TRIM28-dependent SUMOylation protects the adult ovary from activation of the testicular pathway

Gonadal sexual fate in mammals is determined during embryonic development and must be actively maintained in adulthood. In the mouse ovary, oestrogen receptors and FOXL2 protect ovarian granulosa cells from transdifferentiation into Sertoli cells, their testicular counterpart. However, the mechanism underlying their protective effect is unknown. Here, we show that TRIM28 is required to prevent female-to-male sex reversal of the mouse ovary after birth. We found that upon loss of Trim28, ovarian granulosa cells transdifferentiate to Sertoli cells through an intermediate cell type, different from gonadal embryonic progenitors. TRIM28 is recruited on chromatin in the proximity of FOXL2 to maintain the ovarian pathway and to repress testicular-specific genes. The role of TRIM28 in ovarian maintenance depends on its E3-SUMO ligase activity that regulates the sex-specific SUMOylation profile of ovarian-specific genes. Our study identifies TRIM28 as a key factor in protecting the adult ovary from the testicular pathway.

To understand its role in ovarian physiology, we generated a cKO of Trim28 in the somatic 1 compartment of the developing mouse ovary. We observed sex reversal in adult ovaries where the 2 follicular structure progressively reorganized in pseudo-tubules with Sertoli-like cells. We then 3 combined mouse genetic with transcriptomic and genomic approaches to determine the molecular 4 action of TRIM28 and its interplay with FOXL2 in adult ovaries. Our data show that TRIM28 maintain 5 the adult ovarian phenotypes through its SUMO-E3 ligase activity that controls the granulosa cells 6 programme and represses the Sertoli cell pathway.

9
Deletion of Trim28 induces masculinization of adult ovary. 10 Double immunostaining of XX gonads at 13.5 days post-coitum (dpc) showed that TRIM28 is co- 11 expressed with FOXL2 in ovarian pre-granulosa cells, (Fig. S1). To study its role in this crucial ovarian 12 lineage, we generated a mouse line in which Trim28 can be conditionally deleted using the 13 Nr5a1:Cre 21, 22 transgenic line (Trim28 flox/flox ; Nr5a1:Cre referred as Trim28 cKO or cKO in the 14 text/figures). In 13.5 dpc cKO ovaries, nuclear TRIM28 signal was strongly decreased in FOXL2-15 positive pre-granulosa cells, whereas it was still present at heterochromatin foci, and was nearly 16 disappeared at E18.8 ( fig. S1). At birth, XX cKO mice displayed normal external female genitalia, 17 without any obvious ovarian structure abnormality at 3 days post-partum (dpp) ( fig. S2). In FOXL2-18 positive immature granulosa cells, we did not detect any signal for TRIM28 and SOX8/SOX9, two 19 Sertoli cell markers ( fig. S2). Unlike granulosa cells that looked normal at this stage, oocytes were 20 larger, suggesting an early and indirect effect of TRIM28 absence on oogenesis. This suggests that 21 TRIM28 is not required for foetal ovary differentiation. However, as TRIM28 is still expressed in pre- 22 granulosa cells at 13.5 dpc, a potential role in the primary ovarian determination that occurs at ~11.5 23 dpc cannot be excluded. 24 In several follicles of 20 dpp Trim28 cKO ovaries, SOX8 was expressed in groups of cells that stopped 1 expressing FOXL2 (Fig. 1a). Double immunostaining showed that some SOX8-positive cells also 2 expressed SOX9, suggesting that SOX8 expression precedes SOX9, unlike what observed in mouse 3 embryonic testes 23 . As SOX8 and SOX9 are Sertoli cell markers, this suggests that foetal deletion of 4 Trim28 in pre-granulosa cells might induce their reprogramming towards Sertoli cells after birth, as 5 described for Foxl2 deletion 4 and oestrogen receptor double knock-out 2 . 6 In 8-week-old Trim28 cKO mice, ovarian organization was profoundly changed. Medullar follicles had 7 almost completely lost FOXL2 expression, expressed SOX8 and SOX9, and were reorganized into ovaries) that SOX8 might precede SOX9. Conversely, the cortical region presented a less advanced 12 phenotype: as observed in 20 dpp Trim28 cKO ovaries, follicles were still organized, but remodelling had 13 started with groups of cells that stopped expressing FOXL2 and expressed SOX8 and/or SOX9 (fig 14 S3c, f and i). These results show that in Trim28 cKO ovaries, the granulosa-to-Sertoli cell 15 transdifferentiation starts in follicles located in the medulla and then spread to the cortical regions. 16 In parallel, using the Terminal deoxynucleotidyl transferase dUTP nick end labelling (TUNEL) assay, 17 we did not observe any significant increase in apoptosis in 20 dpp and 8-week-old  In 4-month-old Trim28 cKO females, the transdifferentiation of granulosa cells into Sertoli cells was 21 complete: FOXL2 expression has disappeared, and follicles were completely remodelled into tubular 22 structures with cells that expressed the Sertoli cell markers SOX8, SOX9 and DMRT1 (Fig. 1b).

23
Histological analysis confirmed the progressive reorganization of ovarian follicles into tubular 24 6 structures and the transdifferentiation of granulosa cells into cells with a Sertoli cell morphology ( fig.   1 S5). This reorganization was undetectable in 4-week-old Trim28 cKO ovaries but was clearly visible in 2 the medulla at 8 weeks and was completed at 17 weeks. Germ cells (oocytes) were relatively normal in 3 ovaries with a preserved follicular structure but started to degenerate during transdifferentiation. In 8- 4 week-old ovaries in which the medullar part was reorganized into pseudo-tubules, oocytes had 5 disappeared or were degenerating ( fig. S5), and in 17-week-old ovaries they had disappeared.   ). Therefore, the loss of a single Trim28 allele does not 5 cause transdifferentiation of granulosa cells. 6 We next examined the temporal expression of several genes with roles in testicular and ovarian sex 7 determination in 0.5-(15 dpp), 2 and 4-month-old ovaries. Reverse transcription quantitative real-time 8 polymerase chain reaction (RT-qPCR) analysis revealed that in Trim28 cKO ovaries, the mRNA level of 9 most ovarian-specific genes was decreased, with the exception of Rspo1 (Fig. 1c, panel Ovarian genes).

2
As several genes involved in steroidogenesis displayed a modified profile (Fig. 1c, fig. S7a-b for 3 temporal analysis, and RNA-seq data, respectively), we used mass spectroscopy to quantify the 4 production of major steroid hormones in control and Trim28 cKO ovaries and control testes from 7-5 month-old animals (Fig. 1e). Androgen levels (testosterone and androstenedione) in Trim28 cKO ovaries 6 and control testes were similar. Among the oestrogens produced in Trim28 cKO ovaries, estrone was 7 strongly reduced, whereas 17ß-estradiol levels were comparable to those in control ovaries. This can be 8 explained by the persistent expression of Cyp19a1 (the gene encoding the aromatase that catalyses 17ß-9 estradiol production) in Trim28 cKO ovaries ( Fig. 1c) and by the modified expression of genes encoding 10 hydroxysteroid dehydrogenases (HSD) ( Fig. S7a-b). Overall, our results indicate that foetal Trim28 11 deletion induces the masculinization of the steroid production profile in adult ovaries. transcriptomic studies of adult mouse/human testis/ovaries 27, 28, 29 . We then focused on the supporting 2 cell lineages. We identified 3,106 supporting cells that expressed granulosa and/or Sertoli cell markers 3 (n=1,112 in Trim28 cKO ovaries, n=1,446 in control ovaries, and n=548 in control testes) (Fig. 2a). In 4 Trim28 cKO ovaries, transcriptional profiles were asynchronous, some supporting cells were grouped 5 with control granulosa cells and expressed Esr2, Amh, Foxl2, Wnt4, Hsd17b1, and Nr5a2, indicating 6 that they still had a granulosa-like transcriptome (Fig. 2b). However, we also observed a gradient of

2
We then asked whether these intermediate cells resembled embryonic XX or XY supporting cell 3 progenitors 30 that de-differentiated from granulosa cells before differentiating into the Sertoli lineage. 4 We aligned all single cells along a pseudo-time ( Fig. 2d, 2e, fig. S10) 31 , and divided them in three 5 clusters based on their transcriptional profiles (Fig. 2a, right). This allowed us to identify genes that

15
TRIM28 acts in concert with FOXL2 on chromatin 16 As the Trim28 cKO phenotype was similar to that of mice after Foxl2 deletion in adult ovarian follicles 4 , 17 we asked whether these two proteins co-regulated common target genes in the ovary.

18
Immunofluorescence analysis confirmed that TRIM28 and FOXL2 were strongly co-expressed in the 19 nucleus of adult control follicular granulosa cells and to a lesser extent in theca stromal cells. Both 20 were almost undetectable in Trim28 cKO ovaries (Fig. 3a). Next, we performed TRIM28 and FOXL2 21 chromatin immunoprecipitation (ChIP) followed by next-generation sequencing (ChIP-seq) in control 22 ovaries to gain a global view of TRIM28 and FOXL2 co-localization genome-wide. Comparison of the 23 heatmaps of their co-binding to chromatin (Fig. 3b) showed that in ovaries, FOXL2 ChIP-seq reads 24 13 strongly mapped to regions occupied by TRIM28 (Fig. 3b, blue panel). Similarly, TRIM28 ChIP-seq 1 reads strongly mapped to FOXL2 peaks (Fig. 3b, red panel). Analysis of the overlap between TRIM28 2 and FOXL2 peaks confirmed that these proteins shared common genomic targets (62 and 55% 3 respectively, Fig. 3b Venn diagram). TRIM28 and FOXL2 bound to overlapping regions of genes that 4 have a central role in ovarian determination, such as FoxL2, Esr2, Fst (Fig. 3c), and of genes expressed 5 in granulosa cells (Fig. S11). As these genes were downregulated in Trim28 cKO ovaries, this suggests 6 that TRIM28 and FOXL2 positively regulate major granulosa cell genes. For instance, Wnt4, which 7 was downregulated in Trim28 cKO ovaries (Fig 1c), displayed several TRIM28 and FOXL2 peaks in 8 control ovaries (Fig. S11). Conversely, Rspo1, which is upstream of Wnt4 in the ovarian-determining 9 cascade 32 , was upregulated in Trim28 cKO ovaries (Fig 1c). Analysis of the TRIM28/FOXL2 genomic 10 profiles did not highlight any binding on Rspo1 (fig. S11), suggesting that its regulation in the adult 11 ovary is independent of TRIM28 and FOXL2. Moreover, in the absence of TRIM28, Wnt4 expression 12 seems to be independent from Rspo1 expression level.

51
We also analysed DNA motif enrichment for the binding sites of the major granulosa-specific 1 transcription factors (FOXL2 4 , RUNX 22 and ESR1/2 2 ) in TRIM28 and FOXL2 ChIP-seq data, as 2 previously described 8 . We observed a significant enrichment for these motifs in regions bound by 3 TRIM28 and FOXL2 in the ovary compared with regions bound by TRIM28 in bone marrow 35 and 4 thymus 36 (Fig. 3e). This shows that in adult ovaries, both TRIM28 and FOXL2 bind to regions that 5 display a genomic signature with binding sites for major ovarian-specific transcription factors. 6 To confirm that TRIM28 and FOXL2 co-localized on chromatin, we performed FOXL2 ChIP and 7 selective isolation of chromatin-associated proteins (ChIP-SICAP) followed by mass spectrometry that 8 provides only information relative to on-chromatin interactions 37 . We obtained a list of proteins co-9 localized with FOXL2 on ovarian chromatin that we ranked by their relative abundance. TRIM28 was 10 amongst the top 20 FOXL2 interactors, confirming that it is recruited on chromatin regions very close 11 to FOXL2 (Fig. 3f, left). It should be noted that TRIM28 has been recently shown 38 to interact with   S14). RT-qPCR analysis of 8-week-old ovaries (Fig. 4b) (Fig. 4c). Histological analysis ( fig. S15) also showed a similar tissue organization in Trim28 PHD/cKO 6 and Trim2 cKO ovaries. Altogether, these results indicate that the ovarian pathway maintenance in the 7 adult ovary depends on the E3-SUMO ligase activity of TRIM28.

9
To determine whether the global SUMOylation level in the nucleus of granulosa cells was affected in 10 Trim28 PHD/cKO and Trim2 cKO ovaries, we used a confocal microscopy quantitative analysis with anti-11 SUMO1 and -SUMO2/3 antibodies (called here SUMO2 because SUMO2 and 3 cannot be 12 differentiated with antibodies). In both Trim28 PHD/cKO and Trim28 cKO ovaries, SUMO1 and particularly 13 SUMO2 nuclear staining were decreased in ovarian somatic cells (Fig. 4d, left), as confirmed by 14 fluorescence quantification (Fig. 4d, right). This shows that the absence of TRIM28 SUMO-E3 ligase 15 activity in ovarian somatic cells decreased the nuclear level of SUMOylation, confirming the link 16 between TRIM28 and this post-transcriptional modification in vivo.

9
Quantification of SUMO1 and SUMO2 ChIP-seq reads that mapped to hypo-SUMOylated peaks (Fig   10   5a) showed that they were markedly decreased in Trim28 cKO and Trim28 PHD/cKO samples (Fig. 5a, upper 11 panel in blue). Moreover, quantification of TRIM28 ChIP-seq reads from control ovaries showed that 12 they mapped strongly to these regions (Fig 5b, box plots in blue). This shows that in control ovaries,

13
TRIM28 occupies chromatin regions that are hypo-SUMOylated in Trim28 cKO ovaries, strongly 14 implying that TRIM28 is the E3-ligase responsible of their SUMOylation in adult ovary (either auto-

15
SUMOylation or SUMOylation of transcription factors located near TRIM28 on chromatin). For 16 example, many hypo-SUMOylated regions in Trim28 cKO ovaries were occupied by TRIM28 and 17 FOXL2 in control ovaries (Fig. S17) RUNX1 ChIP-seq data in adult ovaries, we performed SUMOylation assays in cells transfected with 1 wild type TRIM28 or the PHD mutant. We observed that TRIM28 wild type, but not the PHD mutant 2 induced SUMOylation of both FOXL2 and RUNX1 ( fig. S18), suggesting that both factors are 3 potential substrates of TRIM28 E3-ligase activity. 4 However, TRIM28-dependent SUMOylation of transcription factors might also occur before their 5 interaction with chromatin because only a fraction (33 to 45%) of hypo-SUMOylated regions in 6 Trim28 cKO and Trim28 PHD/cKO ovaries were occupied by TRIM28 in control ovaries ( fig. S17). We also 7 found a substantial number of SUMO1 or SUMO2 peaks with a significantly stronger signal in 8 Trim28 cKO or Trim28 PHD/cKO than control ovaries (Log2 FC>1, AdjP val >0.05) that we designated as 9 hyper-SUMOylated (Fig. 5b, upper panel, and red spots in fig. S16). ChIP-seq read quantification 10 showed that in Trim28 cKO and Trim28 PHD/cKO ovaries, hyper-SUMOylation (SUMO1 and SUMO2) 11 occurred de novo on regions that were less SUMOylated in control ovaries (Fig. 5a, lower red panels).

12
Moreover, quantification of TRIM28 ChIP-seq reads in control ovaries showed that these hyper-

21
Although more experiments are required to confirm that DMRT1 is SUMOylated, our analysis shows 22 that some hyper-SUMOylated peaks are effectively occupied by DMRT1 and SOX9 during adult 23 reprograming of granulosa to Sertoli cells.

24
Our results showed that downregulation of the ovarian pathway in Trim28 cKO and Trim28 PHD/cKO 1 ovaries allows the activation of another pathway, inducing the de novo SUMOylation of distinct 2 chromatin regions, possibly related to the activated testicular genes. Yet, the RNA-seq analysis of 3 Trim28 cKO ovaries (Data S1) did not highlight the upregulation of any testicular-specific E3-SUMO 4 ligase (e.g. proteins of the PIAS family). This suggests that such ligases are expressed also in granulosa  . S21b), suggesting a more complex regulation. However, most genes were strictly hypo-

11
(74%) or hyper-(75%) SUMOylated, indicating that they belong to distinct pathways. 12 Next, we analysed the SUMOylation status of the genes identified as upregulated or downregulated in specific gene (Fig 2c, fig. S10). We detected TRIM28 and FOXL2 peaks at four different regions of the 10 Cldn11 genomic locus (fig S12), likely to repress its expression. However, the most upstream of these 11 regions, which is an open chromatin region in embryonic gonads 62 , was hyper-SUMOylated in the cKO 12 and PHD mutants (Fig. S23). Therefore, upon disappearance of TRIM28 and/or FOXL2 in mutants, 13 some transcription factors might have access to this potential enhancer, to activate the Cldn11 gene.
14 Overall, the TRIM28 E3-ligase controls the maintenance of granulosa cell fate via the specific 15 SUMOylation of ovarian genes. In its absence, a distinct pathway takes place, leading to the hyper-

16
SUMOylation of some Sertoli cell-specific genes that is correlated with their activation.  question.

23
At the organ level, transdifferentiation is first completed in the medulla and then extends to the cortical 1 region. At week 8 post-partum, mutant ovaries displayed medullar pseudo-tubules and cortical follicles: 2 a two-step process also observed in mice where both oestrogen receptors were knocked out 63 .

3
Interestingly, medullar granulosa cells are mostly derived from bi-potential precursors in which 4 primary sex-determination occurs at 11.5 dpc and that are integrated in follicles at puberty 64, 65 .

5
Conversely, cortical follicle pre-granulosa cells are generated mainly by the celomic epithelium from 6 13.5 dpc until birth and sustain fertility 66, 67 . This suggest that bipotential precursor-derived medullar 7 granulosa cells might be more sensitive to the effect of Trim28 absence/mutation. binds to hypo-SUMOylated peaks, but in a smaller proportion than FOXL2. In the absence of TRIM28, 5 these transcription factors and FOXL2 might lose their capacity to maintain the ovarian programme. 6 Our data support the hypothesis that a TRIM28-dependent programme of SUMOylation maintains the E3 ligase (e.g. TRIM28) might regulate a complete transcriptional programme through activation or 12 repression of target genes.

13
As the Trim28 cKO and Trim28 PHD/cKO ovarian transcriptomes displayed a strong masculinization, we 14 also observed activation of a de-novo pattern of chromatin SUMOylation (i.e. hyper-SUMOylated 15 peaks) that we attributed to the testicular pathway and that is catalysed by a still unknown E3-ligase.

16
These hyper-SUMOylated peaks might represent SUMOylated Sertoli-specific transcription factors, 17 such as DMRT1 or SOX9 that can be SUMOylated 70 . Importantly, by analysing ChIP-seq data 18 obtained by Lindeman and colleagues in ovarian reprograming experiments 60 , we found that a 19 substantial amount of the hyper-SUMOylated peaks from our results co-localized with DMRT1 peaks 20 and to a lesser extend with SOX9 peaks. This shows that both transcription factors are present in hyper-

21
SUMOylated regions and might be SUMOylated (or their partners) independently of TRIM28. SUMO 22 proteomic approaches should answer these questions about hypo-and hyper-SUMOylated peaks.

23
Altogether, our findings suggest a multi-step model. First, in the absence of TRIM28, FOXL2 that co-1 localizes on chromatin with TRIM28 would lose its capacity to maintain the expression of granulosa 2 cell-specific genes. Granulosa cells would differentiate into an intermediate state where they express 3 non-sex-specific markers. Second, this would lead to the de-repression of some Sertoli cell-specific 4 genes, such as Sox8 or Cldn11, allowing progressively the induction of strong activators of the Sertoli 5 cell pathway, such as Dmrt1. To confirm this model, we need now to generate mice lacking both Sox8, 6 Sox9 or Dmrt1 and Trim28.

7
Unlike its role in granulosa cells, TRIM28 is not required for the maintenance of adult Sertoli cells 8 where it is involved in their crosstalk with germ cells 20 and also in SUMOylation 71 . However, as we 9 could not completely abolish TRIM28 protein expression in pre-granulosa cells before 13.5 dpc, we 10 cannot exclude a role in primary sex-determination that occurs at 11.5 dpc. Indeed, in vitro studies 11 have shown that the testis-determining factor SRY, through its interaction with a KRAB-0 protein 72 , 12 might recruits TRIM28 on chromatin to repress ovarian genes 73 . Therefore, more genetic experiments 13 are required to delete Trim28 using Cre drivers that work earlier, as previously described for Gata4 74 .
14 TRIM28 is an important player in ovarian physiology and therefore, might also have a potential role in 15 genetic diseases causing reproductive disorders. TRIM28 has been recently identified as a tumour 16 suppressor in Wilms' tumour, a common paediatric kidney malignancy (reviewed in 75 ). However, no 17 TRIM28 mutation has been described so far in patients with reproductive disorders, such as primordial