggested by its obvious upregulation in plpSCs that appeared at stage VI-VIII (Fig. 1E). Importantly, the nuclear localization of the SOX30 signal is consistent with its predicated role as a TF.

We were interested in identifying TFs responsible for the increases of Sox30 transcripts that we observed from plpSCs to pacSCs. MYBL1 was most likely one of such TFs, as suggested by its expression in mid/late pacSCs and by the decreased expression of Sox30 in the testis of mice with a mutation in the Mybl1 gene (Bolcun-Filas et al., 2011). Moreover, Li et al. profiled the binding sites of MYBL1 in the genome of mouse testicular cells using chromatin immunoprecipitation sequencing (ChIP-seq) (Li et al., 2013). Using their data, we found that the Sox30 promoter contained a ChIP-seq peak with three predicted binding sites of MYBL1 (Fig. 1J). Horvath et al. showed that Rfx2 was a direct target of MYBL1 and we followed their experimental design for ChIP-PCR assays to test the binding of MYBL1 to its predicated binding sites in the promoters of both Rfx2 and Sox30 (Horvath et al., 2009). MYBL1 indeed bound to these two sites but not to a control region of the Tcrd gene (Fig. 1K). MYBL1 also activated the promoter of Sox30, as shown by results obtaining using the luciferase assay (Fig. 1L).

We generated Sox30 KO mice using CRISPR-Cas9 to study its function in meiosis (Fig. 2A). We used two sgRNAs targeting the first exon of Sox30 and acquired a female founder mouse with a 209 bp deletion that resulted in a truncation of SOX30 at amino acid 255 due to a frameshift. Homozygous F3 mutants (KO) were used for phenotypic evaluation (Fig. 2B). Wild-type, heterozygous and homozygous mice were readily identified using genomic PCRs (Fig. 2C). Deletion of the full-length protein was verified in KO mice by western blotting, although it was unclear whether a truncated form was present as the antibody targeted the missing region of SOX30 (Fig. 2D). No off-target mutations were found among the top 10 loci that could be targeted by each sgRNA (data not shown). KO mice were not different from their wild-type or heterozygous littermates in terms of body size, weight or behavior, regardless of their ages. Although female KO mice were fertile, male mice were sterile with smaller testes and epididymis compared with wild-type littermates (Fig. 2E,F).

Hematoxylin and Eosin staining of the testis sections revealed that spermatogenesis of KO mice was arrested at the rST period and multinuclear syncytia were observed in seminiferous tubules (Fig. 2G). Degenerated rSTs and multinucleated giant cells instead of normal sperm were seen in the lumens of the epididymis in KO mice (Fig. 2G). Elevated numbers of apoptotic cells were also observed in testis and epididymis sections of KO mice, as detected by TUNEL (Fig. 2I,J).

Spermiogenesis can be distinguished into 16 steps based on nuclear and acrosomal morphologies of the STs, which are revealed by the staining patterns of Hematoxylin and periodic acid-Schiffs reagent (PAS) (Ahmed and de Rooij, 2009). Normally, a single large pro-acrosomal granule appears adjacent to the nucleus in step 3 STs and flattens to cap the nuclear surface of step 4 spermatids in wild-type mice. However, in Sox30 KO mice, step 4 STs and their progenies were absent, suggesting spermatogenesis arrested at step 3 STs in these mice (Fig. 3).

Chromocenters in the cell nucleus are clusters of constitutive heterochromatin from different chromosomes (Probst and Almouzni, 2011). Male haploid germ cells from rST to sperm usually contain only a single chromocenter in the central area of the nucleus (Brinkley et al., 1986). Several genes, including Trf2, Hmgb2, Brdt and Seipin (Bscl2), have been reported to be involved in the formation of chromocenters in rSTs, as individual knockout of these genes results in multiple chromocenters (Berkovits and Wolgemuth, 2011; Catena et al., 2006; El Zowalaty et al., 2015; Martianov et al., 2002). Chromocenters can be identified by immunostaining of either HP1α or HP1γ (Guenatri et al., 2004; Martianov et al., 2002). Based on HP1α immunostaining results, we found that the percentage of rSTs containing more than one chromocenter was significantly increased in Sox30 KO mice compared with wild-type littermates (Fig. 4A). The identity of the HP1α-stained chromocenter was confirmed by the co-immunostaining of HP1α with either H3K9me2 or H3K9me3 in single rSTs treated with hypotonic solution (Fig. 4B). We next examined the distribution of HA-SOX30 in the nucleus of rSTs treated in this way and found that punctate signals of HA-SOX30 were diffusely distributed throughout the nucleus, as well as enriched in the chromocenter (Fig. 4C). These results suggest that SOX30 may play a role in regulating the structure and/or activity of heterochromatin.

We next conducted RNA-sequencing (RNA-seq) analysis to identify SOX30-regulated genes that were differentially expressed in rSTs of Sox30 KO and wild-type mice. rSTs were isolated by fluorescence-activated cell sorting (FACS) from 23 days postpartum (dpp) wild-type and KO mice, a stage where spermatogenesis progresses to step 3 rSTs with no extra haploid cell types present (Fig. 5A,B). Differential expression analysis revealed that 664 and 121 genes were down- and upregulated, respectively, when Sox30 was knocked out (Fig. 5C, Table S1). In a previous study, we identified 2745 testis-specific mRNAs (Lin et al., 2016). Interestingly, we found that 64% of the downregulated genes were testis specific, and the enrichment based on Fisher's exact test was extremely significant (P<2.2e-16). These genes represent different protein families, including transmembrane (Aqp7, Catsper3, Tmem144), DNA binding [Hmgb4, H1fnt, Fhl5, 1700024P04Rik (H2bl1)], enzymes [Ccdc54 (Spa17), Ppm1j, Adam29, Dpysl3, 1700009N14Rik, Tssk3, Gm1527] and structural components [Lelp1, Fam154a (Saxo1), Gm15319, Pfn3] [see Fig. 5D for the enriched Gene Ontology (GO) terms]. The downregulation of a group of genes, including Sox30 itself, was confirmed by real-time RT-PCR analyses (Fig. 5E). As a large number of lncRNAs are also expressed in STs, we identified 148 and six lncRNA genes were down- and upregulated, respectively, and the downregulation of some lncRNAs was also confirmed by RT-qPCR (Fig. 5F).

Interestingly, Sox30 KO did not change the expression of most key regulators of spermiogenesis, such as Crem, Trf2, Rfx2, Boule, Miwi, Tpap or Ddx25 (Table S1). We also compared Sox30-regulated genes with those regulated by TRF2, RFX2 and CREM-τ. Genes regulated by TRF2 and RFX2 were obtained from DEGs identified between wild-type and KO mice by the original studies using RNA-seq analyses, whereas CREM-τ-regulated genes were identified using ChIP-seq analysis (Kosir et al., 2012; Martianov et al., 2016; Wu et al., 2016). As shown in Fig. 5G, the numbers of genes co-regulated by these TFs were low, suggesting that these key TFs likely act independently during this stage of spermatogenesis.

We have previously developed a computer program to identify TF-binding sites in evolutionarily conserved regions of proximal promoters from −10 kb to 5 kb of the transcription start sites (TSSs) of protein-coding genes (Lin et al., 2016). A site was regarded as a putative SOX30-binding site if it was more than 80% identical to the SOX30 protein-binding motif represented by a position weight matrix (Mathelier et al., 2016; Osaki et al., 1999). As a result, we identified 1935 putative SOX30 target genes (class 1, Table S1), out of which 60 changed their expression levels upon Sox30 KO (class 2, Table S1). We selected 11 class 2 target genes (Catsperb, Ccdc54, Odf3, Rnf151, Sox5, Spata19, Spz1, Dnaja4, Efcab3, Atp1b3 and Pdilt), one class 1 gene (Sox30) and three non-target genes (Nanog, Gfra1 and Oct4) for ChIP-PCR analysis. Two SOX30-binding sites (Sox30M1 and Sox30M2) were identified on the promoter of Sox30 itself and tested independently. Binding sites of 10 genes were significantly enriched in chromatin samples of HA-Sox30 transgenic mice immunoprecipitated using the HA antibody compared with in wild-type mice (Fig. 6A).

As lncRNA genes are poorly conserved during evolution, we directly scanned for any enriched motif in the promoters from −1 kb to +1 kb around TSSs of the 154 lncRNA genes that were differentially expressed in wild-type and Sox30 KO mice. Interestingly, these genes were significantly enriched with binding sites for SOX and CREB family TFs, and 19 lncRNAs had binding sites for both TF families in their promoters (Table S1). We also used ChIP-qPCR to examine the binding of HA-SOX30 to the predicted sites of four lncRNAs and found two binding sites that were positive (Fig. 6B).

We next conducted luciferase assays to examine the activation of proximal promoters of several target genes, including Sox30 itself, by SOX30 protein. A 472 bp Sox30 promoter fragment containing the two predicted binding sites was placed upstream of the luciferase-coding sequence. A fragment of Rfx2 containing a putative SOX30-binding site that is not in the evolutionarily conserved region was also tested. As shown in Fig. 6C, the promoters of Sox30, Ccdc54 and Spata19, but not that of Rfx2, were activated by SOX30. The low activation fold values obtained by these constructs might be due to the lack of other co-activators normally present in germ cells.

Spermatogenesis is an ideal model for studying how a multistep cellular developmental process is controlled by waves of gene expression that are intricately regulated by extracellular and intrinsic signals. In this sense, we report here that SOX30, a TF that is essential for spermiogenesis, regulates the expression of a large number of genes, while being regulated itself by multiple factors, including its own protein product in a step-wise manner.

Sox30 came to our attention because we, in a previous study, found that its transcripts were upregulated by RA in GSCs but its knockdown did not affect meiosis initiation in vitro. In the present study, we not only confirmed the RA-mediated upregulation of Sox30 expression but also found that another TF, MYBL1, also acts as a regulator of Sox30 transcription. During the preparation of this manuscript, another study showed that Sox30 transcripts are elevated immediately after the peak expression of Stra8, which heralds meiosis initiation, in both fetal ovaries and postnatal testis (Feng et al., 2017). Feng et al. also reported the exact phenotypes of Sox30 KO mice that we have observed here. However, probably owing to the lack of good antibodies, the authors failed to examine the localization of SOX30. As delayed translation in spermatogenesis is a prominent feature of postmeiotic genes, we took efforts to generate transgenic mice that expressed HA-tagged SOX30 proteins and found that SOX30 was mainly present in STs. To our knowledge, Sox30 represents the first example of a gene transcript that is upregulated by RA as early as in SG, while the protein executes its function in STs. More interestingly, we found that Sox30 transcription was also regulated by its own protein product. Such a positive self-feedback loop is frequently observed for key regulators of gene expression.

Although transcription is highly active and the transcriptome is highly complex in STs, only a few key TFs (CREM-τ, TRF2, RFX2) have been identified. To be a key regulator, a TF should both be phenotypically essential and control the expression of a large number of genes either directly or indirectly, which we observed for SOX30. For example, all the mice in which these key TFs had been knocked out arrest spermatogenesis in the early rST period. In particular, spermatogenesis of Sox30 KO mice arrest at step 3, earlier than the other three TFs, Crem-τ (step 5), Trf2 (steps 7-14) and Rfx2 (step 7), and all the key RBP regulators, including the Boule (step 6), Miwi (step 4), Rnf17 (step 4) and Ddx25 (step 8) (Blendy et al., 1996; Deng and Lin, 2002; Kistler et al., 2015; Martianov et al., 2001; Nantel et al., 1996; Pan et al., 2005; Shawlot et al., 2015; Tsai-Morris et al., 2004; VanGompel and Xu, 2010; Wasik et al., 2015; Wu et al., 2016; Zhang et al., 2001).

As shown by results here and from previous studies, these key TFs seem to be relatively independent in governing their target genes as the intersection of their target sets are small and they do not mutually regulate each other (Martianov et al., 2016). Furthermore, these key TFs are probably all controlled by other TFs of a higher rank, such as MYBL1 (Bolcun-Filas et al., 2011; Horvath et al., 2009) (Fig. 6D). Such a tree-like regulatory cascade instead of a network enriched in nodes with dual or more parents is theoretically more efficient in operation. Another prominent feature of these key TFs is that their targets are enriched in testis-/ST-specific genes, to which they themselves also belong. How specific gene expression is realized with short promoters and few regulators in male germ cells is still a fascinating unresolved issue; we believe the regulatory pathways, in which SOX30 is involved, may serve as an ideal model to address and resolve matters.

One interesting observation in Sox30 KO mice is that STs with multiple chromocenters are significantly more frequent than in wild-type mice. As chromocenters are foci of aggregated pericentric heterochromatins and are organizing centers of high order chromatin structures, this phenotype suggests that Sox30 may play a role, either directly or indirectly, in the global chromatin re-modeling that occurs during spermiogenesis. The single ST chromocenter is reported to be formed immediately after the second meiotic division (Hoyer-Fender et al., 2000). Therefore, the regulation of chromocenter formation by SOX30 is most likely a direct role of SOX30 and/or its regulated genes, rather than a reflection of arrested differentiation. The KO of a number of genes (Trf2, Hmgb2, Brdt and Seipin) also results in the same multiple chromocenter phenotype (Berkovits and Wolgemuth, 2011; Catena et al., 2006; El Zowalaty et al., 2015; Martianov et al., 2002). However, we found that none of these genes was regulated by SOX30. Therefore, either SOX30 directly regulates chromocenter formation or other mediators have yet to be identified. Noncoding RNAs have recently been reported to be involved in heterochromatin formation (Nishibuchi and Dejardin, 2017). We also tried to identify genomic features, such as localization to pericentric regions or enrichment of retrotransposons, of the SOX30-regulated genes, including lncRNA genes, but unfortunately no significant candidates with these properties were identified.

One caveat for the above discussions on the actions of SOX30 is that they are based on the reliable identification of SOX30-regulated genes. We observed that haploid cells were under-represented in the KO samples compared with the wild-type samples (Fig. 5B), which was caused by germ cell death in KO testes. We tried to isolate rSTs at an earlier stage (21 dpp), but the numbers of cells were too low for regular RNA-seq analyses to be conducted reliably. Fortunately, the types of haploid germ cells at 23 dpp did not differ between KO and wild-type mice based on histological evaluations, indicating that cell loss at this stage does not result in a cell type bias. More importantly, we used the same numbers of rSTs (1000) from KO and wild-type mice for RNA-seq analyses, from which gene expression was calculated in FPKMs (fragments per kilobase per million mapped reads) that are normalized values to eliminate systematic errors caused by variations in the amounts of isolated RNAs and/or in sequencing depths. Based on these considerations, we think that the effect of cell loss in KO testes on the interpretation of expression results, if any, was small. Moreover, we noticed that Feng et al. used 25 dpp mice to examine genes differentially expressed between KO and wild-type mice, and results similar to ours were derived, supporting the argument that cell number reduction in KO mice does not affect the interpretation of the results as long as all cell types are present and the sample size is big enough (Feng et al., 2017).

Based on RNA-seq results and bioinformatic predications, we selected a panel of putative direct target genes of SOX30 for further validation. Through ChIP-PCRs and luciferase assays, we found that both mRNA and lncRNA genes could be direct targets of SOX30. However, it is still too early to fully understand the mechanism by which SOX30 functions as a key TF in spermiogenesis. We believe that it is important to identify SOX30 target sites in the whole genome, including heterochromatins, using high-throughput methodologies, including ChIP-seq technology to address these issues in future studies.

Mice used in this project were approved by the Animal Ethics Committee of the Institute of Zoology, Chinese Academy of Science. All procedures were conducted by following institutional guidelines.

Mice were generated using CRISPR-Cas9 gene-editing technologies (Zhou et al., 2014). Briefly, Cas9 mRNAs were in vitro transcribed from pST1374-Cas9-N-NLS-flag-linker (Addgene 51130) and sgRNAs from pUC57-sgRNA (Addgene 51132). pUC57-sgRNA constructs were made by inserting double-stranded DNA sequences of sgRNAs annealed from synthesized complimentary oligonucleotides into the vector. sgRNA oligos are listed in Table S2. Cas9 mRNA and sgRNAs were injected into zygotes from mating DBA males with superovulated C57BL/6J females. To generate HA-Sox30 mice, single-strand donor oligonucleotides were also included in the mRNAs of Cas9 and sgRNAs. The founders of Sox30 KO mice were of mixed DBA and C57BL/6J genetic backgrounds. The founders were mated with C57BL/6J mice to generate F1 animals, which were then mated to ICR mice to produce F2 offspring. Heterologous F2 males and females were crossed to give rise to F3 mice for phenotypic analyses. As expected, F3 mice of different coat colors, including black, brown and white, were derived and they had the same phenotypes as described in the Results. This shows that genetic background has no obvious effects on the reported phenotypes. Genotyping of the founder mice was conducted by genomic PCR.

Total RNAs were isolated and reverse transcription was conducted by following standard procedures. Real-time PCRs were performed with UltraSYBR Mixture (CW0956, CoWin Biotech) on a LightCycler 480 platform (Roche Diagnostics). Expression values were calculated using the ΔΔCt method with β-actin as the internal control. Oligonucleotide primer sequences are listed in Table S2.

Cell lysates from mouse organs were prepared by homogenizing small pieces of organs with glass homogenizers in RIPA buffer supplemented with Protease Inhibitor Cocktail (Sigma). Lysates were centrifuged at 20,000 g for 10 min at 4°C, and the supernatants were used for western blotting analyses by following standard procedures. Antibodies and dilutions used are listed in Table S3.

cDNAs in eukaryotic expressing vector for TFs Mybl1 (pCR-BluntII-TOPO vector) and Sox30 (pMD18-T vector) were purchased from Source BioScience and Youbio Biological Technology, respectively. Promoter DNAs of Sox30 and Rfx2 were PCR amplified from mouse genomic DNAs isolated from mouse tail tips and were cloned into PGL4.17-Luciferase vector (Addgene, E6721). TF-expressing plasmids, promoter-luciferase plasmids and the internal control pRL-TK-Renilla constructs were co-transfected into HEK293T cells on 96-well plates using X-Transcell reagent (bjyf-Bio technology) following the manufacturer’ s protocol. Cell extracts were prepared 48 h after transfection using the lysis buffer provided in the Dual-Luciferase Reporter Assay System kit (Promega), and luciferase activity was measured on a Synergy Neo2 Multi-Mode Microplate Reader instrument (Bio-Tek) according to the manufacturer’ s protocol. The Renilla luciferase activity was used to normalize the firefly luciferase activity.

rSTs from adult mice were sorted from total testicular cells stained with Hoechst dye using FACS. After a wash with PBS, 10⁵ cells were treated with 200 µl hypotonic extraction buffer [30 mM Tris, 17 mM trisodium citrate dihydrate, 50 mM sucrose, 5 mM EDTA, 0.5 mM DTT and 0.5 mM PMSF (pH 8.2)] at room temperature. The cells were then collected by centrifugation at 600 g for 5 min and resuspended in 100 mM sucrose (pH 8.2). Subsequently, cell suspensions were placed on clean glass slides pre-treated with 0.15% Triton X-100 and 1% paraformaldehyde (PFA) (pH 9.2). Slides were dried for at least 2 h in a closed box with high humidity. The dried slides were stored at −20°C before immunofluorescent staining (Peters et al., 1997).

Haploid cells were sorted from total testicular cells from 23 dpp mice stained with Hoechst dye using FACS. RNA isolation and sequencing library construction were conducted by using the NEBNext Ultra RNA Library Prep Kit for Illumina. Sequencing was carried out on the Illumina HiSeq 2000 platform (Novogene). Data analysis was performed by procedures that have been previously described (Gan et al., 2013b). All sequence data are available in GEO under the accession number GSE113073. Published sequencing data for genes regulated by Crem (GSE29593), Trf2 (GSE79603) and Rfx2 (GSE74961) were downloaded from GEO database.

Chromatin immunoprecipitation (ChIP) assays were performed according to a protocol by Komata et al. (Komata et al., 2014). Briefly, total cells were isolated from mouse testes by following a two-step digestion procedure previously described (Wang et al., 2016). 1×10⁸ cells were fixed with 10 ml 1% formaldehyde in PBS for 10 min at room temperature. The crosslinking reaction was stopped by adding glycine to a final concentration of 125 mM. The cells were then washed twice with ice-cold PBS and resuspended in 1 ml lysis buffer [20 mM Tris-HCl (pH 7.5), 1 mM EDTA, 10 mM NaCl, 0.2% NP-40, 1 mM PMSF] while kept on ice for 10 min. The cells were collected by centrifuging at 2000 g for 5 min and further treated with 700 µl RIPA buffer containing 1 mM PMSF and protease inhibitor cocktail (Sigma). Cell lysates were sonicated (Cole Parmer, UK) at 25% power for 30 cycles (10 s on, 20 s off for each cycle). This resulted in DNA fragments with a size distribution between 0.2 and 1 kb in length when examined by gel electrophoresis. The sonicated cell lysates were centrifuged at 15,800 g for 10 min at 4°C to remove cell debris. A 400 µl aliquot of the supernatants was stored at −80°C. The remaining 300 µl was incubated with 5 µg of anti-HA antibody (Cell Signaling s3724) at 4°C overnight. Next, 30 µl Protein A magnetic beads (Invitrogen) were added to the samples followed by incubation at 4°C for 2 h and three washes with LiCl wash buffer [100 mM Tris (pH 7.5), 1 M LiCl, 1% NP-40]. Chromatins were eluted from the washed beads with elution buffer (1% SDS, 100 nM NaHCO₃) in a 65°C water bath. The eluates were incubated at 65°C overnight to complete the reversal of the formaldehyde crosslinks. DNA fragments were acquired by ethanol precipitation and resuspended in 30 µl of water for PCRs. Real-time PCRs were carried out in a 20 µl reaction volume using 0.5 µl DNA samples following standard procedures.

We thank Mr Qiong Huo, Ms Liming Lv and Mr Ye Tian (Institute of Zoology, Chinese Academy of Sciences) for their help with transgenic mice production. We thank Dr Kyle Miller at the University of Texas for his help on editing the manuscript.