GILZ-dependent modulation of mTORC1 regulates spermatogonial maintenance

Sustained production of spermatozoa is dependent on mitotic germ cells with self-renewal potential known as spermatogonial stem cells (SSCs) (de Rooij, 1998; Kanatsu-Shinohara and Shinohara, 2013). Murine spermatogonia are divided into undifferentiated plus differentiating fractions; SSCs represent a subset of the undifferentiated pool whereas remaining undifferentiated cells act as committed progenitors. Undifferentiated spermatogonia are generated postnatally from foetal germ cells (gonocytes or prospermatogonia) upon migration to the basement membrane of the seminiferous cords. The undifferentiated population contains isolated spermatogonia (A-single or A_s) plus chains of cells interconnected by cytoplasmic bridges. Two-cell chains are known as A-paired (A_pr) whereas chains of four or more cells are known as A-aligned (A_al). Lineage-tracing studies demonstrate that SSCs are marked by GFRα1 and typically A_s and A_pr, whereas most undifferentiated cells, particularly A_al, are committed progenitors and marked by NGN3 (Hara et al., 2014; Nakagawa et al., 2010). NGN3⁺ A_al may revert to SSCs through chain fragmentation, particularly upon tissue damage (Nakagawa et al., 2010). Immunofluorescence demonstrates that RARγ and SOX3 preferentially mark GFRα1⁻ progenitors, although they are detectable at lower levels in SSC-containing fractions (Ikami et al., 2015; Lord et al., 2018; Suzuki et al., 2012). Although transplantation indicates that SSCs are also found in the GFRα1⁻ undifferentiated population (Garbuzov et al., 2018), progenitors of homeostatic testis can exhibit transplantation capacity (Carrieri et al., 2017; Nakagawa et al., 2007). ID4 marks cells with potent transplantation capacity and is primarily restricted to the GFRα1⁺ population (Chan et al., 2014; Helsel et al., 2017), supporting the theory that GFRα1 marks an SSC-enriched fraction.

Spermatogonial differentiation is marked by c-KIT induction and formation of A₁ spermatogonia that undergo mitotic divisions and via A₂, A₃ and A₄, Intermediate (In) and B-type spermatogonia generate meiotic spermatocytes (de Rooij, 1998). Undifferentiated and differentiating spermatogonia are both present within the seminiferous epithelium basal layer. Spermatogenesis is a cyclic process and the seminiferous epithelium can be divided into 12 stages in the mouse (I-XII). Tubules at a given stage contain spermatogonia at a specific differentiation step (de Rooij, 1998). Undifferentiated cells are present at all cycle stages but mitotic activity and the relative proportions of SSCs and progenitors vary.

GDNF is a growth factor produced by Sertoli and peritubular myoid cells within the testis that promotes SSC self-renewal via the GFRα1/RET receptor (Chen et al., 2016; Kanatsu-Shinohara and Shinohara, 2013). Basic fibroblast growth factor (bFGF) is produced by germ and somatic cells within the testis, and independently promotes SSC self-renewal (Ishii et al., 2012; Takashima et al., 2015). Undifferentiated spermatogonia can be cultured in vitro in the presence of GDNF and bFGF (Kanatsu-Shinohara and Shinohara, 2013). Signalling pathways mediating the effects of these growth factors have been elucidated, e.g. the extracellular-regulated kinase (ERK) mitogen-activated protein kinase (MAPK) and phosphoinositide 3-kinase (PI3K)/AKT (Hasegawa et al., 2013; Ishii et al., 2012; Lee et al., 2007; Oatley et al., 2007; Takashima et al., 2015). RAS activation drives SSC self-renewal via induction of cyclins D2 plus E1, and is upstream ERK MAPK and PI3K/AKT (Lee et al., 2009). Aberrant activation of growth factor-dependent signalling can, however, be detrimental. Ablation of phosphatase and tensin homolog (PTEN), a negative regulator of PI3K/AKT, drives SSC exhaustion (Goertz et al., 2011). That growth factor signalling is tightly regulated in SSCs is indicated by the key role of transcription factor FOXO1 in SSC function (Goertz et al., 2011); phosphorylation of FOXO1 by AKT inhibits nuclear localisation and activity (Calnan and Brunet, 2008).

The transcription factor promyelocytic leukaemia zinc finger (PLZF) is expressed by undifferentiated spermatogonia plus early differentiating cells and promotes SSC self-renewal (Buaas et al., 2004; Costoya et al., 2004; Hobbs et al., 2010). SALL4 is expressed in undifferentiated and differentiating spermatogonia, and is crucial for SSC activity plus spermatogonial differentiation (Chan et al., 2017; Hobbs et al., 2012). PLZF and SALL4 modulate responses of undifferentiated spermatogonia to niche factors by regulating pathway components (Chan et al., 2017; Hobbs et al., 2010). PLZF enhances sensitivity of undifferentiated spermatogonia to GDNF through mammalian target of rapamycin complex 1 (mTORC1) inhibition (Hobbs et al., 2010). mTORC1 promotes cell growth and plays crucial roles in stem cell regulation (Laplante and Sabatini, 2012). mTORC1 activity is higher in progenitors than SSCs and aberrant mTORC1 activation promotes SSC exhaustion, indicating a role in differentiation (Hobbs et al., 2015; Wang et al., 2016). Deleting the mTOR kinase blocks spermatogonial proliferation and differentiation, consistent with roles of mTORC1 (Serra et al., 2017). mTORC1 promotes translation of factors required for differentiation, including SOHLH1 and SOHLH2 (Busada et al., 2015).

Appropriate cell signalling is of fundamental importance to SSCs, although signalling modulators in SSCs are poorly characterised. One factor essential for spermatogenesis and implicated in signalling regulation is glucocorticoid-induced leucine zipper (GILZ or TSC22D3). Identified in a screen for genes activated in lymphocytes in response to glucocorticoids, GILZ belongs to the TGFβ-stimulated clone 22 domain (TSC22D) family, of which there are four members (TSC22D1-TSC22D4) (Beaulieu and Morand, 2011; D'Adamio et al., 1997; Kester et al., 1999). TSC22D proteins contain a tuberous sclerosis complex box (TSC) and leucine zipper (LZ) domain. Gilz-null mice demonstrate enhanced T cell responses and haematopoietic defects associated with the ability of GILZ to regulate pathways involved in immune cell activation (Beaulieu and Morand, 2011; Ngo et al., 2013a; Suarez et al., 2012). GILZ inhibits NF-κB by direct binding and mediates anti-proliferative effects of glucocorticoids via interaction with RAS and inhibition of ERK MAPK and PI3K/AKT (Ayroldi et al., 2001; Ayroldi et al., 2007; Bruscoli et al., 2012).

In Gilz-null males, meiotic progression is blocked and spermatozoa are absent (Bruscoli et al., 2012; Ngo et al., 2013b; Romero et al., 2012; Suarez et al., 2012). Spermatogonia are also depleted by 1-2 months of age, indicating defective SSC function (Bruscoli et al., 2012; Ngo et al., 2013b). A study of Gilz-deficient neonatal spermatogonia suggested enhanced growth factor signalling, consistent with the RAS-inhibitory role of GILZ and effects of aberrant signalling on SSCs (Bruscoli et al., 2012). However, GILZ loss caused aberrant FOXO1 activation in neonatal spermatogonia, resulting in a differentiation block and apoptosis, in line with the ability of GILZ to promote FOXO nuclear exclusion (Latré de Laté et al., 2010; Ngo et al., 2013b). Assuming RAS activation upon GILZ loss (Bruscoli et al., 2012), FOXO1 should be inactivated via PI3K/AKT (Calnan and Brunet, 2008), casting doubt on mechanisms of GILZ-dependent SSC regulation. Furthermore, GILZ function in SSCs has been inferred from constitutive knockout mice or upon Gilz deletion in embryonic germ cells (Bruscoli et al., 2012; Ngo et al., 2013b; Romero et al., 2012). In these models, GILZ is absent during SSC development and a potential role for GILZ in establishing the SSC pool may underlie germline failure (Ngo et al., 2013b; Romero et al., 2012).

To define GILZ function in mature SSCs and characterise effectors, we developed an inducible knockout model allowing GILZ ablation in adults. We identified an essential role for GILZ in SSC maintenance through mTORC1 inhibition. GILZ also promoted expression of spermatogenic regulators independently of mTORC1 control. Analysis of GILZ-associated proteins indicated crosstalk between GILZ and other TSC22D proteins in undifferentiated spermatogonia.

Gilz is essential for spermatogenesis but its role in SSCs poorly understood (Ngo et al., 2013b; Romero et al., 2012; Suarez et al., 2012). Whole-mount immunofluorescence of seminiferous tubules from adult wild-type mice demonstrated that GILZ was present in the cytosol of undifferentiated and differentiating spermatogonia (Fig. 1A,B). GILZ was detected in GFRα1⁺ A_s and A_pr plus SOX3⁺ A_al, indicating expression in SSCs and progenitors (Chan et al., 2017; Hara et al., 2014; Suzuki et al., 2012). From Id4^IRES-GFP adult testis, over 85% of GFRα1⁺ but less than 10% of SOX3⁺ cells were ID4⁺ (Fig. S1A,B), supporting use of GFRα1 and SOX3 as stem and progenitor markers, respectively (Best et al., 2014; Helsel et al., 2017). Id4 expression overlapped poorly with RARγ, an alternative marker of progenitors plus early differentiating cells, and was absent from c-KIT⁺ differentiating spermatogonia (Fig. S1A) (Gely-Pernot et al., 2012; Ikami et al., 2015). Notably, GILZ levels in differentiating spermatogonia were lower than those in undifferentiated cells (Fig. 1A,B and Fig. S2A). Reduction in Gilz expression during spermatogonial differentiation was evident by flow cytometry of immunostained testis cell suspensions (Fig. S2B-E). We also confirmed Gilz expression in spermatocytes (Fig. S2F), consistent with roles in mitotic and meiotic germ cells.

To characterise the role of GILZ in mature spermatogonia, we developed an inducible Gilz knockout model (Gilz^TAM-KO) by crossing floxed Gilz mice with a transgenic line expressing tamoxifen (TAM)-regulated Cre from the UBC promoter (UBC-CreER) (Ruzankina et al., 2007). UBC-CreER is active in spermatogonia but not in supporting somatic cells within the testis (Chan et al., 2017). Although UBC-CreER is expressed in multiple tissues (Ruzankina et al., 2007), Gilz function in the germline is cell-autonomous and body-wide Gilz deletion is well tolerated (Bruscoli et al., 2012; Ngo et al., 2013b; Suarez et al., 2012).

Control and Gilz^TAM-KO males were treated with TAM and harvested at multiple timepoints to assess testis phenotype (Fig. 1C). By 7 days post-TAM (D7), Gilz expression was almost entirely lost within PLZF⁺ spermatogonia of Gilz^TAM-KO mice, indicating Gilz deletion in SSCs and progenitors (Fig. 1C). Gilz was also deleted in differentiating spermatogonia (Fig. 1C). Immunofluorescence of Gilz^TAM-KO testis sections showed that populations of PLZF⁺ spermatogonia and VASA⁺ spermatocytes/spermatids appeared normal at D7 (Fig. 1D). However, VASA⁺ germ cells were depleted by D14 and at D28 were absent (Fig. 1D and Fig. S3A), consistent with spermatogenic failure (Bruscoli et al., 2012; Ngo et al., 2013b; Romero et al., 2012). Accordingly, the testis to body weight ratio of Gilz^TAM-KO mice was unaffected at D7 but progressively reduced by D14 and D28 (Fig. S3B). In contrast to rapid meiotic cell depletion after Gilz deletion, PLZF⁺ spermatogonia were abundant at D14 and only depleted by D28 (Fig. 1D,E and Fig. S3A). Using immunofluorescence of Gilz^TAM-KO tubules, we confirmed that almost no SALL4⁺ spermatogonia, including GFRα1⁺ cells, remained at D28 (Fig. S3C). Occasionally, we found small dense clusters of Gilz-deleted SALL4⁺ GFRα1⁻ spermatogonia at D28 (Fig. S3C). Analysis of sections 2 months after Gilz deletion revealed similar lack of germ cells, although some tubules contained VASA⁺ GILZ⁺ cells, indicating that rare Gilz-retaining SSCs repopulated tubules (Fig. S3D). Immunofluorescence for the Sertoli cell marker SOX9 confirmed that most Gilz^TAM-KO tubules at 2 months contained only Sertoli cells (Fig. S3E).

Progressive germ cell loss following Gilz deletion indicated that GILZ was essential for SSC function. To characterise the effects of GILZ loss on SSCs, we performed whole-mount immunofluorescence for the SSC-associated marker GFRα1 plus the spermatogonial marker SALL4 on Gilz^TAM-KO tubules prior to spermatogonial depletion. Although SALL4⁺ spermatogonia were abundant at D7, the number of GFRα1⁺ cells (A_s, A_pr and short A_al) was drastically reduced (Fig. 1F,G and Fig. S3F,G). Isolated A_s and interconnected A_pr plus A_al were identified based on the presence or absence of cell-cell contact, nuclear morphology and physical proximity (Fig. S4A) (de Rooij and Russell, 2000). We confirmed the presence of the intercellular bridge component CEP55 between member cells of A_pr and A_al, supporting these criteria (Fig. S4B) (Iwamori et al., 2010). Although control mice were also treated with TAM, we confirmed that TAM alone did not affect abundance of GFRα1⁺ A_s, A_pr and A_al in wild-type adults (Fig. S5A-C). We concluded that acute Gilz deletion in adults resulted in SSC exhaustion and germline degeneration.

Despite evidence indicating rapid SSC depletion following GILZ loss, PLZF⁺ spermatogonia persisted up to 2 weeks after gene deletion (Fig. 1D,E). As the PLZF⁺ pool includes SSCs, progenitors and spermatogonia at early differentiation stages (Chan et al., 2017; Hobbs et al., 2010), we confirmed the identity of persistent spermatogonia by whole-mount immunofluorescence for progenitor-associated (RARγ) and differentiating (c-KIT) cell markers (Fig. 1B) (Ikami et al., 2015; Schrans-Stassen et al., 1999; Suzuki et al., 2012). RARγ⁺ c-KIT− A_al and c-KIT⁺ spermatogonia were abundant in D7 Gilz^TAM-KO tubules, indicating that progenitors and differentiating cells tolerated GILZ loss (Fig. 2A). Given cycling of the seminiferous epithelium (de Rooij, 1998), RARγ⁺ progenitors in early stage tubules D7 post-TAM should become c-KIT⁺ differentiating cells by D14 and c-KIT+ spermatogonia would generate meiotic spermatocytes. New progenitor cohorts would not be formed because of SSC depletion. RARγ⁺ spermatogonia were present in D14 Gilz^TAM-KO tubules whereas c-KIT⁺ populations were depleted (Fig. 2A), suggesting that Gilz deletion blocks progenitor differentiation, preventing formation of a new round of c-KIT⁺ spermatogonia. RARγ⁺ cells in D14 Gilz^TAM-KO tubules formed unusual dense clusters, supporting defective progenitor function (Fig. 2A). Although RARγ⁺ spermatogonia of controls were typically negative for the proliferation marker KI67 in early and mid-stage tubules, KI67⁺ RARγ⁺ cells were found in equivalent stages of D7 Gilz^TAM-KO tubules (Fig. 2A). By D14, Gilz-deleted RARγ⁺ spermatogonia were KI67^low (Fig. 2A), suggesting dynamic cell cycle changes.

To confirm our observations, we performed immunofluorescence for SOX3, which preferentially marks GFRα1⁻ progenitors (Suzuki et al., 2012). SOX3⁺ A_al were abundant in Gilz^TAM-KO tubules D7 and D14 post-TAM (Fig. 2B). Clusters of SOX3⁺ spermatogonia were also evident at D14. Quantification of SOX3⁺ cells from sections indicated that progenitor populations were unaffected at D7 and D14 after Gilz deletion (Fig. 2C).

Increases in cyclin D1⁺ populations were found in Gilz-null neonatal testis (Bruscoli et al., 2012), suggesting increased mitotic activity of spermatogonia and consistent with enhanced KI67 positivity of progenitors (Fig. 2A). Cyclin D1 marks A-type plus intermediate differentiating spermatogonia and is undetectable in most undifferentiated cells (Fig. S6A). We considered that cyclin D1 might be aberrantly expressed in Gilz-deficient SOX3⁺ progenitors. However, cyclin D1 was not consistently detected in SOX3⁺ A_al of D7 and D14 Gilz^TAM-KO tubules (Fig. 2B). From sections, we found no difference in the percentage of SOX3⁺ cells (encompassing progenitors and early differentiating cells) that were cyclin D1⁺ at D7 and a modest reduction at D14 (Fig. 2D). To confirm altered mitotic activity upon GILZ loss, we analysed levels of phosphorylated retinoblastoma protein (RB) in PLZF⁺ cells (Fig. 2E). RB negatively regulates G1- to S-phase transition, and is phosphorylated and inhibited by cyclin/CDKs to promote cell cycle progression (Giacinti and Giordano, 2006). According to the seminiferous epithelium cycle and PLZF expression pattern (Fig. 1B) (Chan et al., 2017; de Rooij, 1998), the majority of PLZF⁺ cells at late cycle stages (IX-I) will be differentiating A-type spermatogonia, whereas those at early stages (II-IV) will primarily be A_al. At mid-cycle stages (V-VIII), PLZF marks A_al and A₁ spermatogonia. PLZF also marks A_s and A_pr, which are present throughout the epithelium cycle but represent a minor proportion of PLZF⁺ cells. In controls, PLZF⁺ cells at late stages were more frequently P-RB⁺ than those at early and mid-cycle stages, consistent with increased mitotic activity of differentiating versus undifferentiated cells (Fig. 2E,F and Fig. S6B) (de Rooij, 1998). More PLZF⁺ cells at early, mid- and late cycle stages were P-RB⁺ in D7 Gilz^TAM-KO testis versus controls (Fig. 2F and Fig. S6B). This increase in P-RB was most striking at mid-cycle stages when PLZF⁺ progenitors should be quiescent.

Our data indicated that progenitors proliferate excessively upon GILZ loss and differentiation is suppressed. Using intracellular staining and flow cytometry, we confirmed that a higher proportion of undifferentiated spermatogonia (PLZF⁺ c-KIT⁻) from D7 Gilz^TAM-KO testis were KI67⁺ compared with controls (Fig. 2G,H and Fig. S6C). The abundance of spermatogonia at early differentiation stages (PLZF⁺ c-KIT⁺) also decreased, consistent with suppressed differentiation (Fig. S6C,D). Although controls were also treated with TAM, TAM alone did not change relative proportions of PLZF⁺ spermatogonial fractions in wild-type adults or the proportion of undifferentiated cells (PLZF⁺ c-KIT⁻) positive for KI67 compared with vehicle-treated mice (Fig. S6E-H).

Gilz deletion in adults triggered SSC exhaustion plus changes in progenitor activity. We next explored relevant mechanisms. Given connections between GILZ and FOXO factors plus the role played by FOXO1 in SSC maintenance and differentiation, we analysed FOXO1 localisation upon Gilz deletion. In wild-type adults, nuclear (active) FOXO1 was abundant in PLZF⁺ undifferentiated cells (Fig. S7A). FOXO1 was detected in differentiating spermatogonia but was cytosolic, consistent with PI3kinase activation upon differentiation and AKT-dependent FOXO1 nuclear export (Fig. S7A) (Ciraolo et al., 2010; Goertz et al., 2011). By whole-mount analysis, D7 and D14 post-TAM, nuclear FOXO1 was evident in PLZF⁺ undifferentiated spermatogonia of Gilz^TAM-KO testis, indicating that FOXO1 activity was unaffected (Fig. 3A). A switch in FOXO1 localisation from nucleus to cytosol upon differentiation still occurred following Gilz deletion, confirming that GILZ is not required for FOXO1 nuclear exclusion (Fig. 3A), at odds with its neonatal role (Ngo et al., 2013b).

To confirm these observations, we scored FOXO1 localisation in c-KIT⁻ (undifferentiated) and c-KIT⁺ (differentiating) spermatogonia in control and Gilz^TAM-KO sections D7 post-TAM (Fig. S7B,C). In both control and Gilz-deleted testis, FOXO1 was predominantly nuclear or nuclear and cytosolic in the majority of undifferentiated cells, but exclusively cytosolic in most differentiating cells. Given that the subcellular distribution of FOXO1 in undifferentiated and differentiating populations upon GILZ loss was comparable with that of controls (Fig. S7C), we concluded that other effectors were more relevant for GILZ function in adult spermatogonia.

GILZ negatively regulates RAS and downstream AKT and ERK MAPK pathways that activate mTORC1 (Ayroldi et al., 2007; Laplante and Sabatini, 2012). We therefore considered that increased mTORC1 signalling underlies SSC exhaustion upon GILZ loss (Hobbs et al., 2015; Wang et al., 2016). In controls, phosphorylation of ribosomal protein S6 (P-S6), an indicator of mTORC1 activation (Laplante and Sabatini, 2012), was generally undetectable in undifferentiated cells but prominent in differentiating spermatogonia (A₁ to A₄) (Fig. 3B,C and Fig. S7D) (Hobbs et al., 2015; Sharma and Braun, 2018). Strikingly, in Gilz^TAM-KO tubules at D7 post-TAM, PLZF⁺ A_al plus remaining A_s and A_pr were P-S6⁺ (Fig. 3B,C and Fig. S8A,B). Gilz-deleted PLZF⁺ undifferentiated cells were also P-S6⁺ D14 post-TAM (Fig. 3B), confirming that mTORC1 was aberrantly activated.

To test whether mTORC1 hyperactivation upon GILZ loss was responsible for SSC exhaustion, we treated Gilz^TAM-KO and control mice with TAM, then (at D3) started daily treatment with mTOR kinase inhibitor torin 1 or vehicle for 5 days (Fig. 3D). Torin 1 is an ATP competitive inhibitor of mTOR that more fully suppresses mTORC1 function than rapamycin (Thoreen et al., 2012; Thoreen et al., 2009). Torin 1 also targets mTORC2, a distinct mTOR complex that regulates AKT (Laplante and Sabatini, 2012). However, retention of nuclear FOXO1 in undifferentiated spermatogonia upon GILZ loss suggests that AKT, and hence mTORC2, is not aberrantly activated (Fig. 3A) (Calnan and Brunet, 2008). Torin 1 suppressed mTORC1 activity in PLZF⁺ spermatogonia of control and Gilz^TAM-KO testis, as indicated by P-S6 levels (Fig. 3E). Importantly, torin 1 rescued the depletion of GFRα1⁺ A_s and A_pr observed upon GILZ loss, indicating that mTORC1 inhibition by GILZ is essential for SSC maintenance (Fig. 3F,G). Consistent with physiological roles of mTORC1 in SSC differentiation (Busada et al., 2015; Hobbs et al., 2015), torin 1 also enhanced the abundance of GFRα1⁺ cells in controls (Fig. 3F,G). Exhaustion of GFRα1⁺ spermatogonia in TAM-treated Gilz^TAM-KO mice could also be rescued by rapamycin, which directly targets mTORC1 but not mTORC2 (Fig. S9A-C) (Laplante and Sabatini, 2012), supporting the importance of mTORC1 as a GILZ effector in SSCs.

Torin 1 reduced cyclin D1 levels in differentiating spermatogonia of controls, consistent with mTORC1-dependent translation of cyclin D1 mRNA (Fig. S9D) (Averous et al., 2008). Although SOX3⁺ spermatogonia did not exhibit increased cyclin D1 at D7 following Gilz ablation (Fig. 2B,D), the few GFRα1⁺ A_s and A_pr remaining were often cyclin D1⁺ (Fig. 3H). Those in controls were cyclin D1⁻ (Fig. S9D). GFRα1⁺ spermatogonia in torin 1-treated Gilz^TAM-KO mice were cyclin D1⁻, indicating that mTORC1 activation upon GILZ loss promoted cyclin D1 expression in SSCs (Fig. 3H). Consistent with role of cyclin D1 in proliferation, P-RB staining of control spermatogonia was reduced by torin 1 treatment (Fig. S9E). Torin 1 also rescued the increased P-RB positivity of PLZF⁺ spermatogonia following GILZ loss (Fig. 2E,F and Fig. S9E).

Germline ablation of mTORC1 inhibitor TSC2 promotes SSC exhaustion (Hobbs et al., 2015). In contrast to effects of Gilz deletion (Fig. 1D), some tubules in Tsc2 conditional knockout adults retained PLZF⁺ spermatogonia (Hobbs et al., 2015), suggesting that mTORC1 activation is unable to fully deplete SSCs. GILZ may therefore regulate additional pathways involved in self-renewal besides mTORC1. To confirm the kinetics of SSC depletion following TSC2 loss, we crossed Tsc2 floxed mice with the UBC-CreER line (Tsc2^TAM-KO) to allow TAM-induced Tsc2 deletion in adults. In this context, mTORC1 activation was characterised by immunofluorescence for phosphorylated eukaryotic translation initiation factor 4E binding protein 1 (P-4EBP1), a direct mTORC1 target (Hobbs et al., 2015; Kang et al., 2013). In controls, P-4EBP1 was detected within PLZF⁺ A_al and less commonly in A_s and A_pr that include SSCs (Fig. S10A) (Hobbs et al., 2015). In Tsc2^TAM-KO testis D7 post-TAM, PLZF⁺ A_s and A_pr were P-4EBP1⁺, indicating increased mTORC1 signalling (Fig. S10A). The abundance of GFRα1⁺ spermatogonia in D7 Tsc2^TAM-KO tubules was comparable with controls and only reduced by D14 (Fig. S10B,C), indicating that SSC exhaustion upon TSC2 loss occurred more slowly than after GILZ ablation. GFRα1⁺ cells of D7 Tsc2^TAM-KO testis were P-4EBP1⁺, whereas those of controls were mostly P-4EBP1⁻, confirming aberrant mTORC1 activation (Fig. S10B). Distinct kinetics of GFRα1⁺ cell failure in Tsc2 and Gilz knockout models suggests that GILZ regulates SSC function through mTORC1-dependent and mTORC1-independent pathways. Alternatively, Gilz deletion results in more substantial mTORC1 activation than TSC2 loss.

To gain insight into GILZ function, we isolated undifferentiated spermatogonia from control and Gilz^TAM-KO mice at D7 post-TAM for RNA-Seq analysis. Undifferentiated cells were sorted according to E-cadherin expression, a surface marker of this population (Tokuda et al., 2007) (Fig. 4A,B). Consistent with persistence of E-cadherin in differentiating spermatogonia, E-cadherin⁺ cells were divided into c-KIT⁺ (differentiating) and c-KIT⁻ (undifferentiated) fractions (Fig. 4C and Fig. S11A). E-cadherin⁺ c-KIT⁻ cells were positive for α6-integrin, a SSC marker (Fig. 4C and Fig. S11A) (Takubo et al., 2008). By qRT-PCR, expression of undifferentiated (Plzf and Pou5f1) and SSC-associated (Gfra1) markers was enriched in the E-cadherin⁺ c-KIT⁻ α6-integrin⁺ population from wild-type adults, confirming undifferentiated cell isolation (Fig. S11B). Moreover, the ability to generate undifferentiated spermatogonial colonies in vitro was essentially limited to E-cadherin⁺ c-KIT⁻ α6-integrin⁺ cells (Fig. S11C,D) (Hobbs et al., 2010). The relative abundance of E-cadherin⁺ c-KIT⁻ α6-integrin⁺ cells was reduced in Gilz^TAM-KO testis, suggestive of undifferentiated cell depletion (Fig. 4D). Numbers of gated undifferentiated cells were not significantly reduced (Fig. S11E).

Three-hundred and seven genes were aberrantly expressed in E-cadherin⁺ c-KIT⁻ α6-integrin⁺ undifferentiated cells upon Gilz deletion (false discovery rate<0.05 and fold change>1.5) (Fig. S12A and Table S1). SSC-associated genes (Gfra1, Ret, T, Shisa6, Eomes, Lhx1, Foxc2, Etv5 and Egr2) were expressed at lower levels in Gilz-ablated samples, whereas expression of progenitor-associated genes (Nanos3, Neurog3, Rarg, Sohlh1, Sohlh2, Piwil4 and Sox3) plus general undifferentiated cell markers (Zbtb16/Plzf, Sall4 and Foxo1) were not significantly altered upon GILZ loss (Fig. 4E and Fig. S12A) (Carrieri et al., 2017; Oatley et al., 2006; Tokue et al., 2017; Wei et al., 2018). These gene expression changes are consistent with SSC depletion within the undifferentiated pool and persistence of progenitors. Genes involved in the cell cycle (Foxm1, Plk1, Mki67, Nasp, Ckap2l, Birc5, Tktl2 and Ccng1) were upregulated in Gilz^TAM-KO samples, mirroring elevated proliferation (Fig. 2G and Fig. 4E). Grb10 was upregulated upon Gilz deletion, a mediator of negative feedback between mTORC1 and upstream pathways (Fig. 4E) (Yu et al., 2011). The insulin signalling components Insr and Irs2 were downregulated, consistent with negative feedback of the PI3kinase/AKT pathway through FOXO1 inactivation, despite the observed retention of nuclear FOXO1 in Gilz-deleted undifferentiated spermatogonia (Fig. 3A and Fig. S7B,C) (Puig and Tjian, 2005). Many genes involved in spermatogenesis (Zmym3, Usp9x, Utp14b, Baz1a, Dmrtb1, Fus, Glis3 and Nr6a1) were differentially expressed upon Gilz deletion (Fig. 4E and Table S1).

Analysis of differentially expressed genes by gene ontology (GO) revealed enrichment of genes in multiple processes, including differentiation, signal transduction, proliferation and metabolism (Fig. 4F). GSEA suggested altered activity of various pathways, including the FOXM1 network, insulin signalling and Myc (Fig. S12B). Identities of genes differentially expressed following GILZ loss are supportive of SSC exhaustion, increased progenitor proliferation, aberrant signalling and disrupted spermatogenesis.

Gilz deletion caused downregulation of genes essential for spermatogenesis, including Zmym3 and Usp9x (Fig. 4E and Fig. S12A). Zmym3 encodes a chromatin-binding protein that promotes DNA repair (Leung et al., 2017). Males lacking Zmym3 are infertile due to a block in meiosis (Hu et al., 2017). Nuclear ZMYM3 is detected in PLZF+ and SOX3+ spermatogonia plus subsets of GFRα1⁺ and c-KIT⁺ cells, indicating predominant expression in progenitors (Fig. 4G and Fig. S13A) (Hu et al., 2017). ZMYM3 levels were markedly lower in PLZF⁺ spermatogonia of Gilz^TAM-KO versus control testis D7 post-TAM, confirming that GILZ loss results in Zmym3 downregulation (Fig. 4E,G). USP9X is a deubiquitylase that regulates cellular pathways by modulating protein stability (Bridges et al., 2017; Premarathne et al., 2017). Germline Usp9x deletion blocks spermatogenesis and is associated with spermatocyte apoptosis (Kishi et al., 2017). USP9X colocalised with GILZ in the cytosol of spermatogonia and spermatocytes, and was expressed throughout the spermatogonial hierarchy (Fig. 4H and Fig. S13B). Importantly, USP9X levels were substantially reduced in spermatogonia and spermatocytes of D7 Gilz^TAM-KO testis, confirming that GILZ promotes Usp9x expression (Fig. 4H). Given the spermatogenic failure of Gilz knockout mice resembles that of Zmym3 and Usp9x null models, ZMYM3 and USP9X may be key GILZ effectors. USP9X also negatively regulates mTORC1 (Agrawal et al., 2012); Usp9x downregulation upon Gilz deletion could also contribute to observed mTORC1 dysregulation and SSC exhaustion.

To assess whether mTORC1 hyperactivation upon Gilz deletion was responsible for Zmym3 and Usp9x downregulation, we performed immunofluorescence on sections from vehicle- and torin 1-treated Gilz^TAM-KO mice. Expression of Zmym3 in PLZF⁺ spermatogonia was not obviously rescued by torin 1 (Fig. 4I). Torin 1 slightly increased the proportion of Gilz^TAM-KO SOX3⁺ spermatogonia expressing Zmym3 but substantially fewer SOX3⁺ cells were ZMYM3⁺ than in controls (Fig. 4J). Torin 1 did not restore Usp9x expression in PLZF⁺ spermatogonia upon GILZ loss (Fig. S13C). GILZ therefore promotes Zmym3 and Usp9x expression independently of its ability to regulate mTORC1.

To gain insight into mechanisms by which GILZ regulates SSCs and progenitors, we established cultures of undifferentiated spermatogonia from untreated Gilz^TAM-KO adults (Fig. 5A) (Chan et al., 2017; Hobbs et al., 2010). Transplantable cells are present at low frequency in this culture and the bulk of cells resemble progenitors (Nagano and Yeh, 2013). Gilz was deleted in Gilz^TAM-KO cultures upon 4-hydroxy-TAM treatment whereas markers of undifferentiated spermatogonia (GFRα1, PLZF, SALL4, E-cadherin) were maintained (Fig. 5A and Fig. S14A). Gilz ablation suppressed cell growth (Fig. 5B), consistent with disrupted SSC activity. Importantly, Gilz deletion resulted in increased mTORC1 signalling in vitro as in vivo, as indicated by elevated P-S6 (Fig. 5C). P-4EBP1 was not consistently increased, suggesting differential regulation of mTORC1 effectors (Fig. S14B). mTORC1 integrates multiple regulatory inputs, including AKT and ERK MAPK (Hobbs et al., 2010; Laplante and Sabatini, 2012). Phosphorylated (active) ERK MAPK was increased upon GILZ loss but AKT activity (P-AKT) was not consistently altered (Fig. 5C and Fig. S14B). Blocking ERK MAPK with an inhibitor rescued increased P-S6 following Gilz deletion, demonstrating that GILZ regulates mTORC1 through ERK MAPK (Fig. S14C). As a control, torin 1 suppressed mTORC1 activation in Gilz-deleted cells (Fig. S14C). In agreement with in vivo analysis, ZMYM3 and USP9X were reduced upon Gilz deletion (Fig. 5D). Inhibition of ERK MAPK or mTOR did not rescue Zmym3 and Usp9x expression in TAM-treated Gilz^TAM-KO cells (Fig. S14C), confirming that GILZ promotes expression in a mTORC1-independent manner.

GILZ regulates targets through direct binding (Beaulieu and Morand, 2011), e.g. GILZ interacts with RAS and disrupts coupling to ERK MAPK and PI3K/AKT pathways (Ayroldi et al., 2007). GILZ binds and inhibits RAF, suggesting that GILZ regulates ERK MAPK through multiple mechanisms (Ayroldi et al., 2002). Identification of binding partners in cultured spermatogonia was predicted to reveal mechanisms by which GILZ regulates SSC and progenitor function.

Analysis of lysates under non-denaturing conditions indicated that GILZ was present in high molecular weight complexes and not as free monomer (Fig. 5E). To characterise GILZ-interacting proteins in cultured spermatogonia, GILZ was isolated by immunoprecipitation and associated proteins identified by mass spectrometry. GILZ bound other TSC22D family proteins (1, 2 and 4), the adaptor NRBP1 plus elongins B and C (ELOB/C) (Fig. S14D). Heterodimerisation of GILZ with TSC22D family proteins is predicted (Kester et al., 1999). NRBP1 may be a component of an E3 ubiquitin ligase containing ELOB/C and CUL5 (Kerr and Wilson, 2013). Both TSC22D2 and TSC22D4 are reported to interact with the NRBP1-ELOB/C complex (Wilson et al., 2012). Although GILZ interacts with RAS and RAF (Ayroldi et al., 2007; Ayroldi et al., 2002), we did not identify these components in mass spectrometry or by GILZ immunoprecipitation and western blot (Fig. S14E). Given the bulk of these cultures resemble progenitors, we cannot exclude the possibility that identified complexes are specific for progenitors and GILZ associates with distinct proteins in SSCs.

TSC22D proteins can be expressed as multiple isoforms containing distinct combinations of domains (Bruscoli et al., 2012; Khoury et al., 2008; Soundararajan et al., 2007). From mass spectrometry of GILZ-interacting proteins, peptides from TSC22D proteins could be assigned to various isoforms (Table S2). Combined with molecular weight analysis by western blot (Fig. S14F), we defined isoforms in spermatogonia (Fig. 5F). Long isoforms of GILZ/TSC22D3 (234 residues) and TSC22D2 (769 residues) were present, plus a short isoform of TSC22D1 (143 residues). Mass spectrometry suggested multiple TSC22D4 isoforms were expressed but available antibodies could not detect all isoforms (Table S2). A 387-residue TSC22D4 isoform was present (Fig. 5F) (Kester et al., 1999). Using GILZ immunoprecipitation and western blot, we confirmed the interaction of GILZ with other TSC22D proteins in spermatogonia (Fig. 5G). Select TSC22D isoforms contain a sequence (Motif 2) that mediates interactions with NRBP1 (Gluderer et al., 2010; Nie et al., 2015). GILZ lacks this motif and may bind NRBP1 indirectly via heterodimerisation with other TSC22D proteins (Fig. 5F). Although functions of the TSC22D-NRBP1-ELOB/C complex are poorly understood, we verified GILZ interaction with NRBP1, ELOB and ELOC by immunoprecipitation and western blot (Fig. 5H).

TSC22D proteins heterodimerise and contain conserved domains, suggesting functional redundancy, e.g. both GILZ and short isoforms of TSC22D1 inhibit RAS (Ayroldi et al., 2007; Nakamura et al., 2012). We therefore tested whether Tsc22d1 overexpression rescued increased ERK MAPK and mTORC1 activation upon GILZ loss. Increases in P-ERK and P-S6 upon Gilz deletion were comparable in control and Tsc22d1-transduced cells, indicating that TSC22D1 cannot compensate for GILZ (Fig. 5I). As control, spermatogonia transduced with Gilz did not activate ERK MAPK and mTORC1 following endogenous Gilz deletion (Fig. 5I). Although Tsc22d1 could not rescue mTORC1 activation in Gilz-deleted cells, it partially restored cell growth, suggesting coordinate regulation of other pathways by GILZ and TSC22D1 (Fig. S14G). Notably, TSC22D2 colocalised with GILZ in the cytosol of cultured undifferentiated spermatogonia and expression was unchanged by Gilz deletion (Fig. S14H). From wild-type adult whole-mount tubules, we confirmed that Tsc22d2 expression was prominent in the GFRα1⁺ SSC-containing population and PLZF⁺ A_s and A_pr (Fig. S14I). Our data suggest non-redundant roles for multiple TSC22D proteins in undifferentiated cell function and spermatogenesis.

To characterise GILZ function in mature SSCs, we developed an inducible knockout model allowing Gilz ablation in adults. We confirm an absolute requirement for GILZ in SSC maintenance and roles in progenitor proliferation and differentiation. We provide evidence that enhanced mTORC1 activation following GILZ loss underlies SSC exhaustion plus increased progenitor proliferation, and we link aberrant ERK MAPK signalling to mTORC1 dysregulation. Additionally, we uncover a role for GILZ in expression of USP9X and ZMYM3, essential regulators of spermatogenesis (Hu et al., 2017; Kishi et al., 2017). USP9X can also regulate mTORC1, suggesting that GILZ controls mTORC1 through multiple mechanisms (Fig. 6) (Agrawal et al., 2012; Bridges et al., 2017).

SSC function is dependent on niche-derived growth factors that induce self-renewal genes via ERK MAPK and PI3K/AKT (Kanatsu-Shinohara and Shinohara, 2013). However, these pathways also stimulate mTORC1, which promotes proliferation and differentiation (Hobbs et al., 2015; Serra et al., 2017; Wang et al., 2016). We propose that GILZ is an essential rheostat of growth factor-dependent signalling in SSCs (Fig. 6). In the presence of GILZ, response of SSCs to growth factors is restricted to allow induction of self-renewal factors, while limiting mTORC1 activation that triggers proliferation and differentiation. When GILZ is absent, the response of SSCs to niche factors is enhanced, resulting in excessive mTORC1 stimulation and SSC exhaustion. Mechanisms by which GILZ restricts signalling in SSCs warrant additional study. Although our data do not support direct regulation of RAS by GILZ (Bruscoli et al., 2012), we found that GILZ operates in a complex with other TSC22D proteins plus NRBP1 and ELOB/C in cultured undifferentiated spermatogonia. Our study confirms the conserved nature of TSC22D protein interactions and provides a basis for studies on their roles as signalling modulators (Gluderer et al., 2010; Wilson et al., 2012).

Regulation of mTORC1 is crucial for maintenance of stem cell function (Kalaitzidis et al., 2012; Magri et al., 2011; Rodgers et al., 2014). Deletion of Tsc1 or Tsc2, which encode a complex that inhibits mTORC1 via modulating activity of RHEB, disrupts SSC self-renewal (Hobbs et al., 2015; Wang et al., 2016). Although TSC1/2 ablation phenocopies effects of GILZ loss on SSC activity, kinetics of SSC exhaustion appears distinct. Germ cell loss following conditional Tsc2 deletion from embryonic stages is only evident by adulthood and PLZF+ spermatogonia persist in many tubules (Hobbs et al., 2015). In contrast, acute GILZ loss depleted the GFRα1⁺ SSC-containing population within 1 week and germline degeneration was complete within 1 month. We considered that chronic TSC2 loss might activate compensatory mechanisms that mitigate the effects of mTORC1 deregulation (Hobbs et al., 2015). However, GFRα1⁺ spermatogonial depletion was only evident 2 weeks after acute Tsc2 deletion in adults, indicating that GILZ loss results in more rapid SSC depletion. This may reflect the mechanism by which GILZ regulates mTORC1. GILZ inhibits mTORC1 through ERK MAPK, which regulates mTORC1 at multiple levels, including inhibitory phosphorylation of TSC2 and phosphorylation of mTORC1 component RAPTOR (Carriere et al., 2011; Laplante and Sabatini, 2012). Interestingly, ERK-dependent RAPTOR phosphorylation is linked with spermatogonial proliferation (Wang et al., 2017). Aberrant ERK MAPK activation might therefore promote mTORC1 activation more robustly than loss of TSC2 alone and hence have more dramatic effects on SSCs.

One gene upregulated in Gilz-deleted undifferentiated spermatogonia was Grb10, which encodes a negative regulator of insulin and insulin-like growth factor signalling (Plasschaert and Bartolomei, 2015). GRB10 is phosphorylated and stabilised by mTORC1, leading to suppression of upstream pathways (Hsu et al., 2011; Yu et al., 2011). GRB10 thus acts in an mTORC1-dependent negative-feedback loop. Grb10 is also induced by mTORC1 (Yu et al., 2011), consistent with increased expression upon GILZ loss. GRB10 induction in SSCs would limit responses to growth-promoting stimuli, potentially contributing to SSC depletion (Kubota et al., 2004). Ablating Grb10 in haematopoietic stem cells enhances regenerative capacity via loss of inhibitory effects on c-KIT dependent signalling (Yan et al., 2016). Given that GRB10 binds c-RET (Pandey et al., 1995), a component of the GDNF receptor, increased Grb10 expression upon GILZ loss may suppress SSC self-renewal by inhibiting the GDNF response. Plzf deletion also activates mTORC1-driven negative feedback in undifferentiated spermatogonia and suppresses GDNF responsiveness (Hobbs et al., 2010). Activation of mTORC1-dependent negative-feedback loops in SSCs may represent a key mechanism by which mTORC1 promotes differentiation.

GILZ is essential for meiotic progression but underlying mechanisms are unclear, given that mTORC1 dysregulation does not block spermatogenesis (Hobbs et al., 2015; Ngo et al., 2013b; Romero et al., 2012; Wang et al., 2016). We demonstrate that GILZ loss suppresses expression of the transcription factor ZMYM3 and the deubiquitylase USP9X. Knockout of Zmym3, which is primarily expressed in progenitors, results in meiotic arrest and spermatocyte apoptosis (Hu et al., 2017). USP9X is present throughout the spermatogonial pool and germline Usp9x deletion triggers spermatocyte depletion (Kishi et al., 2017). GILZ may therefore regulate spermatogenesis through ZMYM3 and USP9X. However, spermatogonial populations are preserved in Zmym3 and Usp9x knockout models (Hu et al., 2017; Kishi et al., 2017), indicating that other effectors are relevant for GILZ function in SSCs. Interestingly, USP9X regulates pathways with known or potential roles in SSCs by deubiquitylating and stabilising protein components, e.g. Wnt/β-catenin, Notch and Hippo (Premarathne et al., 2017; Toloczko et al., 2017; Yang et al., 2016). USP9X binds and inhibits mTORC1 in muscle progenitors (Agrawal et al., 2012), suggesting that GILZ-dependent mTORC1 inhibition is mediated in part through USP9X (Fig. 6). Conversely, USP9X promotes mTORC1 activity in neural progenitors by stabilising RAPTOR (Bridges et al., 2017). USP9X therefore modulates mTORC1 in a cell-type dependent manner. Further assessment of USP9X function is needed to confirm targets in SSCs. Given GILZ localisation to the cytosol and limited evidence for transcriptional roles, the manner by which GILZ regulates Usp9x and Zmym3 requires clarification.

GILZ interacted with other TSC22D proteins and the NRBP1-ELOB/C complex in undifferentiated spermatogonia. Functions of TSC22D proteins are potentially regulated by homo- and heterodimerisation with family members via the LZ domain (Gluderer et al., 2010; Kester et al., 1999). Dynamic changes in dimer composition are associated with alterations in cell activity (Hömig-Hölzel et al., 2011). Heterodimerisation of GILZ with TSC22D proteins may represent a mode of control of GILZ activity in SSCs, and vice versa. Besides GILZ, roles of other TSC22D proteins in SSCs and fertility are undefined. TSC22D1 modulates haematopoietic progenitor function, TSC22D2 is implicated in intestinal progenitor activity and TSC22D4 regulates neural differentiation (Canterini et al., 2012; Nakamura et al., 2012; Wilson et al., 2012). Roles for multiple TSC22D proteins in SSC function might therefore be predicted. Selected TSC22D isoforms contain a domain mediating interaction with NRBP1 and Drosophila studies have demonstrated that orthologues of TSC22D1 and NRBP1 form a growth-promoting complex (Gluderer et al., 2010; Nie et al., 2015; Wu et al., 2008). NRBP1 is a subunit of an E3 ubiquitin ligase, containing ELOB/C and CUL5, that targets TSC22D2 for degradation (Kerr and Wilson, 2013; Wilson et al., 2012). Although GILZ associated with NRBP1 and ELOB/C in spermatogonia, CUL5 was not detected and the relevance of this complex remains poorly appreciated.

Our study highlights the roles played by GILZ at different stages of spermatogenesis. GILZ is involved in SSC maintenance, progenitor proliferation and differentiation, and meiotic progression. Besides defining GILZ as a modulator of mTORC1 in undifferentiated spermatogonia, we identify mTORC1-independent effectors through which GILZ can control spermatogenesis. Given roles of mTORC1 plus GILZ target USP9X in immune cells, an appreciation of GILZ function in spermatogonia can provide insight into mechanisms by which GILZ regulates immune responses (Beaulieu and Morand, 2011).

Mice carrying floxed Gilz alleles have been previously described and were maintained on a mixed C57BL6/CBA background (Ngo et al., 2013b). Gilz is X-linked and Gilz conditional knockout mice were generated by crossing female Gilz^flox/flox mice with male mice expressing tamoxifen (TAM)-regulated Cre from the ubiquitin c promoter (UBC-CreER). Gilz^flox/Y UBC-CreER mice were used as experimental mice and Gilz^flox/Y mice without the UBC-CreER transgene as controls. For gene deletion, 6- to 8-week-old adult mice were injected for 2 consecutive days with 2 mg/kg TAM (Sigma) in sesame oil intraperitoneally (Matson et al., 2010). Torin 1 powder (Selleckchem) was dissolved in 100% N-methyl-2pyrrolidone (Sigma) to 25 mg/ml and then further diluted 1:4 with sterile 50% PEG400 (Sigma) to a final concentration of 5 mg/ml. Eight-week-old Gilz^flox/Y and Gilz^flox/Y UBC-CreER mice were first injected with 2 mg/mouse of TAM for 2 consecutive days to induce Gilz deletion then at day 3 post-TAM, mice were treated with vehicle or torin 1 at 20 mg/kg for 5 consecutive days by intraperitoneal injection. Testes were harvested 3 h after the last treatment. Mice were treated with rapamycin (4 mg/kg) or vehicle according to the torin 1 treatment regimen, as described (Hobbs et al., 2010). C57BL6 wild-type adults were used between 8 and 10 weeks of age. Id4^IRES-GFP mice are previously described and were maintained on a mixed FVBN/CBA background (Best et al., 2014). The Monash University Animal Ethics Committee approved all animal experiments.

Mouse testes were fixed with 4% paraformaldehyde (PFA) overnight at 4°C, transferred into 30% sucrose in phosphate-buffered saline (PBS) for cryoprotection, embedded in OCT compound (Tissue-Tek) and cut into 8 μm sections. The sections were blocked in PBS supplemented with 2% bovine serum albumin (BSA) (Sigma) and 10% foetal bovine serum (FBS) (GE Healthcare) prior to incubation overnight with primary antibodies diluted in blocking solution. For whole-mount immunofluorescence (Chan et al., 2017), testes were detunicated, and seminiferous tubules teased apart and rinsed in PBS on ice. Tubules were fixed with 4% PFA for 6 h at 4°C and washed in PBS prior to blocking in 0.3% PBS Triton X-100 (PBSX) supplemented with 10% FBS and 2% BSA. Tubules were incubated overnight at 4°C with primary antibodies diluted in 0.3% PBSX containing 1% BSA. Samples were washed in 0.3% PBSX and primary antibodies detected with appropriate Alexa Fluor-conjugated secondary antibodies (Jackson ImmunoResearch). Tubules were mounted in Vectashield mounting medium (Vector Labs). For immunofluorescence of cultured cells, undifferentiated spermatogonia were cultured on Lab-Tek chamber slides coated with Geltrex (Thermo Fisher Scientific) and washed in PBS prior to fixing with 4% PFA for 15 min. The slides were washed with PBS prior to permeabilising in 0.3% PBSX supplemented with 10% normal donkey serum (Sigma) and 2% BSA. Slides were incubated overnight at 4°C with primary antibodies diluted in PBS supplemented with 1% normal donkey serum. Slides were washed in PBS and primary antibodies detected with appropriate Alexa-Fluor-conjugated secondary antibodies (Thermo Fisher Scientific and Jackson ImmunoResearch, 1:500). DNA was counterstained with DAPI prior to mounting with Vectashield mounting medium (Vector Labs). Primary antibodies were as follows: goat anti-PLZF (AF2944, 1:500), anti-GFRα1 (AF560, 1:250), anti-SOX3 (AF2569, 1:250), anti-c-KIT (AF1356, 1:250) and anti-E-cadherin (AF748, 1:250) (all R&D Systems); rabbit anti-phospho-RPS6 (Ser235/236) clone D57.2.2E (1:300), anti-phospho-4E-BP1 (Thr37/46) clone 236B4 (1:500), anti-RARγ1 clone D3A4 (1:500), anti-DDX4 clone D10C5 (1:500), anti-phospho-RB (Ser807/811) clone D20B12 (1:500), anti-FOXO1 clone C29H4 (1:500) and anti-CEP55 clone D1L4H (1:250) (all Cell Signaling Technology); mouse anti-GILZ clone G-5 (Santa Cruz, 1:300); rabbit monoclonal anti-cyclin D1 clone SP4 (Novus Biologicals, 1:250); rabbit anti-SALL4 (ab29112, 1:2000), anti-ZMYM3 (ab19165, 1:500), anti-USP9X (ab19879, 1:300) and chicken anti-GFP (ab13970, 1:5000) (all Abcam); rat anti-KI67 clone SolA15 (eBioscience, 1:250); rabbit anti-TSC22D2 (A304-067A, 1:1000) (Bethyl Laboratories); and rabbit anti-SOX9 (ab5535, 1:1000) (Chemicon). Image analysis was performed with a Zeiss LSM780 FCS confocal microscope at the Monash University Micro Imaging facility.

The numbers of PLZF⁺ and SOX3⁺ spermatogonia per 100 µm tubule perimeter were scored from immunostained sections as described previously (Garcia et al., 2014). Seminiferous epithelium staging of DAPI-counterstained sections was based on established criteria (Fig. S6B) (Ahmed and de Rooij, 2009). Immunostained whole-mount tubules were assigned to a grouping of consecutive stages based on relative density and morphology of differentiating and undifferentiated spermatogonial populations as described previously (Chan et al., 2017; Phillips et al., 2010). A_s A_pr and A_al spermatogonia were distinguished in whole-mount tubules according to evidence of cell-cell contact, nuclear morphology, physical proximity and spatial arrangement (Fig. S4A and Fig. S8A) (Chan et al., 2017; de Rooij and Russell, 2000). A proportion of scored A_pr may be ‘false pairs’: two daughter A_s generated from a recent A_s division that have yet to migrate away from each other and therefore appear interconnected (de Rooij and Griswold, 2012).

For isolation of undifferentiated spermatogonia according to E-cadherin, single-cell suspensions were prepared from adult testis by digestion with type II collagenase (Sigma) as described previously (Tokuda et al., 2007). Cells were stained with the following antibodies: phycoerythrin (PE)-conjugated anti-E-cadherin clone DECMA-1 (1:250), allophycocyanin (APC)-conjugated anti-c-KIT clone 2B8 (1:500), fluorescein isothiocyanate (FITC)-conjugated anti-CD49f (integrin α6) clone GoH3 (1:250) (eBioscience or Biolegend). DAPI was used for live/dead cell discrimination. Cells were sorted with a BD Influx Cell Sorter (BD Biosciences) at the Monash Flowcore facility. Analysis of fixed and permeabilised testis cells for PLZF, c-KIT and KI67 has been previously detailed (Chan et al., 2017). Rabbit anti-GILZ (Santa Cruz FL-134, 1:300) and rabbit anti cyclin D1 clone SP4 (Novus Biologicals, 1:200) were detected using Alexa 488-conjugated secondary antibody (Thermo Fisher Scientific, 1:500). Cells were analysed on a LSR Fortessa X-20 and data analysis performed using FlowJo software.

RNA was extracted from flow-sorted testis cells using TRIzol LS (Life Technologies) and Direct-zol RNA MiniPrep kits (Zymo Research), including removal of contaminating DNA by in-column DNase I digestion. RNA quality was assessed using Bioanalyzer and RNA quantity by Qubit. All samples were processed starting with 2 ng of total RNA. SPI- amplified cDNA was quantitated and 100 ng sheared using Covaris sonication and processed as per the Nugen Ovation RNA-Seq system V2 protocol. Seven cycles of amplification were used to minimise amplification effects. Libraries were checked for size using Bioanalyzer and all were ∼310 bp within the expected range. Libraries were quantified by Qubit and qPCR and one equimolar pool was made based upon qPCR results. Following denaturation 200 pM of library pool was clustered in one lane of an Illumina HiSeq 3000 (100 bp paired end) eight-lane flowcell using cBot. RNA-sequencing was performed at the Medical Genomics Facility, Monash Health Translation Precinct. The Monash Bioinformatics Platform used the RNAsik pipeline to process data. A raw counts file was generated by featureCounts and uploaded to Degust for subsequent analysis (degust.erc.monash.edu). An established weighted analysis method implemented within the ‘limma’ statistical package was used to identify differentially expressed genes (DEGs) (Liu et al., 2015). Cut-off for DEGs was false discovery rate (FDR)<0.05 and fold change (FC)>1.5. Data have been deposited in the GEO repository (under accession number GSE107893). DEGs were classified using PANTHER (protein analysis through evolutionary relationships) (www.pantherdb.org) (Mi et al., 2016).

Isolated cells were lysed in TRIzol LS reagent (Thermo Fisher Scientific), and RNA purified and DNase treated using a Direct-zol RNA Miniprep kit (Zymo Research). A Tetro cDNA synthesis kit (Bioline) was used for cDNA synthesis and quantitative PCRs were run on a Mic qPCR Cycler (Bio Molecular Systems) using Takara Sybr Premix Ex Taq II (Clontech). Primer sequences were as follows: actin forward (FW), GGCTGTATTCCCCTCCATCG; actin reverse (RV), CCAGTTGGTAACAATGCCATGT; Plzf FW, CTCCGTAAGCGTCCCCTCTGC; Plzf RV, GGTGCAGGCTAGCACCGTCC; Pou5f1 FW, CAGCCAGACCACCATCTGTC; Pou5f1 RV, GTCTCCGATTTGCATATCTCCTG; Gfra1 FW, CACTCCTGGATTTGCTGATGT; Gfra1 RV, AGTGTGCGGTACTTGGTGC; Kit FW, GCCACGTCTCAGCCATCTG; Kit RV, GTCGGGATCAATGCACGTCA.

Undifferentiated spermatogonia were cultured on mitomycin-inactivated mouse embryonic fibroblast (MEF) feeder cells in StemPro-34 media (Thermo Fisher Scientific) supplemented with 10 ng/ml GDNF, 10 ng/ml bFGF, 20 ng/ml EGF, 25 μg/ml insulin and other additives (Chan et al., 2017; Hobbs et al., 2010). To establish cultures, undifferentiated spermatogonia were enriched from testis cell suspensions by CD9 selection using an EasySep biotin selection kit (Stem Cell Technologies) and biotinylated CD9 antibody (Biolegend clone MZ3, 1:500) (Kanatsu-Shinohara et al., 2004). Cells were used up to passage 10-12. For gene knockout, Gilz^TAM-KO cells were treated with 0.2 μM of 4-hydroxytamoxifen for 4 days (media refreshed after 2 days), then grown in standard media for an additional 4 days before harvesting for analysis. Prior to using undifferentiated cell cultures for western blot, mass spectrometry and immunoprecipitation, cells were harvested using trypsin/EDTA and MEF feeder cells depleted by plating on tissue culture plates for 30 min then non-adherent spermatogonia removed from adherent MEFs. For experiments performed under growth factor-reduced conditions for analysis of signalling pathway activity, cells were incubated for 24 h in media without GDNF, bFGF, EGF and insulin supplements. The ERK pathway inhibitor PD0325901 and the mTOR inhibitor torin 1 (both from Selleckchem) were first dissolved in DMSO according to the manufacturer's instructions then diluted in media to final concentrations of 5 µM and 0.5 µM, respectively.

Tsc22d1 and Tsc22d3 cDNA (Origene) was sub-cloned using standard PCR-based methods into a modified pCCL-hPGK-P2A-tdTomato vector (Dull et al., 1998). Undifferentiated spermatogonial cultures were infected with lentiviral-containing supernatant prepared as described previously (Chan et al., 2017; Hobbs et al., 2010). Cells infected with pCCL-hPGK-Luc2-P2A-tdTomato were used as controls. Cells were sorted twice according to tdTomato to ensure selection of infected cells, and expanded prior to treatment with 4-hydroxytamoxifen as described above. Cells were harvested at day 4 post-TAM then plated at 1×10⁵ cells per well of a 12-well plate for cell growth analysis or in six-well plates for western blot and allowed to form colonies for 10-14 days prior to analysis.

Immunoprecipitation and western blotting were performed as previously described (Chan et al., 2017; Hobbs et al., 2012). Antibodies are as detailed above and as follows: rabbit anti-S6 (5G10, 1:1000), anti-phospho-RPS6 (Ser235/236) clone D57.2.2E (1:2000), anti-ERK1/2 (137F5, 1:1000), anti-phospho-p44/42 MAPK (Thr202/Tyr204) clone D13.14.4E (1:2000), anti-AKT (C67E7, 1:1000), anti-phospho-AKT (Ser473) (D9E, 1:2000), anti-4EBP1 (53H11, 1:2000), anti-phospho-4E-BP1 (Thr37/46) clone 263B4 (1:2000) and anti-DDX4 clone D10C5 (1:1000) (all Cell Signaling); mouse anti-β-ACTIN clone (Sigma, 1:2000); rabbit anti-USP9X (ab19879, 1:1000) and rabbit anti-ZMYM3 (ab15165, 1:1000) (both Abcam); mouse anti-GILZ clone G-5 (Santa Cruz, 1:2000); goat anti-TSC22D1 (R&D Systems, 1:1000); rabbit anti-TSC22D2 (A304-067A, 1:1000) (Bethyl Laboratories); and rabbit anti-TSC22D4 (ARP37339, 1:250) (Aviva Systems Biology). Band intensity was quantified using ImageJ.

Mass spectrometry was performed as previously described (Chan et al., 2017). GILZ complexes were immunoprecipitated from cultured spermatogonial lysates with Dynabeads coupled with rat anti-GILZ clone CFMKG15 (eBiosciences, 10 µg) antibody using a Dynabeads Antibody Coupling Kit (Thermo Fisher). Dynabeads coupled to non-specific rat IgG were used for control immunoprecipitation. Proteins were eluted using 2,2,2-trifluoroethanol (Acros Organics) and 1 mM Bond-Breaker TCEP (Thermo Fisher) then digested overnight with trypsin (Sigma) at 37°C. Eluted peptides were characterised by tandem mass spectrometry collected with a NanoLC/OrbiTRAP (Thermo) at the Bio21 Proteomics Facility. Peptide identification was performed with the Mascot software against MSDB and UniProt databases. The peptide count threshold was one peptide with a confidence level of 95%.

Assessment of statistical significance was performed using an unpaired two-tailed t-test (GraphPad Prism). Associated P values are indicated as follows: *P<0.05; **P<0.01, ***P<0.001; not significant (ns) P>0.05. For mouse experiments, no statistical method was used to predetermine sample sizes and no specific randomisation or blinding methods were used.

We acknowledge the facilities and technical assistance of Monash Animal Research Platform, Monash FlowCore, Monash Bioinformatics Platform, Monash Micro Imaging, Monash Health Translation Precinct Medical Genomics Facility and the Mass Spectrometry and Proteomics Facility of Bio21 Institute. We thank Alex Swarbrick and Jane Visvader for providing Id4^IRES-GFP mice. The Australian Regenerative Medicine Institute is supported by grants from the State Government of Victoria and Australian Government.