DDX5 plays essential transcriptional and post-transcriptional roles in the maintenance and function of spermatogonia

Mammalian spermatogenesis is sustained by mitotic germ cells with self-renewal potential known as undifferentiated spermatogonia. Maintenance of undifferentiated spermatogonia and spermatogenesis is dependent on tightly co-ordinated transcriptional and post-transcriptional mechanisms. The RNA helicase DDX5 is expressed by spermatogonia but roles in spermatogenesis are unexplored. Using an inducible knockout mouse model, we characterise an essential role for DDX5 in spermatogonial maintenance and show that Ddx5 is indispensable for male fertility. We demonstrate that DDX5 regulates appropriate splicing of key genes necessary for spermatogenesis. Moreover, DDX5 regulates expression of cell cycle genes in undifferentiated spermatogonia post-transcriptionally and is required for cell proliferation and survival. DDX5 can also act as a transcriptional co-activator and we demonstrate that DDX5 interacts with PLZF, a transcription factor required for germline maintenance, to co-regulate select target genes. Combined, our data reveal a critical multifunctional role for DDX5 in regulating gene expression programmes and activity of undifferentiated spermatogonia.

T he maintenance of adult tissue integrity and function is often dependent upon a population of resident stem cells that self-renews and generates differentiating cells. The balance between stem cell self-renewal and differentiation can be regulated through extrinsic stimuli, including growth factors produced in a supportive niche, as well as cell intrinsic factors such as transcription factors and post-transcriptional gene regulatory mechanisms. The mammalian testis is an example of such a tissue where tightly co-ordinated transcriptional and posttranscriptional mechanisms harmonise to ensure lifelong maintenance of fertility through the continual production of mature spermatozoa 1 .
In mammals, spermatozoa are produced through the complex process of spermatogenesis that is maintained by a population of cells with self-renewal potential referred to as undifferentiated spermatogonia. During spermatogenesis, undifferentiated spermatogonia undergo mitotic divisions followed by induction of differentiation that ultimately results in production of meiotic spermatocytes. After two meiotic reduction divisions, round spermatids are formed that undergo significant morphological changes during a process known as spermiogenesis to ultimately produce haploid spermatozoa 1 . The entire process of spermatogenesis occurs within seminiferous tubules of the testis and is supported by Sertoli cells, essential somatic components that produce GDNF, a growth factor critical for spermatogonial maintenance 2 . In homoeostatic testis, only a proportion of undifferentiated spermatogonia function as stem cells, while the remainder act as differentiation-committed progenitors. However, progenitor populations can revert to a stem cell state following germ cell depletion and induction of a regenerative response 3,4 . Self-renewing cells within the steady-state undifferentiated spermatogonial pool are marked by expression of Gfra1, Nanos2, and Id4 while committed progenitors express Ngn3, Sox3, and Rarg 3-8 . Spermatogonial differentiation is marked by induction of the cell surface receptor c-KIT and a series of coordinated mitotic divisions prior to meiosis 2,9 . Progression of germ cells through the differentiation pathway is associated with incomplete cytokinesis following cell division, a conserved feature of germline systems between species 7 . Within the undifferentiated compartment, the most primitive cells generally exist as single cells termed A single (A s ) that can divide to produce A s daughters for self-renewal or pairs (A paired or A pr ) and subsequently longer chains of cells that reach 16 cells or more (A aligned or A al ) and are typically differentiation-primed 7 (Fig. 1a). Importantly, in the presence of GDNF and bFGF growth factors, undifferentiated spermatogonia can be cultured long-term in vitro while maintaining stem cell potential 10 .
In addition to extrinsic stimuli, the maintenance and function of undifferentiated spermatogonia are dependent on a complex network of cell intrinsic factors that co-ordinately regulate transcription. Amongst these, transcription factors such as PLZF and SALL4 are important for spermatogonial self-renewal and differentiation, respectively [11][12][13][14] . PLZF maintains self-renewal capacity by enhancing sensitivity of undifferentiated spermatogonia to GDNF through indirect inhibition of mTORC1 signalling [14][15][16] . mTORC1 is a key regulator of mRNA translation, as well as cellular metabolism and aberrant activation of this pathway drives differentiation commitment of spermatogonia in vivo 14,17 . SALL4 was demonstrated to modulate PLZF function in undifferentiated spermatogonia and is required for their differentiation 13 . More recently, we have demonstrated that SALL4 is also required in a PLZF-independent manner for longterm maintenance of undifferentiated spermatogonia activity and regeneration following germline depletion 12 . Other important transcriptional regulators of spermatogonial self-renewal and differentiation have been identified such as ETV5 18 and SOHLH1/2 8 , respectively; however, despite the uncovering of numerous factors, much remains to be understood regarding their interplay in this intricate process. Besides regulation at the transcriptional level, periods of transcriptional quiescence occurring in spermatogenesis during meiotic homologous recombination and spermiogenesis necessitates post-transcriptional regulation of gene expression 1 . Numerous regulators of alternative pre-mRNA splicing, mRNA export and stability, as well as translational control have been shown to be essential for timely gene expression during spermatogenesis 1 . The RNA-binding protein NANOS2 maintains the undifferentiated state of spermatogonia by recruiting differentiation-associated mRNAs to ribonucleoprotein complexes and inhibiting their translation 19 . NANOS2 also directly inhibits mTORC1 through sequestration of the mTOR kinase, thus robustly suppressing translation of mRNAs required for differentiation 19 . More recently, BCAS2 has been shown to be an essential regulator of alternative pre-mRNA splicing in spermatogonia 20 . Loss of BCAS2 in male mice results in infertility as a consequence of aberrant splicing of genes necessary for meiotic initiation 20 . While spermatogonial fate is regulated by both transcriptional and post-transcriptional mechanisms, involvement of specific RNA processing factors in the maintenance and function of undifferentiated spermatogonia is poorly appreciated.
DEAD box helicase 5 (DDX5) is an ATP-dependent RNA helicase originally described as important for the unwinding of RNA secondary structure, but now known to be involved in multiple aspects of RNA processing including splicing, mRNA export, transcript stability, rRNA biogenesis, and microRNA processing [21][22][23][24][25] . DDX5 belongs to the DEAD box protein family, which is characterised by a conserved Asp-Glu-Ala-Asp (DEAD) motif within a helicase core 22 . Although the helicase core is largely conserved amongst DEAD box family members, highly variable N-terminal and C-terminal regions confer additional roles to these proteins including the ability to bind DNA and act as transcriptional co-factors in conjunction with partner transcription factors 22,26 . In particular, DDX5 acts as a transcriptional co-factor for MYOD and RUNX2, key regulators of myogenesis and skeletal development, respectively 27,28 . DDX5 has also been implicated in progression of breast and prostate cancer due to its ability to act as a co-activator of oestrogen and androgen receptor signalling 22,29 .
Previous studies have suggested a role for DDX5 in the male germline; DDX5 was found to be specifically expressed in human spermatogonia through validation of single cell RNA-sequencing (RNA-seq), and the RNA-binding protein RBM5 was found to interact with DDX5 in mouse round spermatids and be required for spermatogenesis 30,31 . DDX5 has also been suggested to be a negative regulator of WNT signalling in the GC-1spg germ cell line 32 . Despite its expression within the male germline, a precise role for DDX5 has not been elucidated and its involvement in maintenance of spermatogenesis is unknown. In this study, we identified DDX5 as an interacting partner of PLZF in spermatogonia and found that DDX5 is indispensable for maintenance of spermatogenesis in mice. We demonstrate that DDX5 has a multifaceted role within undifferentiated spermatogonia that includes both transcriptional and post-transcriptional functions. We show that DDX5 is important for multiple RNA processing steps including pre-mRNA splicing of key genes required for spermatogenesis, as well as the regulation of mRNA nuclear export and transcript stability of cell cycle-related genes. In addition, we provide evidence that DDX5 is involved in transcriptional regulation together with PLZF. Our data uncovers an essential role for DDX5 within undifferentiated and differentiating spermatogonia, and characterises DDX5 as a critical regulator of male fertility. I n p u t I n p u t I P : I g G I P : I g G I P : α P L Z F I P : α D D  14 . To understand further the role of PLZF within the male germline, we performed PLZF immunoprecipitation (IP) on lysates from cultured wildtype murine undifferentiated spermatogonia followed by mass spectrometry (MS) analysis to identify interacting proteins 12 . Including the known PLZF-interacting protein SALL4 13 , we identified putative PLZF binding partners (Supplementary Data 1) with DDX5 amongst the top hits defined by presence in each of three independent samples. We confirmed this interaction by reciprocal co-IP and western blot where we observed bands at expected molecular weights when probing for PLZF (75 kDa) and DDX5 (68 kDa) (Fig. 1b). Next, we assessed in vivo expression of DDX5 in mouse seminiferous tubules by immunofluorescence (IF) of testis sections and found nuclear DDX5 expression within all spermatogenic cells (spermatogonia, spermatocytes, spermatids) marked by the germ cell marker VASA 33,34 and distinguished according to their location within the seminiferous tubule 35 (Fig. 1a, c). In addition, we examined expression of DDX5 in non-human primate (marmoset) and human testis sections and observed a comparable expression pattern in spermatogenic cells to that of mouse, including within PLZF-positive cells ( Supplementary Fig. 1a, b). As spermatogonia reside adjacent to the seminiferous tubule basement membrane, whole mount IF can be used to distinguish subsets of spermatogonia through the presence of specific markers plus according to their morphology and topology 7 . Using this method, we examined expression of DDX5 within selfrenewing, committed progenitor, and differentiating spermatogonia. First, we found DDX5 expressed within supporting Sertoli cells as shown by co-localisation with the Sertoli cell marker, SOX9 36 (Fig. 1d). Next, we observed DDX5 expression localised to the nuclei of GFRα1-positive (self-renewing, A s and A pr ) 3 , SOX3-positive (committed progenitor, A al ) 8 , and c-KIT-positive (differentiating) 1,9 spermatogonia, with GILZ used as a broad spermatogonial marker 37 (Fig. 1a, e). We concluded that DDX5 is expressed throughout the male germline and in supporting somatic cells.
DDX5 is essential for the maintenance of spermatogenesis. To define the functional role of DDX5 in the adult male germline, we sought to ablate Ddx5 using a conditional knockout model. Previously, we have used transgenic mice containing a tamoxifeninducible Cre recombinase under control of the ubiquitin c promoter (UBC-Cre ERT2 ) 38 to drive efficient Cre-LoxP-mediated gene recombination in spermatogonia, while meiotic and testis somatic cells remain mostly unaffected 12 . We crossed UBC-Cre ERT2 mice with previously described Ddx5 flox/flox mice 39 to generate a tamoxifen-inducible Ddx5 knockout line (Ddx5 flox/flox ; UBC-Cre ERT2 ). We treated 8-week-old to 12-week-old male Ddx5 flox/flox ; UBC-Cre ERT2 mice (Ddx5 TAM-KO ) and Ddx5 flox/flox control mice with tamoxifen (TAM) and collected testis samples 5, 7, 14, and 30 days (D) post-treatment. Analysis of testis sections by IF revealed a notable loss of PLZF-positive spermatogonia by D7 following TAM-induced Ddx5 ablation (Fig. 2a). To verify loss of all PLZF-positive spermatogonial subsets, we stained testis sections for markers of self-renewing (GFRα1), progenitor (SOX3) and differentiating (c-KIT) cells ( Supplementary Fig. 2). We did not observe any Ddx5-ablated cells remaining in these spermatogonial populations by IF at D7 post-TAM treatment, besides few c-KIT-positive cells adjacent to the basement membrane of seminiferous tubules ( Supplementary Fig. 2c). In addition, we quantified Ddx5 ablation and total cell numbers for Sertoli cells, spermatocytes, and round spermatids by IF at D7 (Supplementary Fig. 3). We found no significant difference in the number of Sertoli cells, spermatocytes or round spermatids between control and TAM-treated Ddx5 TAM-KO testis crosssections. Moreover, we found no significant differences in the proportion of those cells positive for DDX5, suggesting the model used in this study does not result in significant Ddx5 ablation within testis cells other than spermatogonia ( Supplementary  Fig. 3). Interestingly, in both control and TAM-treated Ddx5 TAM-KO testis we noted that DDX5 was downregulated in a subset of SYCP3-positive meiotic cells proximal to the basement membrane, suggesting dynamic regulation of DDX5 during meiosis ( Supplementary Fig. 3b). Next, we analysed populations of spermatogonia isolated from whole testis samples up to D14 post-TAM by intracellular staining and flow cytometry according to expression of PLZF and c-KIT 12,13,17 . We observed significant reduction of undifferentiated (PLZF + c-KIT − ), early differentiating (PLZF + c-KIT + ) and late differentiating spermatogonia (PLZF low c-KIT + ) in Ddx5 TAM-KO testes by D5 post-TAM compared with control, with these significant reductions persisting through to D14 ( Supplementary Fig. 4a, b).
By D30 post-Ddx5 ablation, analysis of testis cross-sections by IF revealed seminiferous tubules completely devoid of germ cells as indicated by the absence of VASA-positive cells and a Sertoli cell-only phenotype (Fig. 2a). Whole mount IF of seminiferous tubules at D30 post-Ddx5 ablation confirmed significant loss of PLZF-positive spermatogonia, with only Ddx5-retaining Sertoli cells visible in most TAM-treated Ddx5 TAM-KO tubules (Fig. 2b). Only tubules where rare Ddx5-retaining PLZF-positive spermatogonia were present showed indications of germ cell repopulation ( Supplementary Fig. 4c). At D30, TAM-treated Ddx5 TAM-KO testes weighed significantly less than control testes as anticipated (0.12 ± 0.01% of total body mass versus 0.30 ± 0.01%, mean ± SEM, n = 18 control, n = 20 TAM, P < 0.0001, two-tailed unpaired t-test) (Fig. 2c). Quantification of PLZF-positive cells per tubule cross-section demonstrated significant reduction in numbers of spermatogonia at D5, D7, D14, and D30 in TAMtreated Ddx5 TAM-KO testes compared with controls (Fig. 2d). Besides this testis phenotype, no other overt phenotype was observed in Ddx5 TAM-KO mice and no significant difference in body mass was discerned at any time point (Fig. 2e). This is of interest given that UBC-Cre ERT2 is broadly expressed in adults 38 , and TAM treatment of Ddx5 TAM-KO mice would ablate Ddx5 in multiple tissues besides the testis. Our data indicate that DDX5 plays critical roles in maintenance of spermatogenesis and its loss results in rapid and profound depletion of adult spermatogonia.

DDX5 is indispensable for the maintenance of spermatogonia.
Having demonstrated the requirement of DDX5 in maintenance of spermatogonia in vivo, we sought to explore mechanisms underlying DDX5 function and confirm its cell-autonomous role in the germline using an in vitro system 4,14 . Therefore, we established cultures of undifferentiated spermatogonia from untreated Ddx5 TAM-KO mice and induced Ddx5 ablation by treatment with 4-hydroxytamoxifen (TAM) 12 . Cultured Ddx5 TAM-KO spermatogonia treated with vehicle alone were used as controls. Ddx5 was efficiently ablated in Ddx5 TAM-KO spermatogonia following TAM and resulted in pronounced depletion of cultured cells such that only few small colonies of PLZFpositive Ddx5-ablated cells remained at D2 post-treatment (Fig. 3a). These results are consistent with a critical cellautonomous role for DDX5 in spermatogonial maintenance. In contrast, mouse embryonic fibroblasts (MEFs) established from Ddx5 TAM-KO embryos and treated with TAM persisted in culture and continued to proliferate despite loss of Ddx5 suggesting a specific requirement for DDX5 within spermatogonia (Fig. 3b). It was noted that expression of DDX17, a functionally co-operative paralog of DDX5 26 , was upregulated in Ddx5 TAM-KO MEFs following TAM when compared with vehicle-treated MEFs (Fig. 3b, c). However, no corresponding increase in DDX17 was observed in Ddx5 TAM-KO spermatogonia and although the ratio of p72 and p82 isoforms appeared slightly altered upon Ddx5 loss, this was not statistically significant (Fig. 3b, c and Supplementary Fig. 5). These data suggest that loss of DDX5 function in MEFs may be compensated for through upregulation of DDX17, whereas its function is indispensable in spermatogonia. Despite significant loss of cultured spermatogonia upon Ddx5 ablation, we were able to extract RNA from remaining Ddx5ablated cells for gene expression analysis. TAM-treated and vehicle-treated Ddx5 TAM-KO spermatogonia were FACS-purified at D2 based on expression of EpCAM to eliminate potential feeder cell contamination 12 . We performed whole transcriptome analysis using RNA-seq to gain insight into the consequences of Ddx5 loss in undifferentiated spermatogonia. We found that Ddx5 loss resulted in differential expression of 6934 genes (false discovery rate <0.05) ( Fig. 3d and Supplementary Data 2). We confirmed downregulation of Ddx5 in TAM-treated samples and found aberrant expression of a number of key genes required for maintenance and function of spermatogonia. Key stem-associated and progenitor-associated genes such as Nanos2, Gfra1, Sohlh1, Plzf (Zbtb16), and Gilz (Tsc22d3) were downregulated and these changes confirmed by RT-qPCR in independent samples (Fig. 3d, Supplementary Fig. 6a and Supplementary Data 2). However, a number of spermatogonia-associated and spermatogenesisassociated genes were either upregulated (e.g., Rarg, Kit, Thy1, T, Lhx1), downregulated (Nanos3, Rec8, Dnmt3b, Sycp3), or unchanged (Sall4, Pou5f1), suggesting an overall dysregulation of gene expression rather than transition to a particular state (e.g., stem, differentiating, meiotic induction) (Fig. 3d, Supplementary  Fig. 6a and Supplementary Data 2). We also noted upregulation of Ddx17 in TAM-treated Ddx5 TAM-KO spermatogonia by RNAseq analysis; however, this could not be confirmed by RT-qPCR and corresponding levels of DDX17 protein were unchanged ( Fig. 3b-d, Supplementary Fig. 5 and Supplementary Fig. 6a). Moreover, the severity of Ddx5 TAM-KO phenotype in spermatogonia indicates that DDX17 cannot compensate for DDX5 function in undifferentiated spermatogonia.
Concerning the general mechanism by which spermatogonia were lost following Ddx5 ablation, we observed upregulation of Trp53 and genes encoding effectors of p53-mediated apoptosis (Bax, Puma, Noxa, Gadd45b) 40,41 by RNA-seq analysis (Fig. 3e) and we confirmed upregulation of Puma and Gadd45b by RT-qPCR in independent samples ( Supplementary Fig. 6b). To confirm loss of Ddx5 TAM-KO spermatogonia occurred by apoptosis, we treated cells with TAM in combination with the pan-caspase inhibitor, Z-VAD-FMK. At D2 post-treatment, TAM-treated cells were almost completely lost, whereas significantly more cells remained when treated in combination with Z-VAD-FMK. As expected, control cells treated with vehicle alone or in combination with Z-VAD-FMK showed increased cell numbers (Fig. 3f). By IF, we confirmed successful gene ablation in TAM-treated Ddx5 TAM-KO spermatogonia under all conditions and increased levels of apoptotic marker cleaved capsase-3 compared with controls ( Fig. 3g). Importantly, we confirmed that the substantial increase in cleaved caspase-3-positive cells evident upon Ddx5 deletion was inhibited by Z-VAD-FMK ( Fig. 3f, g). Although we could not confirm upregulation of all genes identified in our RNA-seq analysis by RT-qPCR, these data indicate that following Ddx5 ablation, undifferentiated spermatogonia are lost due to p53-mediated apoptosis.
Next, we aimed to determine whether the helicase activity of DDX5 is required for survival of spermatogonia, as previous studies investigating the role of DDX5 in other systems have shown helicase-dependent and independent functions 22 . We transduced Ddx5 TAM-KO cultured spermatogonia with lentiviral constructs containing wildtype DDX5 (DDX5-WT) or a helicaseinactive form of DDX5 42 (DDX5-NEAD) prior to TAM treatment. Although we were unable to elicit expression of DDX5 constructs to the same level as endogenous DDX5 and observed substantial apoptosis in all conditions ( Supplementary  Fig. 6c), we nevertheless observed~5-fold higher cell numbers remaining in the DDX5-WT condition and~3-fold higher in the DDX5-NEAD condition compared with control transduced cells at D2 post-TAM (Fig. 3h). These findings suggest that although the helicase activity of DDX5 is not entirely dispensable for survival of undifferentiated spermatogonia, there exists a helicaseindependent function for DDX5 within spermatogonia. Moreover, Ddx5 is specifically required for survival of undifferentiated spermatogonia, with its loss leading to aberrant expression of spermatogonial genes and rapid induction of apoptosis.
DDX5 is required for splicing of key genes in spermatogonia. DDX5 is involved in numerous transcriptional and posttranscriptional gene regulatory processes across various biological systems through interactions with other proteins [21][22][23][24]26 . To gain additional insight into DDX5 function in undifferentiated spermatogonia, we characterised DDX5 interacting proteins by IP and MS analysis from lysates of cultured wildtype cells. We identified 445 candidate proteins that interacted with DDX5 (Supplementary Data 3) and subjected these results to gene ontology (GO) term overrepresentation analysis according to classification by molecular function and biological process using PANTHER 43 . For GO molecular function and GO biological f Quantification of cell fold recovery following tamoxifen-induced Ddx5 ablation (Ddx5 TAM-KO ) in cultured spermatogonia at D2 versus vehicle-treated control (Control) and effects of treatment with pan-caspase inhibitor Z-VAD-FMK (with DMSO used as a control). All differences between groups are statistically significant (P < 0.05, Mann-Whitney U test, n = 4 biologically independent cell lines per condition, mean ± SEM shown). g Immunofluorescence of cultured spermatogonia upon Ddx5 ablation at D1 depicting an increase in caspase-mediated apoptosis. Cleaved caspase-3 (cCASP3) is used as a marker of apoptotic cells, with SALL4 used as a marker of spermatogonia. Inhibition of apoptosis using the pan-caspase inhibitor Z-VAD-FMK prevents loss of spermatogonia upon Ddx5 ablation. Nuclei are counterstained with DAPI (DNA). All scale bars = 100 μm. h Quantification of cell fold recovery at D2 in cultured murine spermatogonia transduced with wildtype DDX5 (WT), helicase-inactive mutant DDX5 (NEAD) or tdTomato control constructs prior to tamoxifen-induced Ddx5 ablation at D0. *P < 0.05; n.s.: not significant. Mann-Whitney U test, n = 4 biologically independent cell lines per condition. Mean ± SEM shown NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-09972-7 ARTICLE NATURE COMMUNICATIONS | (2019) 10:2278 | https://doi.org/10.1038/s41467-019-09972-7 | www.nature.com/naturecommunications process, we found significant overrepresentation of terms related to RNA binding and post-transcriptional processes, respectively ( Fig. 4a and Supplementary Data 4 and 5). To confirm interactions identified by MS analysis that implicate DDX5 in RNA processing steps, we performed DDX5 IP from cultured wildtype spermatogonia followed by western blot analysis. We confirmed interaction of DDX5 with ELAVL1 (mRNA stability), PABP1 (mRNA stability/export), and SRSF3 (splicing/mRNA export) (Fig. 4b). Our data indicates that DDX5 is primarily involved in RNA processing in undifferentiated spermatogonia.
Given these findings and that DDX5 is implicated in splicing within other systems 21,25,44 , we sought to establish a role for DDX5 in splicing within undifferentiated spermatogonia. We used the MISO (mixture-of-isoforms) probabilistic framework 45 to analyse our RNA-seq data for identification of alternative splicing events upon loss of Ddx5. We discovered 667 alternative splicing events upon loss of Ddx5 in spermatogonia (Bayes factor >10, ΔPSI ≥0.20) with the majority of events (= 406/667) annotated as cassette exon, where differential expression of specific exons was identified between control and TAM-treated Ddx5 TAM-KO samples ( Fig. 4c and Supplementary Data 6). From these events, we selected candidates based on involvement in spermatogenesis (Dazl, Ehmt2) [46][47][48] and key cellular processes such as DNA repair/ homologous recombination (Rad50, Rad18) 49,50 and autophagy (Becn1) 51 . Loss of Ddx5 resulted in differential expression of exon 8 in Dazl (increased exon inclusion), exon 10 in Ehmt2 (increased exon skipping), exon 15 in Rad50 (increased exon skipping), exon 10 in Rad18 (increased exon inclusion), and exon 3 in Becn1 (increased exon skipping) (Fig. 4d). We confirmed these candidates were bound to DDX5 by RNA immunoprecipitation (RIP)-PCR (Fig. 4e), suggesting DDX5 is directly involved in their splicing, and we validated these alternative splicing events by PCR (Fig. 4f). Other candidates were confirmed to be differentially spliced upon Ddx5 loss (e.g., Fmr1) but were not detectably bound to DDX5, suggesting their splicing is indirectly regulated (Supplementary Fig. 7a-c). In addition, we confirmed that Hprt was not bound to DDX5 as a negative RIP control (Supplementary Fig. 7c).
Interestingly, the DNA repair genes Rad18 and Rad50 are required for maintenance of spermatogenesis in mice and loss of function of either gene results in infertility 49,50 . When we assessed gene expression of mis-spliced candidates, we found Rad50 to be substantially downregulated upon Ddx5 loss according to RNAseq ( Fig. 4g), and we confirmed this finding by RT-qPCR ( Supplementary Fig. 7d). By coding sequence analysis of Rad50 mRNA, we found that skipping of exon 15 results in the introduction of a premature termination codon ( Supplementary  Fig. 7e) and is likely to result in nonsense mediated decay of the transcript 52 . Accordingly, we confirmed a decrease in RAD50 protein in TAM-treated Ddx5 TAM-KO spermatogonia compared with controls by western blot (Fig. 4h) and IF (Fig. 4i). These findings demonstrate that DDX5 is an important trans-acting regulator of splicing in undifferentiated spermatogonia and its loss results in dysregulated splicing of key genes required for spermatogenesis. Moreover, p53-mediated apoptosis of spermatogonia following ablation of Ddx5 may be a direct consequence of splicing defects 53 .
DDX5 regulates cyclin mRNA export and stability. To more fully understand the transcriptional consequences of Ddx5 loss in spermatogonia, we performed pathway analysis using the Ingenuity platform on differentially expressed genes identified by RNA-seq. The top two pathways identified for top canonical pathways and top tox lists (i.e., top biological "toxicity" associated with the observed gene expression changes) were found to be cell cycle-related pathways ( Fig. 5a and Supplementary Data 7). We observed aberrant expression of a number of genes involved in cell cycle regulation upon Ddx5 ablation in spermatogonia, including significant upregulation of the cell cycle arrest gene Cdkn1a (p21) ( Fig. 5b and Supplementary Fig. 8a). However, core cell cycle-related genes were not predicted to be aberrantly spliced (Supplementary Data 6). Cell cycle analyses of surviving TAMtreated Ddx5 TAM-KO cells by flow cytometry revealed a significant decrease in the proportion of cells in S-phase and an increase in the proportion of cells in G2/M-phase compared with controls (Fig. 5c). These data suggest that Ddx5 is specifically required in undifferentiated spermatogonia to maintain the expression of genes important for cell cycle and its loss results in cell cycle arrest at both G1 and G2 checkpoints.
CCNA2 regulates G1/S and G2/M transitions of the cell cycle, and is strongly expressed in spermatogonia 54 . Expectedly, given decreased Ccna2 expression observed upon Ddx5 ablation, we observed a decrease in CCNA2 level in TAM-treated cultured Ddx5 TAM-KO spermatogonia compared with controls (Fig. 5d, e and Supplementary Fig. 8a). However, we noted that despite arrest at both checkpoints, positive regulators of the cell cycle such as Ccnd1 and Ccnd2 55 were upregulated in TAM-treated Ddx5 TAM-KO spermatogonia (Fig. 5b, c and Supplementary  Fig. 8a). When analysing protein levels, we found that CCND1 and CCND2 were decreased upon Ddx5 ablation in contrast to respective mRNA levels ( Fig. 5b, d, e), suggesting posttranscriptional regulation of these genes was perturbed. To investigate whether DDX5 was directly regulating cyclin expression post-transcriptionally, we performed RIP-qPCR in cultured wildtype spermatogonia using an antibody against DDX5. We analysed cyclin mRNAs that were differentially expressed according to RNA-seq (Fig. 5b). For each transcript, we determined fold enrichment versus pull down with immunoglobulin G (IgG) control and found Ccnb2 (~70-fold), Ccnd1 (~40fold), Ccna2 (~30-fold), and Ccnd2 (~12-fold) to be the most strongly enriched transcripts (Fig. 5f).
Given our observation of increased Ccnd1 and Ccnd2 mRNA expression yet decreased levels of corresponding protein in Ddx5ablated spermatogonia, we considered that loss of DDX5 function results in impaired nuclear export of these transcripts. Restricted localisation of DDX5 to nuclei in spermatogonia ( Fig. 1e) together with binding of cyclin transcripts to DDX5 suggested direct regulation of cyclin expression by post-transcriptional processes other than translation. We assessed nuclear export of Ccnd1 and Ccnd2 mRNA transcripts by determining nuclearcytoplasmic ratios at D2 post-TAM in cultured Ddx5 TAM-KO spermatogonia. We confirmed significant accumulation of both Ccnd1 and Ccnd2 mRNA in the nucleus of Ddx5-ablated spermatogonia (Fig. 5g), indicating that DDX5 directly regulates export of these transcripts. In contrast, nuclear export of Ccna2 mRNA was not significantly disrupted upon Ddx5 loss (Supplementary Fig. 8b).
Ccna2 transcript is bound to DDX5 and both mRNA plus protein levels of CCNA2 are downregulated upon Ddx5 loss, suggesting a distinct mode of post-transcriptional regulation by DDX5. Consequently, we assessed stability of Ccna2 mRNA upon Ddx5 loss in Ddx5 TAM-KO spermatogonia treated with the transcription inhibitor actinomycin D. However, there was no evident difference in Ccna2 mRNA stability between control and TAM-treated samples (Fig. 5h), suggesting that Ccna2 downregulation occurs as an indirect transcriptional consequence of Ddx5 ablation. Importantly, we observed a significant reduction in levels of Ccnb2 mRNA upon Ddx5 loss after 4 h of actinomycin D treatment, indicating that stability of Ccnb2 transcript was reduced (Fig. 5h). Given significant reduction in Ccnb2 expression upon Ddx5 loss ( Fig. 5b and Supplementary Fig. 8a) and association of Ccnb2 mRNA with DDX5, our results indicate that the stability of Ccnb2 transcript is directly enhanced by DDX5. We used Myc as an assay control and observed mRNA stability comparable to previous reports in both control and TAM-treated cells 56 (Supplementary Fig. 8c).
These findings demonstrate that DDX5 is required for posttranscriptional regulation of cyclin mRNAs in undifferentiated spermatogonia. Furthermore, that cell cycle arrest observed upon Ddx5 loss is, at least in part, a consequence of impaired export and stability of specific cyclin transcripts.  . ILF3 (interleukin enhancer binding factor 3, also known as NF90 or NFAR) is an RNA-binding protein important for transcriptional and post-transcriptional regulation of Il2 in T cells and plays key regulatory roles in cell cycling [58][59][60][61] .
In light of the identified functional connection between DDX5 and ILF3, we examined protein expression of ILF3 in male germline cells by IF. We observed ILF3 expression in all PLZFpositive cultured undifferentiated spermatogonia (Fig. 6a), as well as in vivo co-localisation with DDX5 in the nucleus of spermatogonia, spermatocytes, spermatids, and Sertoli cells of mouse seminiferous tubules (Fig. 6b). Previous studies have demonstrated that ILF3 is involved in post-transcriptional regulation of cyclin mRNAs 57,62 . Therefore, we performed ILF3 RIP-qPCR experiments in cultured wildtype spermatogonia to identify cyclin mRNAs bound to ILF3 and elucidate a potential co-regulatory role with DDX5. We found strong enrichment of Ccnb2 (~1124-fold), Ccna2 (~718-fold), Ccnd1 (~84-fold), Ccne1 (~73-fold), and Ccnd2 (~62-fold) transcripts (versus IgG control), whereas others such as Ccnd3 and Ccna1 showed minimal enrichment or did not appear to be bound to ILF3, respectively (Fig. 6c).
ILF3 regulates the stability of Ccne1 mRNA by direct binding to its 3'-UTR and post-transcriptionally regulates Ccnd1 through interactions with IGF2BP3 57,62 . Given the DDX5/ILF3 interaction and known role of ILF3 in cyclin transcript regulation, we sought to determine whether loss of Ilf3 in undifferentiated spermatogonia would result in a phenotype similar to that observed upon loss of Ddx5. We therefore used a lentiviral doxycycline (DOX)-inducible CRISPR/Cas9 system 63 to ablate Ilf3 in cultured wildtype spermatogonia (referred to as Ilf3 -/cells). At D4 post-DOX treatment, we confirmed loss of Ilf3 by RT-qPCR and at the protein level by IF and western blot (Fig. 6d, f, g). In contrast to Ddx5 loss, Ilf3 -/spermatogonia persisted in culture and continued to proliferate. However, we observed a significantly reduced growth rate of Ilf3 -/cells compared with DOX-treated wildtype cells when analysed D2 post-treatment (Fig. 6d, e). Analysis of gene expression in Ilf3 -/spermatogonia by RT-qPCR did not find any significant change in expression of analysed cyclin genes compared with controls (Fig. 6f). However, CCND1 protein was substantially decreased upon loss of Ilf3 (Fig. 6g); a phenotype analogous to observed Ccnd1 mRNA export defects upon Ddx5 loss. However, we were unable to confirm consistent changes in Ccnd1 nuclear-cytoplasmic ratio upon Ilf3 loss, suggesting alternate post-transcriptional regulatory mechanisms are in play ( Supplementary Fig. 9).
The difference in severity of cellular phenotypes and disruption in cyclin expression upon loss of Ddx5 and Ilf3 suggests that regulation of cyclin genes in undifferentiated spermatogonia and consequent effects on cell cycling are not dependent on a DDX5/ ILF3 complex. Rather, DDX5-dependent post-transcriptional regulation is important for cyclin mRNA expression and cell cycle regulation whereas ILF3 appears dispensable, with the exception of Ccnd1.
DDX5 and PLZF share DNA binding sites and co-regulate Ilf3. Previous studies in other systems have shown that DDX5 interacts with and modulates activity of transcriptional regulators including MYOD, RUNX2, the tumour suppressor p53, and oestrogen receptor alpha 22 . Considering the ability of DDX5 to interact with transcription factor PLZF (Fig. 1b), we investigated whether DDX5 acts as a PLZF co-factor in undifferentiated spermatogonia. We therefore performed ChIP-sequencing (ChIPseq) for DDX5 on cultured wildtype undifferentiated spermatogonia and identified 101 putative binding sites, of which 31 were found within gene promoter regions ( Fig. 7a and Supplementary Data 8). Motif analysis of bound regions identified enrichment of the E-box-containing CLEAR element (5'-GTCACGTGAC-3'), a binding motif for the autophagy regulator TFEB 64 (Fig. 7b). Amongst our ChIP-seq analysis, we found DDX5 bound within the promoter regions of autophagy-related genes Lamp1, Gns, Hexa, Tmem55b, and Atp6v1h (Fig. 7c). Of these genes, only Hexa (1.12-fold downregulated), Tmem55b (1.13-fold upregulated), and Atp6v1h (1.15-fold upregulated) were differentially expressed according to RNA-seq analysis in cultured Ddx5ablated spermatogonia compared with controls (false discovery rate <0.05) (Fig. 7d); however, these minor changes could not be confirmed by RT-qPCR ( Supplementary Fig. 10d). Furthermore, we were unable to establish an interaction between DDX5 and TFEB by co-IP, indicating that DDX5 is unlikely to be a TFEB cofactor ( Supplementary Fig. 10a). Therefore, while recruitment of DDX5 to CLEAR motifs is compelling, our data indicate that Fig. 4 DDX5 is an essential regulator of splicing in undifferentiated spermatogonia. a Gene ontology (GO) term enrichment analysis of data obtained from DDX5 immunoprecipitation followed by mass spectrometry using wildtype murine cultured undifferentiated spermatogonia (n = 3 independent experiments). b DDX5 immunoprecipitation followed by western blot in wildtype cultured spermatogonia confirming the interaction of DDX5 with proteins involved in pre-mRNA splicing (SRSF3), maintenance of mRNA stability (ELAVL1, PABP1, ILF3, ILF2, IGF2BP3), mRNA export (PABP1, SRSF3, ILF2, ILF3, IGF2BP3) and translation (IGF2BP3). Representative of n = 2 independent experiments. c Summary of differential splicing analysis performed between control and Ddx5-ablated cultured spermatogonia. Numbers of predicted alternative splicing events in each category upon Ddx5 deletion are indicated. d Visualisation of differential splicing analysis of RNA-sequencing data comparing control and Ddx5-ablated spermatogonia. Tracks are shown for selected candidate genes (left). Red arrows indicate differentially spliced exon. Schematics of alternative splicing events are shown (blue and yellow rectangles) (middle). Change in "percent spliced in" between conditions is shown as a value below splicing schematics (ΔPSI) and in bar charts (right). #: changes meet analysis cut-offs (ΔPSI >0.20, Bayes Factor ≥10). PSI ± 95% confidence interval shown. e RNA-immunoprecipitation using DDX5 antibody in cultured wildtype spermatogonia followed by PCR and gel electrophoresis for differentially spliced candidates depicted in d. Representative of n = 3 independent samples per condition. f PCR validation of differentially spliced candidates identified in d. CTL vehicle-treated control spermatogonia, KO Ddx5-ablated cultured spermatogonia. Differentially spliced exons are depicted by "Δexon number". Representative of n = 3 independent samples. Position of the PCR primers used are depicted in d in orange beneath schematic showing gene structure. g Bar chart of RNA-sequencing analysis for selected candidates. Fold change in Ddx5-ablated spermatogonia relative to vehicle-treated control. # denotes FDR <0.05. Mean ± SEM shown; n = 4 independent biological replicates per condition. h RAD50 protein levels by western blot in vehicle-treated control (Control) versus Ddx5-ablated cultured spermatogonia (Ddx5 TAM-KO ). *P < 0.05; two-tailed unpaired t-test; mean ± SEM shown; n = 4 independent biological replicates per condition. Representative western blot shown on right. i RAD50 immunofluorescence of vehicle-treated control (top) and Ddx5-ablated (bottom) cultured spermatogonia at D2 post-treatment. Representative of n = 4 independent biological replicates per condition. Scale bars = 100 μm DDX5 is not an essential transcriptional regulator of autophagyrelated genes in spermatogonia.
To investigate the potential role of DDX5 as a PLZF transcriptional co-factor in spermatogonia, we compared DDX5 ChIP-seq and RNA-seq datasets to identify DDX5-bound genes differentially expressed upon Ddx5 loss. Forty-nine genes were potentially transcriptionally regulated by DDX5, with the number of upregulated and downregulated genes (25 versus 24, respectively) suggesting DDX5 may act as both an activator and repressor of transcription ( Fig. 7e and Supplementary Fig. 10b, c). Next, we compared these 49 genes with a published PLZF ChIPseq study 16 to identify shared binding sites. We found DDX5 and PLZF overlapping binding sites within promoter regions of Amdhd2, Ilf3, Arhgap12, Tmem55b, 4933434E20Rik, and Zbtb25 ( Fig. 7f and Supplementary Data 8). Next, we determined changes in expression of these genes according to RNA-seq and found significant downregulation of Amdhd2 (1.53-fold), Ilf3 (1.44fold), and Arhgap12 (1.31-fold), and upregulation of Tmem55b (1.13-fold, as described earlier), 4933434E20Rik (1.16-fold) and Zbtb25 (1.24-fold) upon Ddx5 loss. However, only changes in Ilf3 and Amdhd2 expression could be confirmed by RT-qPCR (Fig. 7d, g and Supplementary Fig. 10e). We examined expression of these DDX5-bound and PLZF-bound genes in undifferentiated spermatogonia isolated from testes of Plzf-null mice as described 11,12 . However, from this panel of genes only Ilf3 was significantly downregulated in Plzf-null cells compared with wildtype controls (Fig. 7h). We identified DDX5 binding to the promoter of Ilf3 as one of the top peaks by ChIP-seq and confirmed that this site overlapped with a previously identified PLZF-binding site (Fig. 7i). In addition to downregulation of Ilf3 mRNA, we found reduced levels of ILF3 protein at D2 in TAM-treated versus vehicle-treated Ddx5 TAM-KO spermatogonia (Fig. 7j). Finally, we performed siRNA-mediated Plzf knockdown in cultured spermatogonia (with non-targeting siRNA used as control) followed by DDX5 ChIP-qPCR to determine whether DDX5 recruitment to gene promoters was PLZF-dependent. First, we confirmed efficient Plzf knockdown by western blot (Supplementary Fig. 11a). Next, we performed DDX5 ChIP in control versus Plzf knockdown spermatogonia and determined enrichment by qPCR for peak regions identified by DDX5 and PLZF ChIP-seq 16 (Supplementary Fig. 11b). We selected candidate PLZF-DDX5 co-regulated genes Ilf3, Amdhd2 and Lamp1 for analysis, plus uniquely DDX5-bound and PLZFbound genes Zmym6 and Tex13b 16 , respectively, as controls. However, we found no difference in DDX5 enrichment at Ilf3, Amdhd2 and Lamp1 promoters between control and Plzf siRNA conditions, suggesting that DDX5 recruitment to these target genes is PLZF-independent ( Supplementary Fig. 11c).
We conclude that while DDX5 and PLZF are both bound at the Ilf3 promoter and co-regulate Ilf3 expression, other factors besides PLZF mediate DDX5 recruitment to this gene. Furthermore, our data indicates that DDX5 is a multifaceted regulator of ILF3 function given its association with the ILF3/ILF2/IGF2BP3 protein complex and ability to regulate Ilf3 expression.

Discussion
Spermatogenesis is a complex process dependent on the coordinated regulation of gene expression at transcriptional and post-transcriptional levels 1 . It has been demonstrated that the transcription factor PLZF is important for self-renewal of undifferentiated spermatogonia, the population of adult stem and progenitor cells that enable lifelong male fertility 11,14 . More recently, the role of various RNA-binding proteins during spermatogenesis has been uncovered, particularly during meiotic and post-meiotic stages of spermatogenesis where periods of transcriptional quiescence demand mechanisms for post-transcriptional control of gene expression 1 . However, few RNA-binding factors with specific functions in undifferentiated spermatogonia have been identified. NANOS2 19 , DND1 65 , and BCAS2 20 have been described as important RNA-binding proteins within undifferentiated spermatogonia and each has a unique role in post-transcriptional regulation, highlighting the complexity of this process and the need for further investigation.
In this study, we have identified an interaction between PLZF and the RNA helicase DDX5. We demonstrate that DDX5 is expressed throughout the male germline and its loss has catastrophic effects on spermatogenesis. Using a murine knockout model, we show that Ddx5 ablation results in rapid loss of spermatogonia in vivo and in vitro. Moreover, we demonstrate that Ddx5 ablation causes aberrant expression of genes involved in maintenance of spermatogonia, cell cycle, and apoptosis. Ddx5 is particularly essential for spermatogonial function; the inducible Cre driver used to delete Ddx5 in vivo is active in multiple tissues 38 , yet TAM-treated Ddx5 TAM-KO mice did not display any overt phenotype besides germline depletion.
While a specific role for DDX5 within the male germline has not been described previously, a study detailed DDX5 expression in subsets of spermatogenic cells in humans 30 . Consistent with this study, we identified robust expression of DDX5 in PLZFpositive spermatogonia of humans and non-human primates although DDX5 was also detected in spermatocytes, spermatids, and somatic cell populations within the testis. We observed a ubiquitous expression of Ddx5 in testis cells of mice and previous studies also identified DDX5 within other murine spermatogenic cell types 31 , confirming its broad expression within the testis. Nevertheless, expression of DDX5 in primate spermatogonia suggests that the essential role of DDX5 in mouse spermatogonial maintenance is conserved in other mammalian species.
In this study, we observed significant loss of spermatogonia upon ablation of Ddx5 both in vivo and in vitro. In contrast, Ddx5 appeared dispensable within fibroblasts and its loss potentially compensated for by upregulation of the related helicase DDX17 26 . Previous studies have identified functional redundancy between DDX5 and DDX17 in regulation of splicing, ribosome biogenesis, cell proliferation, and transcriptional co-activation [21][22][23]26,27 . However, studies employing individual knockdown of Ddx5 and Ddx17 have also provided evidence for non-overlapping roles 22 . For example, DDX5 is important for the p53-mediated DNA damage response in cancer cell lines, whereas Ddx17 knockdown had no effect on this process 42 . Although Ddx17 expression is detected in undifferentiated spermatogonia and upregulated at the transcript level upon Ddx5 deletion according to RNA-seq, our findings support a critical role for DDX5 in spermatogonia that evidently cannot be compensated for by DDX17.
Through MS analysis, we have shown that DDX5 interacts with a number of proteins involved in post-transcriptional gene regulatory processes including pre-mRNA splicing, maintenance of mRNA stability, and mRNA export. Although the role of DDX5 in these processes is somewhat appreciated from other a Pathway analysis of differentially expressed genes identified by RNA-sequencing comparing control and Ddx5-ablated cultured spermatogonia. b Differentially expressed cell cycle-related genes identified by RNAsequencing comparing control and Ddx5-ablated cultured spermatogonia. Fold change in Ddx5-ablated spermatogonia shown relative to control. Mean ± SEM shown; n = 4 independent biological replicates per condition. # denotes FDR <0.05. c Representative flow cytometry plots (left) showing Ddx5 loss results in cell cycle arrest in cultured undifferentiated spermatogonia at D1 post-ablation. Control: vehicle-treated control spermatogonia; Ddx5 TAM-KO : tamoxifen-induced Ddx5-ablated spermatogonia. Representative of n = 3 independent experiments. Histogram (right) depicts quantification of cell proportions for each cell cycle phase determined by flow cytometry. *P < 0.05; **P < 0.01. Two-tailed unpaired t-test, n = 3 independent experiments per condition. d Representative western blot showing reduced CCNA2, CCND1 and CCND2 protein upon loss of Ddx5 (Ddx5 TAM-KO ) compared with control in cultured undifferentiated spermatogonia at D2 post-tamoxifen treatment. Representative of n = 3 independent experiments. e Quantification of western blots represented in d. Mean ± SEM shown; *P < 0.05; **P < 0.01; Two-tailed unpaired t-test. n = 4 independent samples per condition. f RNA immunoprecipitation using DDX5 antibody in cultured wildtype undifferentiated spermatogonia followed by RT-qPCR for cyclin genes. Fold enrichment versus IgG control shown. Mean ± SEM shown; n = 2 independent experiments. g Nuclear-cytoplasmic ratio of Ccnd1 and Ccnd2 transcripts determined by RT-qPCR of control and Ddx5-ablated (Ddx5 TAM-KO ) cultured undifferentiated spermatogonia subcellular fractions. Fold change relative to control shown. Mean ± SEM shown; *P < 0.05; Mann-Whitney U test; n = 4 independent samples per condition. h Analysis of mRNA stability over time following inhibition of transcription by actinomycin D treatment in control versus Ddx5-ablated (Ddx5 TAM-KO ) cultured undifferentiated spermatogonia. mRNA expression determined by RT-qPCR (normalised to Actb) relative to initial levels at time 0 h. *P < 0.05; Two-way ANOVA with Bonferroni multiple comparisons test; mean ± SEM shown; n = 4 independent samples per condition systems 21,24,29 , it is well known that cell type-specific post-transcriptional regulation occurs and defects in ubiquitously expressed RNA-binding proteins can lead to disease phenotypes within specific tissues 66 . Here, we have identified an essential posttranscriptional regulator in spermatogonia that acts at multiple stages of RNA processing.
Our results support an essential role for DDX5 in splicing regulation within spermatogonia. It is known that splicing can be regulated in a tissue-specific manner and mediators of alternative splicing within the testis have been described 1,67 . We propose that DDX5 is central to the maintenance of an appropriate splicing programme through interactions with other known splicing regulators in spermatogonia and its loss leads to aberrant splicing of key spermatogenic and cellular genes. These splicing defects are likely to result in severely compromised gene expression or the translation of inappropriate protein isoforms, which may account for the rapid loss of spermatogonia observed upon Ddx5 ablation.
Interestingly, we found the important germ cell genes Dazl and Ehmt2 to be aberrantly spliced in Ddx5-ablated spermatogonia, similar to a recent study that identified a role for BCAS2 in splicing within spermatogonia 20 . Dazl encodes an RNA-binding protein essential for germ cell maintenance and meiotic induction 46 and knockout mice are known to be infertile 48 . Our results indicate that Ddx5 loss results in decreased levels of the Dazl-Δ8 isoform. To our knowledge, a role for Dazl-Δ8 in the male germline has not been defined. However, Dazl-Δ8 binds unique targets in mouse embryonic stem cells; 68 therefore, its loss may have important consequences in spermatogonia that warrants further investigation. Similarly, a specific role for Ehmt2-Δ10 in  male germ cells is unknown, although the role of Ehmt2 as an essential epigenetic regulator during embryogenesis and spermatogenesis is defined 47 . A distinct role for Ehmt2-Δ10 is described in neuronal differentiation 69 , suggesting additional roles in other systems may yet be uncovered.
In addition to Dazl and Ehmt2, we found that DDX5 was important for the splicing of double-strand break (DSB) repair genes Rad18 and Rad50. It is known that DSB repair is pivotal for meiotic homologous recombination 70 and both Rad18 and Rad50 mutant mice are infertile due to spermatogenic defects 49,50 . Rad18 and Rad50 are expressed in spermatogonia and spermatogenic failure in these models is suggested to be due to impaired spermatogonial stem cell maintenance and pre-meiotic defects, respectively 49,50 . Therefore, aberrant splicing of these genes in Ddx5-ablated spermatogonia leading to impaired protein function or expression likely results in deleterious effects. We also found aberrant splicing of Becn1, an important regulator of autophagy 51 . Although a specific role for Becn1 in spermatogonia has not been defined, autophagy is a critical cellular process that is important throughout spermatogenesis during normal and pathological conditions 71 .
Importantly, major spliceosome defects are suggested to be associated with non-obstructive azoospermia in humans and prevent spermatogonial differentiation in Drosophila 72 . Disruption of the pre-mRNA splicing machinery is known to result in p53 activation, p21 upregulation and cell cycle arrest in G1 phase 53 , all of which were observed within Ddx5-ablated spermatogonia in our study. Therefore, the rapid and significant apoptosis of spermatogonia that occurs upon Ddx5 loss may also be due to its role as core spliceosome component. However, we suggest that the negative effects of Ddx5 deletion on cell proliferation are due to defects in specific DDX5-mediated posttranscriptional regulatory mechanisms.
We found that loss of Ddx5 in spermatogonia results in cell cycle arrest at both G1/S and G2/M stages. Several cell cyclerelated genes were aberrantly expressed, including strong upregulation of Cdkn1a (p21), a negative regulator of the G1/S checkpoint. Interestingly, we found a number of cyclin mRNA transcripts bound to DDX5 and showed that nuclear export and stability of a subset of these were regulated by DDX5. Ccnd1 and Ccnd2 transcripts were retained aberrantly in the nucleus upon Ddx5 ablation and protein levels were significantly reduced. It has been shown that RNA-binding proteins can have roles in splicing and mRNA nuclear export of different transcripts within a single cell type that contributes to overall cellular function 56 . Besides their general requirement for G1/S-phase progression 54 , CCND1 and CCND2 are dynamically regulated during spermatogenesis and are predominately expressed in specific subsets of spermatogonia. CCND2 has been shown to be important for long-term maintenance of spermatogonial stem cell self-renewal 73 , whereas Ccnd1 expression is predominantly associated with differentiating subsets of spermatogonia 37 . Therefore, we suggest that impaired spermatogonial function as a result of CCND1 and CCND2 loss may contribute to the overall Ddx5-knockout phenotype in addition to expected effects on G1/S-phase transition. Importantly, we demonstrate that DDX5 is required for efficient nuclear export of these transcripts and maintenance of CCND1 and CCND2 protein in spermatogonia.
We found that DDX5 bound to Ccna2 and Ccnb2 transcripts in spermatogonia and these genes were downregulated upon Ddx5 ablation. Ccna2 is expressed strongly in spermatogonia and is required for progression through G1/S and G2/M checkpoints, whereas CCNB2 functions only at the G2/M transition 54 . The binding of DDX5 to these transcripts and its absence at Ccna2 and Ccnb2 promoters by ChIP-seq coupled with differential expression upon Ddx5 loss suggests a post-transcriptional regulatory role; however, we did not find defects in mRNA export or transcript stability of Ccna2 in Ddx5-ablated spermatogonia. We suggest that Ccna2 downregulation is an indirect transcriptional consequence but nevertheless can contribute to the G2/M arrest observed in spermatogonia upon Ddx5 loss. We cannot exclude that other RNA-binding proteins are involved in maintaining the stability of particular transcripts, either independently or in association with DDX5. For example, we found DDX5 interacts with ELAVL1 in spermatogonia and this protein is known to be important for maintenance of mRNA stability through 3'-UTR-mediated interactions 74 . In contrast, we found that the stability of Ccnb2 mRNA was significantly reduced in Ddx5-ablated spermatogonia and Ccnb2 was strongly downregulated compared with control spermatogonia. Given its central importance in G2/ M transition, we propose that decreased Ccnb2 transcript stability is a main contributor to G2/M cell cycle arrest we observe in Ddx5-ablated spermatogonia.
Notably, we found that DDX5 interacts with ILF3, a posttranscriptional regulator of cyclin mRNA transcripts 57,58,62 , and demonstrated that ILF3 associates with cyclin mRNAs in spermatogonia. While expression of cyclins at the mRNA level appeared unaffected upon Ilf3 deletion, CCND1 protein was substantially reduced and cell growth inhibited. This indicates that ILF3 is involved in post-transcriptional regulation of Ccnd1 and plays a role in cell cycle regulation alongside DDX5. A specific role for ILF3 has not been described in spermatogonia; however, its role in cyclin regulation and cell cycle progression has been shown in other systems, with ILF3 acting independently or in concert with binding partners ILF2 and IGF2BP3 [57][58][59]62 .
Our data indicates that DDX5 is present in a complex with ILF3, ILF2, and IGF2BP3 and that together they bind a number of cyclin mRNA transcripts. However, whether this complex is absolutely required for regulation of expression of these mRNAs remains unclear as independent ablation of Ddx5 and Ilf3 did not produce overlapping effects on gene expression, with the exception of Ccnd1. Both DDX5 and ILF3 evidently regulate Ccnd1 post-transcriptionally, although the role of DDX5 appears more significant. Given findings of this study and previously described functions of DDX5 and ILF3, our observation of increased or unchanged cyclin mRNA expression yet decreased protein levels suggests disruption of mRNA nuclear export or translation mechanisms. Defects in translation of cyclin mRNAs may be unlikely, as both DDX5 and ILF3 appear exclusively nuclear within spermatogonia. However, DDX5 and ILF3 evidently function through different post-transcriptional mechanisms; therefore, further studies are required to define the significance of DDX5-ILF3 interaction and precise post-transcriptional mechanisms by which ILF3 controls cyclin expression in spermatogonia.
Besides its known roles in RNA processing, previous studies have identified DDX5 as a co-factor of transcriptional regulators 21,22 . The ability of DDX5 to interact with PLZF indicated that DDX5 is a transcriptional co-factor for PLZF in spermatogonia. Through ChIP-seq experiments, we found that DDX5 is present at a number of gene promoters overlapping with previously identified PLZF binding sites 16 , including that of Ilf3. Individual loss of Ddx5 and Plzf in spermatogonia resulted in reduced llf3 expression, suggesting a co-ordinated role in regulation of Ilf3 expression. We propose that PLZF/DDX5-dependent expression of Ilf3 is required in part to maintain proliferative capacity of undifferentiated spermatogonia through posttranscriptional regulation of cell cycle-related genes such as Ccnd1. Interestingly, despite robust interaction between PLZF and DDX5 in cultured undifferentiated spermatogonia, DDX5 is only recruited to a small subset of PLZF target genes. Molecular mechanisms mediating selective association of DDX5 with PLZF at specific target genes remain unclear and DDX5 recruitment to PLZF-target gene promoters was not strictly dependent on PLZF, suggesting its recruitment is mediated through other factors. For example, DDX5 can interact with RNA polymerase II and is suggested to function as a link between transcriptional and posttranscriptional processes 75 . Notably, we found RNA polymerase subunits within our list of DDX5 interacting proteins identified by IP-MS, suggesting this mechanism may be conserved in spermatogonia. This suggests that the described association of DDX5 with multiple transcriptional regulators across distinct systems may occur due to a broader role in transcription initiation and/or elongation. Accordingly, DDX5 (in association with DDX17) is important for formation of the transcription initiation complex (including recruitment of RNA polymerase II) at promoters of skeletal muscle-related genes 27 . It is conceivable that this mechanism is conserved in other cell systems and dependent on tissue-specific transcriptional regulators that bind proximal promoter elements. Although PLZF was found unnecessary for recruitment of DDX5 to shared target gene Ilf3, such recruitment may be dependent on other transcriptional regulators that are yet to be characterised. Motif analysis of our DDX5 ChIP-seq dataset suggests that E-box binding proteins are likely candidates and numerous E-box binding motifs are found within the Ilf3 proximal promoter. Nevertheless, our data indicate that PLZF and DDX5 interact at the Ilf3 promoter and are both required for its appropriate expression in spermatogonia. Given that transcription regulation-independent roles for PLZF have been described 76 , it is also possible that the interaction of PLZF with DDX5 is relevant for non-transcriptional PLZF functions. Notably, the majority of genes dysregulated upon loss of Ddx5 in spermatogonia are not bound by either DDX5 or PLZF and are not differentially spliced according to our analyses. This indicates that loss of Ddx5 results in many indirect transcriptional consequences. The even distribution of upregulated and downregulated genes upon Ddx5 loss in SPCs precludes a global defect in mRNA stability.
The TFEB transcription factor is a master regulator of lysosome biogenesis and autophagy and binds target promoters through the CLEAR sequence motif 64 . Our ChIP-seq analysis found DDX5 bound to promoters of multiple TFEB target genes in cultured undifferentiated spermatogonia and the CLEAR sequence element was enriched at DDX5 binding sites. However, we were unable to confirm a specific role for DDX5 as a co-factor of TFEB and transcriptional regulator of autophagy-related genes. One exception could be Amdhd2, a PLZF target gene that was cobound by DDX5 and is a candidate TFEB target 77 . Amdhd2 is involved in amino sugar metabolism but despite potential regulation by TFEB, specific roles in autophagy or lysosome function are unexplored 78 . Moreover, while Amdhd2 was significantly downregulated upon Ddx5 deletion, whether DDX5 promoted Amdhd2 expression by acting as a co-factor of TFEB rather than PLZF remains to be determined. Notably, we found that DDX5 regulates splicing of the autophagy-associated gene Becn1 suggesting DDX5 may play a multifaceted role in autophagy. These findings can form the basis of future studies investigating additional mechanisms of autophagy regulation.
Considering our data demonstrating the role of DDX5 in splicing, mRNA export and transcript stability, as well as recruitment to multiple gene promoters, we propose a model whereby DDX5 maintains function of undifferentiated spermatogonia by regulating gene expression programmes through multiple distinct mechanisms (Fig. 8). Specifically, that DDX5 plays essential roles in post-transcriptional gene regulation as a trans-acting mediator of splicing and through regulation of mRNA stability and export independently and via interactions with other RNA-binding proteins such as ILF3. Furthermore, we propose that DDX5 and PLZF co-regulate transcription of Ilf3, thereby maintaining cell cycle progression through downstream effects on cyclin mRNA transcripts (Fig. 8). However, given that DDX5 recruitment to the Ilf3 promoter does not appear to be dependent on PLZF, further study is required to confirm significance of the DDX5-PLZF interaction or identify other transcriptional regulators mediating DDX5 recruitment.
In conclusion, our study provides evidence of an essential role for Ddx5 during in vivo spermatogenesis and in maintenance of spermatogonia. Moreover, through a combination of MS and ChIP-seq approaches, we provide evidence for undefined transcriptional roles and identify DDX5 molecular targets. Altogether, our findings present DDX5 as a multidimensional regulator of gene expression in spermatogonia and an essential factor for male fertility. Given that mammalian spermatogenesis is an ideal model for the study of adult stem cells, an understanding of the important factors and mechanisms governing this process may have implications beyond reproductive biology.

Methods
Mouse maintenance and treatment. Ddx5 flox/flox and Plzf -/mice were established in previous studies 11,39 . UBC-Cre ERT2 mice 38 were obtained from the Jackson Laboratory. For in vivo gene deletion experiments, 8-week-old to 12-weekold male mice were injected with 2 mg tamoxifen (Sigma-Aldrich) dissolved in a 90% sesame oil (Sigma-Aldrich)/10% ethanol vehicle solution for 2 consecutive days 12 . All animal experimentation was conducted in accordance with institutional ethical requirements and approved by the Monash University Animal Ethics Committee.
Whole mount and frozen section immunofluorescence. For immunofluorescence staining of whole mount samples, testes were detunicated and seminiferous tubules washed in ice cold phosphate-buffered saline (PBS). Tubules were fixed with 4% paraformaldehyde (PFA) for 6 h and washed with PBS prior to blocking with 10% v/v foetal bovine serum (FBS) (In Vitro Technologies), 2% w/v bovine serum albumin (BSA) (Sigma-Aldrich) and 0.3% v/v Triton X-100 (Sigma-Aldrich) in PBS. Tubules were incubated overnight at 4°C with primary antibodies diluted in PBS supplemented with 0.3% v/v Triton X-100 and 1% w/v BSA. Tubules were washed with PBS supplemented with 0.3% v/v Triton X-100. Tubules were then incubated for one hour at room temperature with Alexa Fluor-conjugated secondary antibodies diluted in PBS supplemented with 0.3% v/v Triton X-100 and 1% w/v BSA.
Cell culture and immunofluorescence. Murine undifferentiated spermatogonia cultures were established from 6-week-old mice following enzymatic digestion of whole testis extracts with collagenase and trypsin 12 . Undifferentiated spermatogonia were cultured on mitomycin-inactivated mouse embryonic fibroblast (MEF) feeder cells and maintained in StemPro-34 medium (Thermo Fisher Scientific) supplemented with 10 ng/ml GDNF, 10 ng/ml bFGF, 20 ng/ml EGF, 25 μg/ml insulin plus other additives 13 . For culture of Ddx5 flox/flox ; UBC-Cre ERT2 MEFs, embryos were collected at E13.5 and MEFs isolated according to a standard protocol 80 . MEFs were maintained in DMEM (Thermo Fisher Scientific) supplemented with 10% v/v FBS, 1X non-essential amino acids (Thermo Fisher Scientific), 1× L-glutamine (Thermo Fisher Scientific) and 1× Penicillin/Streptomycin (Thermo Fisher Scientific). For gene knockout experiments, cells were treated with medium supplemented with 0.2 μM 4-hydroxytamoxifen (4-OHT) (Sigma-Aldrich) for four consecutive days. Following treatment, 4-OHTsupplemented medium was replaced with standard culture medium until the required collection time point. For subsequent analyses, cells were collected by dissociation with 0.5% v/v trypsin-EDTA (Thermo Fisher Scientific).
Immunofluorescence on cultured cells was performed using LabTek chamber slides coated with Geltrex Matrix (Thermo Fisher Scientific) 12  Apoptosis inhibitor experiments. Undifferentiated spermatogonia were cultured and treated with 0.2 μM 4-OHT as described above (see Cell culture and immunofluorescence). To inhibit apoptosis, Z-VAD(OMe)-FMK (Z-VAD-FMK) (Cell Signalling Technology) reconstituted in DMSO was added at a final concentration of 50 μM in conjunction with 4-OHT treatment. An equivalent amount of DMSO was used as a vehicle control. Following 4-OHT treatment, medium was replaced with standard culture medium supplemented with 50 μM Z-VAD-FMK until collection for further analyses or fixation for immunofluorescence.
Mass spectrometry and gene ontology analysis. Lysates from cultured undifferentiated spermatogonia were prepared using lysis buffer containing 50 mM Tris-HCl pH 8.0, 150 mM NaCl, 5 mM EDTA pH 8.0, 0.5% v/v Nonidet P-40 (Sigma-Aldrich), 1 mM dithiothreitol (Sigma-Aldrich), and protease inhibitor cocktail (Roche) 76 . DDX5 and PLZF complexes were immunoprecipitated with goat anti-DDX5 antibody (Abcam, #ab10261) and goat anti-PLZF antibody (R&D Systems, #AF2944), respectively, coupled with Dynabeads using a Dynabead Coupling Kit (Thermo Fisher Scientific) according to manufacturer's instructions. Elution and preparation of samples for mass spectrometry analysis was performed as described previously at the Bio21 Proteomics Facility 12 . Gene ontology analysis was performed on mass spectrometry data with PANTHER 43 v13.0 using PANTHER Overrepresentation Test (Fisher's Exact with FDR multiple test correction, FDR <0.05) and complete GO annotations for molecular function and biological process.
RNA extraction and RT-qPCR. RNA was isolated using TRIzol reagent (Thermo Fisher Scientific) and purified using a Direct-zol MiniPrep kit (Zymo Research) according to the manufacturer's instructions, including treatment with DNase I. cDNA was synthesised using Tetro cDNA synthesis kit (Bioline). qPCR was performed using SYBR Premix Ex Taq II (Takara) and run on a Light Cycler 480 (Roche) or Mic qPCR Cycler (Bio Molecular Systems). Primer sequences used were obtained from the Harvard PCR Primer Bank (http://pga.mgh.harvard.edu/ primerbank/) and are included in Supplementary Table 1. Differences between conditions were calculated using the ΔΔCt method.
RNA-sequencing and analysis. RNA was isolated from FACS-purified cultured spermatogonia as described above. RNA quality was evaluated using Bioanalyzer and RNA quantity was determined using Qubit. Samples were prepared starting with 1 μg of total RNA and libraries generated according to the Illumina Stranded mRNA protocol using 12 cycles of amplification. Library size (~350 bp) was checked using Bioanalyzer and quantified with Qubit and qPCR. Based on qPCR results, a single equimolar pool was prepared and denatured with a final mRNA stability Splicing  84,85 . Genes with a false discovery rate (FDR) <0.05 were considered to be differentially expressed. Pathway analysis was performed with Ingenuity Pathway Analysis (IPA) (QIAGEN Inc., https://www. qiagenbioinformatics.com/products/ingenuity-pathway-analysis) using all differentially expressed genes (FDR <0.05). Differential splicing analysis was performed using the MISO v0.5.4 probabilistic framework 45 (in the configuration where filter_results = TRUE) and the mm10 alternative event annotation (v2.0). For each alternative splicing event type, pooled treatment replicates were compared with combined control replicates using miso_compare and results filtered using filter_events (num-inc 1, num-exc 1, numsum-inc-exc 10, delta-psi 0.20, bayes-factor 10). PCR validation of differential splicing analysis. RNA was extracted from vehicle-treated and TAM-treated Ddx5 TAM-KO cultured undifferentiated spermatogonia and cDNA synthesised as described above (see Cell culture and immunofluorescence and Apoptosis inhibitor experiments). Primers for validation of differentially spliced exons detected by MISO were designed using Primer3 (http:// primer3.ut.ee). Primers were designed within constitutive exons flanking the exon of interest. Standard PCR for analysis by gel electrophoresis was performed using JumpStart REDTaq ReadyMix Reaction Mix (Sigma-Aldrich) according to the manufacturers' instructions and visualised by running on a 2% agarose gel. Primer sequences used are found in Supplementary Table 1. Uncropped gels are presented in Supplementary Fig. 12.
Nuclear-cytoplasmic fractionation. Vehicle-treated and TAM-treated Ddx5 TAM-KO cultured undifferentiated spermatogonia were treated with Z-VAD-FMK as described above (see Cell culture and immunofluorescence and Apoptosis inhibitor experiments). Following dissociation with trypsin, cells were feeder depleted according to differential adhesion to cell culture dishes. For nuclear-cytoplasmic fractionation and RNA extraction, dissociated spermatogonia were pelleted by centrifugation and resuspended in PBS once to wash, then centrifuged again and supernatant removed. Cells were lysed by resuspension in hypotonic lysis buffer containing 10 mM HEPES pH 7.4 (Thermo Fisher Scientific), 10 mM NaCl, 3 mM MgCl 2 and 0.313% IGEPAL CA-630 (Sigma-Aldrich) in DNase/RNase-free deionised water. Samples were incubated for 5 min on ice and centrifuged for 8 min at 2000 × g at 4°C. The supernatant containing the cytoplasmic fraction was collected and the nuclear pellet was washed once in hypotonic lysis buffer before centrifuging again for 5 min at 2000 × g at 4°C. The supernatant was removed and resulting nuclear pellet and previously collected cytoplasmic fraction were processed for RNA extraction as described above (see RNA extraction and RT-qPCR). Nuclear-cytoplasmic ratios were calculated using the ΔΔCt method. mRNA stability assays. Vehicle-treated and TAM-treated Ddx5 TAM-KO cultured undifferentiated spermatogonia were treated with Z-VAD-FMK as described above (see Cell culture and immunofluorescence and Apoptosis inhibitor experiments). For mRNA stability assays, cells were treated with 10 μM actinomycin-D (Sigma-Aldrich) for up to four hours, with cell samples collected at 0, 2, and 4 h posttreatment. At each time point and following dissociation with trypsin, cells were feeder depleted according to differential adhesion to cell culture dishes prior to RNA extraction, cDNA synthesis and RT-qPCR analysis as described above (see RNA extraction and RT-qPCR).
Cloning of Ddx5 constructs and lentiviral transduction. DDX5 wildtype and DDX5 helicase mutant constructs were generated and described in a previous study by Bates et al. 42 . DDX5 constructs were subcloned into a PCCL-hPGK vector 86 modified to express a tdTomato fluorescent protein tag. First, the P2A-tdTomato sequence from pCDH1-EF1-Luc2-P2A-tdTomato vector (a gift from Kazuhiro Oka, Addgene plasmid #72486) was amplified by PCR using complimentary primers that included a BglII, BamHI and SalI recognition site at the 5′ end and XhoI recognition site at the 3′ end (F: 5′-GAA GAT CTG GAT CCG TCG ACG GAA GCG GAG CTA CTA A-3′; R: 5′-CGC CGC TCG AGT TAT TAC TTG TAC AGC TCG TC-3′). The PCR product was cloned by BglII/XhoI restriction digestion and ligation into BamHI/SalI sites of the PCCL-hPGK vector. As a result, original BamHI and SalI sites present in the PCCL-hPGK vector were destroyed and new BamHI and SalI sites created preceding P2A-tdTomato (new vector referred to as PCCL-hPGK-P2A-tdTomato). DDX5 constructs were then amplified by PCR using complimentary primers that included a BglII recognition site at the 5' end and SalI site at the 3′ end (F: 5′-GGA AGA TCT GCC ATG GAA CAA AAA CTC ATC TCA G-3′; R: 5′-GCC GAC GTC GAC TTG GGA ATA TCC TGT TGG C-3′). The resulting PCR product was cloned by BglII/SalI restriction digest and ligation into the BamHI/SalI sites of PCCL-hPGK-P2A-tdTomato vector. All PCR, digestion and ligation reactions were performed using appropriate commercially available kits (New England Biolabs). PCR product purification was performed using PureLink PCR Purification kit (Thermo Fisher Scientific). All plasmid transformations were performed using One Shot TOP10 chemically competent E. coli according to manufacturer's instructions (Thermo Fisher Scientific) and colonies were selected based on ampicillin resistance. All DNA purification was performed using PureLink HiPure Plasmid Midiprep kits (Thermo Fisher Scientific). Sanger sequencing was performed to confirm successful cloning of each construct. For lentivirus production, HEK293T cells were transfected with cloned DDX5 constructs (10 µg) and packaging plasmids pMDL (5 µg), pRSV-rev (2.5 µg) and pVSV-G (3 µg) using Lipofectamine-2000 (Thermo Fisher Scientific). Lentiviral-containing supernatant was collected in spermatogonia culture medium 48 and 72 h after transfection. Undifferentiated spermatogonia cultures were infected with lentiviral-containing supernatant supplemented with 8 μg/ml polybrene (Millipore). Successfully infected cells were selected by FACS according to expression of tdTomato (see Fluorescence-activated cell sorting purification) and re-plated for subsequent analyses. Cells infected with PCCL-hPGK-tdTomato were used for control experiments.
ChIP-sequencing and motif analysis. Chromatin immunoprecipitation was performed using digested chromatin from two independently derived lines of cultured wildtype undifferentiated spermatogonia prepared with SimpleChIP Enzymatic Chromatin IP Kit (Cell Signalling Technology #9003) 12,14 . The antibody used for immunoprecipitation was: goat anti-DDX5 (Abcam ab10261 Lot# GR183991-1). Samples were quantified by Qubit and 20 ng of DNA was used as starting amounts for all libraries. Samples were sheared by Covaris to~200-300 bp prior to library generation. Libraries were constructed using Ovation Ultralow System V2 using Nugen protocol M01379v1, 2014. Libraries were initially quantified by Qubit and size profile determined using Bioanalyzer followed by quantification by qPCR and pooled in equimolar ratio. Upon qPCR, 200 pM of a library pool was used for cluster generation with c-bot (Ilumina Protocol 15006165 v02 Jan 2016). Two lanes of Illumina HiSeq3000 (Ilumina Protocol 15066493 Rev A, February 2015) were used for 50 bp single end reads. ChIP-sequencing was performed at the Medical Genomics Facility, Monash Health Translation Precinct (MHTP) and data was processed by the Monash University Bioinformatics Platform. Raw reads from the ChIP-seq libraries were mapped to the UCSC mouse genome (mm10) using the bowtie2 tool. Good alignment rates were obtained across all libraries with over 97% mapped reads. Unmapped and low mapping quality reads were filtered out using samtools. ChIP-seq analysis was performed using the HOMER tool (http://homer. ucsd.edu/homer/). The first step of the HOMER ChIP-seq analysis creates a tag directory for each ChIP-seq library that generates quality control matrices and bedGraph files. Peak identification was performed with HOMER's findPeaks function using histone parameters. Peak annotation of individual samples and peaks common between 2 samples was performed with HOMER annotatePeaks.pl. Peak annotation determines the distance to the nearest Transcription Start Site (TSS) and assigns the peak to that gene followed by genomic annotation of the region occupied by the centre of the peak. The list of peaks and associated genes are listed in Supplementary Data 8. Data is deposited in the GEO repository (GSE110347). Motif analysis was performed on ChIP-seq data using the RSAT peak-motifs tool (https://rsat01.biologie.ens.fr/rsat/peak-motifs_form.cgi).
Plzf knockdown and ChIP-qPCR. Enzymatically dissociated wildtype undifferentiated spermatogonia were seeded on to a mitotically-inactivated MEF feeder layer at 1 × 10 6 cells per well in a 6-well tissue culture plate with 2 ml final volume of culture medium (see Cell culture and immunofluorescence). siRNA knockdown was performed immediately following passage using ON-TARGETplus Mouse Zbtb16 siRNA SMARTpool (Horizon Dharmacon, #L-040219-01-0005) or control ON-TARGETplus Non-targeting Pool (Horizon Dharmacon, #D-001810-10-05). For transfection (per well), 150 μl Opti-MEM Reduced Serum Media (Thermo Fisher Scientific) was combined with 9 μl Lipofectamine RNAiMAX Reagent (Thermo Fisher Scientific) in one tube and 150 μl Opti-MEM Reduced Serum Media (Thermo Fisher Scientific) was combined with 60 pmol siRNA in a separate tube. Tubes were combined and incubated for 5 min at room temperature to form siRNA-lipid transfection complexes. After 5 min, 300 μl transfection complex mixture was added to each well. Cells were incubated for 18 h at 37°C, 5% CO 2 then culture medium was replaced. Cells were incubated for an additional 24 h before collection for ChIP experiments. 4 × 10 6 cells were used for each ChIP. Knockdown efficiency was confirmed by western blot. Chromatin immunoprecipitation was performed using digested chromatin prepared using SimpleChIP Enzymatic Chromatin IP Kit (Cell Signalling Technology #9003) 12,14 . The antibody used for immunoprecipitation was: goat anti-DDX5 (Abcam ab10261 Lot# GR183991-1). Purified eluted DNA was used for subsequent qPCR analysis using primers designed within peak regions identified by HOMER as described above (see ChIP-sequencing and motif analysis), and control primers designed within intergenic regions upto 2 kb upstream of identified peaks. Primers were designed using Primer3 and are listed in Supplementary Table 2. RNA immunoprecipitation. Cultured wildtype murine undifferentiated spermatogonia were used for all RNA immunoprecipitation (RNA-IP) experiments. Cells were cultured and collected as described above (see Cell culture and immunofluorescence). RNA immunoprecipitation was performed using an Imprint RNA Immunoprecipitation Kit (Sigma-Aldrich) according to the manufacturer's insructions. Five micrograms of antibody or rabbit IgG was used for immunoprecipitation. Antibodies used were: rabbit anti-DDX5 (Abcam, #ab21696) and rabbit anti-ILF3 (Abcam, #ab92355). RNA extraction was performed following RNA-IP using TRIZol (Thermo Fisher Scientific) and purified with Direct-zol Miniprep Kit (Zymo Research). cDNA synthesis was performed on total extracted RNA using Tetro cDNA Synthesis Kit (Bioline). qPCR was performed using SYBR Premix Ex Taq II (Takara) and run on a Mic qPCR Cycler (Bio Molecular Systems). Using cycle threshold values, fold enrichment versus IgG only pull-down was calculated for candidate genes. Standard PCR for analysis by gel electrophoresis was performed as described above (see PCR validation of differential splicing analysis). Primer sequences used are found in Supplementary Table 1. Uncropped gels are presented in Supplementary Fig. 12.
Generation of inducible Ilf3 −/− spermatogonia by CRISPR/Cas9. CRISPR/ Cas9 targeting mouse Ilf3 was performed using the inducible lentiviral system described by Aubrey et al. 63 . sgRNAs to mouse Ilf3 were designed by CCTop software (https://crispr.cos.uni-heidelberg.de/). One sgRNA (5′-TCCCGTTACTGCCGTCTACCTCCA-3′) targeting an exonic sequence common to all known transcript variants of Ilf3 was selected. Oligonucleotides containing this sgRNA sequence were synthesised (Integrated DNA Technologies) including a 5' "TCCC" 4 bp overhang for the complementary sequence and a ′5 "AAAC" 4 bp overhang for the reverse complementary sequence. Annealed oligonucleotides were cloned into BsmBI restriction site of the lentiviral vector for doxycycline (DOX)inducible sgRNA expression (FgH1tUTG-eGFP expression) provided by Dr. Marco Herold. Sanger sequencing was performed to ensure sgRNA sequence was ligated into FgH1tUTG. Lentiviral production and transduction of cells were performed as described above. Cultured wildtype spermatogonia were initially infected with FUCas9mCherry and subsequently infected with the sgRNA construct. Cells were then sorted with BD Influx Cell Sorter (BD Biosciences) for mCherry (Cas9) and eGFP (sgRNA). Expression of sgRNA was induced by treatment with 2 μg/ml of doxycycline hyclate (Sigma-Aldrich D9891). Knockout of Ilf3 in spermatogonia was confirmed by immunofluorescence and qRT-PCR.
Statistical analyses. Assessment of statistical significance was performed using two-tailed unpaired t-tests, one-way ANOVA with Tukey multiple comparisons test or two-way ANOVA with Bonferroni multiple comparisons test for normally distributed data and two-tailed Mann-Whitney U tests or Kruskal-Wallis with Dunn's multiple comparisons tests for data that were not normally distributed. P-values less than 0.05 were considered statistically significant. All statistical analyses including normality tests were performed using GraphPad Prism v7. For mouse experiments, no statistical method was used to predetermine sample sizes and no specific randomisation or blinding methods were used.

Data availability
PLZF and DDX5 mass spectrometry data have been deposited in the MassIVE repository under accession code MSV000083476 [https://doi.org/10.25345/C57P6V]. DDX5 RNAseq data has been deposited in the Gene Expression Omnibus (GEO) database under accession code GSE127275 and DDX5 ChIP-seq data has been deposited in GEO under accession code GSE110347. All data are available within the article or Supplementary Information or from corresponding author upon reasonable request.