EZHIP constrains Polycomb Repressive Complex 2 activity in germ cells

The Polycomb machinery is required for the proper orchestration of gene expression by virtue of its critical role in maintaining transcriptional silencing. It is composed of several chromatin modifying complexes, including Polycomb Repressive Complex 2 (PRC2), which deposits H3K27me2/3. Here, we report the identification of a new cofactor of PRC2, EZHIP (EZH1/2 Inhibitory Protein), expressed predominantly in the gonads. EZHIP limits the enzymatic activity of PRC2 and lessens the interaction between the core complex and its accessory subunits, but does not interfere with PRC2 recruitment to chromatin. Deletion of Ezhip leads to a global increase in H3K27me2/3 deposition both during spermatogenesis and at late stages of oocyte maturation. This alteration of the epigenetic content of mature oocytes does not affect the initial number of follicles but is associated with a reduction of follicles in aging mice. We provide evidences that mature oocytes Ezhip -/- are not fully functional and that fertility is strongly impaired in Ezhip -/- females. Altogether, our study uncovers EZHIP as a novel functional player in the comprehensive chromatin remodeling that occurs in the gonads.


INTRODUCTION
Early in development, cells commit to specific lineages and acquire precise identities that require maintenance throughout the lifespan of the organism. Polycomb group proteins play an important role in this process by maintaining transcriptional repression through the regulation of chromatin structure 1 . In mammals, this machinery is composed of two main complexes: Polycomb Repressive Complex 1 and 2 (PRC1 and 2). The core PRC2 complex is composed of four subunits: the catalytic subunit EZH1/2, SUZ12, EED and RbAp46/48 1 . PRC2 catalyzes the di-and tri-methylation of lysine 27 on histone H3 (H3K27me2/3), an enzymatic activity which is required for its function. Indeed, the mutation of lysine 27 of histone H3 to arginine leads to loss of gene repression and mutant flies display a phenotype similar to deletion of PRC2 components 2 . H3K27me3 is generally enriched around the promoter of transcriptionally silent genes and contributes to the recruitment of PRC1 1 . H3K27me2 is widely distributed, covering 50-70% of histones, and its role is less defined but may be to prevent aberrant enhancer activation 3 .
The question of how PRC2 is targeted to chromatin and how its enzymatic activity is controlled has received ongoing attention 4 . Cumulative evidence suggests that PRC2 may not be actively recruited to chromatin and that instead its activity is promoted by the recognition of its own mark H3K27me3, ubiquitination of lysine 119 of H2A, GC-richness or by condensed chromatin 4 . Conversely, some histone modifications negatively influence PRC2 function, particularly those associated with active transcription, such as H3K4me3 and H3K36me3 4 . PRC2 binding to chromatin may also be inhibited by DNA methylation 5 , although other reports suggest that PRC2 is compatible with DNA methylation 6 .
A number of accessory subunits have now been shown to influence PRC2 function 4 . Recent comprehensive proteomic analyses suggest that they might form around two main PRC2 subtypes, PRC2.1 and PRC2.2 7 . The subunit SUZ12 plays a central role by orchestrating the cofactor interactions 8 . PRC2.1 includes one of the three Polycomb-like proteins (PHF1, MTF2 or PHF19) together with the recently identified PRC2 partners EPOP and PALI1 9,10 . The three Polycomb-like proteins harbor one Tudor domain and two PHD finger domains each 4

. Their
Tudor domain is able to recognize H3K36me3 decorated genes, which could be important for PRC2 association with transcribed targets 4 . The function of EPOP remains ambiguous since, in vitro, it stimulates PRC2 catalytic activity while, in vivo, it limits PRC2 binding, likely through interaction with Elongin BC 11 . In contrast, PALI1 is required for H3K27me3 deposition both in vitro and in vivo 10 . The other complex, PRC2.2, includes JARID2 and AEBP2 subunits in equal stoichiometry 12,13 . Both are able to stimulate PRC2 catalytic activity in vitro with JARID2 being also able to bind nucleosomes 14 . JARID2 also appears to be necessary for PRC2 targeting at its loci, possibly through its DNA binding domain or as a result of its methylation by PRC2 4 . AEBP2 binds to DNA in vitro, but appears to negatively modulate PRC2 in vivo 15,16 . Of note, AEBP2 was reported to stimulate PRC2 through a mechanism independent of PRC2 allosteric activation 17,18 . While we now have a good picture of the accessory subunits interacting with PRC2, their precise roles are only partially understood. This might be due to compensatory mechanisms, such that interfering both with PRC2.1 and PRC2.2 is required to inhibit PRC2 recruitment 19 as observed upon loss of SUZ12 20 .
The regulation of chromatin structure in germ cells is pivotal as these cells are the bridge between generations and therefore potential vector of epigenetic information. In particular, H3K27me3 has been shown to be involved in parental imprinting 21,22,23 . Yet, in contrast to the extensive characterization of PRC2 in models such as mouse embryonic stem cells (ESC), much less is known about the regulation of its enzymatic activity in germ cells. Deletion of PRC2 core components during spermatogenesis results in the progressive loss of germ cells, indicating that its activity is required for this process 24,25 . At later stages of spermatogenesis, when round spermatids differentiate into mature sperm, histones are progressively replaced by protamines.
A variable fraction of the genome retains a nucleosomal structure (1% in mice, 10-15% in human), with histones carrying post-translational modifications, including H3K27me3 26 .
During oogenesis, histones are maintained and H3K27me3 is detected throughout this process [27][28][29] . However, H3K27me3 displays a peculiar pattern of enrichment in the growing oocyte, showing broad enrichment in intergenic regions and gene deserts (reviewed in 30,31 ). Genetic interference with PRC2 function in growing oocytes does not prevent their maturation but has been linked to a postnatal overgrowth phenotype in the progeny 32 possibly through the control of imprinting 23 .
Here, we report the identification of a new cofactor of PRC2, EZHIP, functioning primarily in mammalian gonads and show that this cofactor limits PRC2-mediated H3K27me3 deposition.
Inactivation of this cofactor results in excessive deposition of this mark, altering the epigenetic content of oocytes and impairing mouse female fertility.

Identification of a new cofactor of PRC2 in the gonad
PRC2 recruitment and enzymatic activity is controlled by a set of cofactors interacting in a partially mutually exclusive manner with the core subunit SUZ12 but little is known about its regulation in germ cells. To tackle this question, we first focused on testes (more abundant material than ovaries) and took advantage of knock-in mouse models expressing an N-terminal Flag-tagged version of either EZH1 or EZH2 from their respective endogenous locus (this study 33 ). We verified the expression of the tagged-EZH1 by western blot on mouse testis nuclear extract and were able to detect the presence of a slowly migrating polypeptide, which is specifically pulled down by Flag-Immunoprecipitation (Flag-IP) (Supplementary Figure 1A).
We then isolated nuclei from adult mouse testes (WT-control, EZH2-Flag or EZH1-Flag),  Figure 1B). Interestingly, our experiments also reveal the existence of a new partner, the uncharacterized protein AU022751 (ENSMUST00000117544; NM_001166433.1), which we retrieved in both EZH1 and EZH2 pull-downs. We referred to this new cofactor as "EZHIP" for EZH1/2 Inhibitory Protein. Of note, this protein was previously identified in PRC2 interactomes of mouse embryonic stem cells but its function was not further investigated 13,34,35 .
In order to confirm this interaction, we overexpressed Flag-tagged versions of the mouse and human homologs in HeLa-S3 cells (Supplementary Figure 1C) and performed IP followed by mass spectrometry. These reverse IPs confirmed the interaction between PRC2 and EZHIP ( Fig. 1B and Supplementary Figure 1D). Additional putative partners were identified in both IPs, but with the exception of USP7, they were not common to both homologs. We therefore did not pursue their study further. Importantly, these reverse-IPs also indicate that EZHIP interacts with both PRC2 complex subtypes.
EZHIP is located on the X chromosome. In most species, it is a mono-exonic gene -that may indicate that it was generated by retroposition-but in the mouse, splicing also creates a shorter isoform. Using PAML (Phylogenetic Analysis by Maximum Likelihood), we observed that EZHIP homologs are present across Eutheria but we did not identify any homologs outside of this clade based on either sequence conservation or on synteny. EZHIP genes have rapidly evolved both at the nucleotide and amino acid levels, the rodent homologs being particularly distant from the rest ( Fig. 1C and Supplementary Figure 1E). This contrasts with the other PRC2 components, such as EZH2, which are highly conserved across mammals (Supplementary Figure 1E). No known protein domain was predicted for EZHIP and the only distinguishing feature is a serine-rich region (Fig. 1C in green) including a short amino-acid stretch that is fully conserved in all orthologs identified (Fig. 1C in purple). To characterize Ezhip expression, we performed RT-qPCR on various tissues (3-month-old females and males). Ezhip mRNA expression was particularly high in ovaries; it was also expressed in testes and much less in other tissues (Fig. 1D). Of note, Ezhip transcript level appears at least ten-fold higher than any PRC2 core components or cofactors in oocytes (Supplementary Figure 1F). Ezhip's pattern of expression is distinct from that of Ezh2, which is expressed tissue-wide, with the strongest expression observed in spleen. Analysis of public gene expression datasets from fetal gonads 36 indicates Ezhip is preferentially expressed in E13.5 primordial germ cells (PGCs) compared to somatic cells, correlating with germ cell markers such as Piwil2 or Prdm14 (Supplementary Figure 1G). Interestingly, Ezhip belongs to a set of genes referred to as "germlinereprogramming-responsive" that become active following PGC DNA demethylation 37 , as they are associated with strong CpG island promoters. Similarly, in humans EZHIP is highly transcribed in male and female PGCs from week 5 until week 9 of pregnancy, while almost absent in ESCs and somatic cells (Supplementary Figure 1H) 38 . We confirmed this observation at the protein level by performing immunohistochemistry on sections of testes and ovaries of human origin. hEZHIP protein was detected in male germ cells inside the seminiferous tubules, especially in spermatogonia and round spermatids (Fig. 1E). In ovaries, EZHIP antibody stained primordial follicles and oocytes (red arrows) but not the external follicle cells in contrast to EZH2 antibody, which stained both zones (Fig. 1F). To summarize, EZHIP is a novel cofactor of PRC2 in placental mammals. It is a fast-evolving protein with no known protein domain, it is expressed primarily in PGCs during development and remains present in the adult gonad.

EZHIP is a negative regulator of PRC2 activity
To study the molecular role of EZHIP, we sought a model cell line that would express this factor endogenously. The EZHIP transcript is undetectable from most cell lines, with the exception of U2OS, an osteosarcoma-derived cell line (Supplementary Figure 2A). We used genome editing to generate U2OS clonal cells that were knockout for EZHIP or for EED as a control for PRC2 inactivation (U2OS EZHIP -/-and U2OS EED -/-respectively). Both cell lines were viable and had not overt phenotype. Western blot showed that deletion of EED destabilized the other PRC2 core components, such as EZH2, while inactivation of EZHIP had no discernible effect on the accumulation of these proteins ( Fig. 2A). We then assessed H3K27 methylation and observed a robust increase in H3K27me2/3 upon EZHIP deletion while H3K27me1 was stable and H3K27ac slightly reduced (Fig. 2B). Interestingly, H3K27me3 level was very low in U2OS compared to extract prepared from HEK-293T cells, which do not express EZHIP (Supplementary Figure 2B). To confirm that EZHIP deletion was directly responsible for the increased H3K27me3 in U2OS, we stably restored its expression using either full-length (FL) or deletion mutants ( To determine whether these alterations of PRC2 activity translate into aberrant gene expression, we analyzed the transcriptome of U2OS in the different genetic contexts described above by RNA-seq (Fig. 2D). Only a few genes were differentially expressed in U2OS EED -/-as compared to WT (FDR<0.05), whereas approximately 500 genes were differentially expressed in EZHIP -/-versus WT. The majority of which were downregulated, as expected considering the global gain of H3K27me3 repressive mark. Of note, gene ontology analysis of the genes downregulated upon EZHIP-knockout did not reveal any robust categories (Supplementary Figure 2E). Altogether, these results reveal that EZHIP inhibits the activity of PRC2 thus altering gene expression profile.

Interplay between H3K27 me2 &me3 upon expression of EZHIP
Having shown the effects of EZHIP on PRC2 activity at the global level, we then investigated how this affects the chromatin landscape locally. First, we analyzed H3K27me3 genomic distribution in the absence of EZHIP by chromatin immunoprecipitation followed by sequencing (ChIP-seq). We used the U2OS EED -/-in which H3K27me3 is not detectable as a negative control and compared it to the U2OS wild-type WT and EZHIP -/-. Replicates were well correlated and the U2OS WT and U2OS EED -/-clustered together, away from the U2OS EZHIP -/-(Supplementary Figure 3A). This agrees with our earlier observation that H3K27me3 is very low in U2OS, as in the EED knockout (Supplementary Figure 2B). In contrast, there was a genome-wide increase in H3K27me3 deposition upon deletion of EZHIP (Fig. 3A), as demonstrated by the large number of peaks detected in this context (Supplementary Figure 3B).
To further characterize the role of EZHIP, we analyzed the genome wide distribution of H3K27me2, H3K27ac and H2Aub in U2OS WT versus U2OS EZHIP -/-by "CUT&RUN" 39 .
Replicates clustered together as expected (Supplementary Figure 3C) and the correlation matrix revealed that H3K27me2 in the U2OS WT clusters with H3K27me3 in the EZHIP -/-condition suggesting that EZHIP limits the conversion of H3K27me2 into H3K27me3.
This global correlation is confirmed when zooming on a specific region as illustrated by genome browser screen shot (e.g. DRGX, Fig 3B). At this locus, H2Aub and H3K27ac displayed only modest variations and their enrichment appeared not particularly affected by the deletion of EZHIP. Focusing on peaks that gain H3K27me3 upon deletion of EZHIP (Fig. 3C), we noticed a general decrease of H3K27me2 with a slight depletion around the peaks of H3K27me3 when comparing U2OS WT to U2OS EZHIP -/- (Fig. 3D). H3K27ac which is known to anticorrelate with H3K27me2/3 enrichment was low and appeared to slightly decrease in the absence of EZHIP. Regarding H2Aub, the enrichment of the mark seemed rather insensitive to the deletion of EZHIP (Fig. 3D). We conclude that EZHIP limits the activity of PRC2 favoring the deposition of H3K27me2 at regions normally enriched for H3K27me3. Of note, this altered chromatin landscape is reminiscent of what has been described upon expression of H3K27M mutant oncogenic histone 40 .

EZHIP impairs PRC2 activity but not its binding to chromatin
Considering the inhibitory action of EZHIP on PRC2, we next sought to explore the underlying mechanisms. First, we hypothesized that EZHIP could limit PRC2 binding to chromatin. To test this hypothesis, we performed CUT&RUN against SUZ12 to monitor PRC2 recruitment to chromatin, comparing U2OS WT, U2OS EED -/-and U2OS EZHIP-/-. Focusing again on DRGX, we observed that SUZ12 enrichment is lost in the absence of EED, whereas SUZ12 is enriched both in U2OS WT and U2OS EZHIP -/- (Fig. 4A). This result held true when we analyzed all the peaks that gain H3K27me3 in the absence of EZHIP (Supplementary Figure   4A). Overall, we observed a slight increase of SUZ12 enrichment in particular at the regions flanking the peaks. Since our attempts to immunoprecipitate EZHIP were unsuccessful, we used immunofluorescence (IF) to evaluate its colocalization with EED and H3K27me2. The specificity of EZHIP antibody by IF is demonstrated by the lack of signal in U2OS EZHIP -/-(Supplementary Figure 4B). In U2OS WT, EZHIP staining appeared as a diffuse nuclear staining which overlaps partially with the signal detected for EED ( Fig 4B). Of note, EZHIP staining tends to be excluded from the bright dots detected with the anti-H3K27me2 antibody.
Since EZHIP modestly impacts PRC2 binding to chromatin but H3K27me3 deposition is impaired, this suggested that EZHIP may instead interfere with PRC2 enzymatic activity. To test this hypothesis, we first evaluated whether a titration of purified EZHIP (Supplementary Figure 4C) inhibited the enzymatic activity of the recombinant PRC2 core complex in a histone methyltransferase assay. However, even at molar excess, EZHIP did not impact the enzymatic activity of PRC2 (Supplementary Figure 4D). We then reasoned that EZHIP might regulate PRC2 activity only in the presence of its cofactors. To test this hypothesis, we purified the core PRC2 and its cofactors from U2OS and U2OS EZHIP -/-cells that stably over-express a Flagtagged version of EZH2 (Supplementary Figure 4D). EZH2 was immunoprecipitated and further purified through an ion-exchange column before monitoring its activity on native histones. While we observed very low methyltransferase activity towards H3 with PRC2 purified from WT cells, the complex purified from U2OS EZHIP -/-was much more active Overall, PRC2 displayed the same composition (Supplementary Figure 4F); however, the stoichiometry of the cofactors appeared substantially different in the absence of EZHIP (labelfree quantification based on iBAQ). Namely, several cofactors -AEBP2, JARID2 and PALI1were present at a higher stoichiometry in the IPs from EZHIP -/-cells. (Fig. 4D). We confirmed this result by co-IP/WB investigating the interaction of AEBP2 and JARID2 with EZH2 in IPs performed with nuclear extracts prepared from U2OS wild type or EZHIP -/-cells (Fig. 4E).
Our results suggest that EZHIP does not prevent PRC2 binding to chromatin but limits the stimulatory action of cofactors, such as AEBP2 and JARID2 on its enzymatic activity.

Ezhip -/-males are fertile despite H3K27me3 increase
To study the role of EZHIP in a more physiological environment, we generated a knockout We first investigated the expression of Ezhip during spermatogenesis in the different subpopulations of germ cells sorted from adult mice based on staining for α6-integrin, the tyrosine kinase receptor c-Kit, and DNA content, as previously described 41,42 . Ezhip was mostly expressed in spermatogonia (α6-integrin positive, Supplementary Figure 5E). Its expression was very low in spermatocytes I and II, consistent with the global transcriptional inactivation of the X chromosome at these stages 43 , in contrast to Ezh2 expression, which increases at the final stages of differentiation (4n, 2n and n; Supplementary Figure 5E).
We then tested whether deletion of Ezhip could enhance H3K27me3 deposition during spermatogenesis, as it does in U2OS cells. For this, we probed nuclear extracts from whole testes of adult mice by western blot. As shown in Figure  This suggest that EZHIP does not regulate H3K27me3 deposition in somatic cells of the testis.
Indeed, in Dnmt3l mutant testes that are germ cell-free 45 Figure 5G). Finally, we evaluated sperm quality through analysis of computer-assisted spermatozoa images. Spermatozoa motility felt within normal standards, although spermatozoa from Ezhip -/Y males showed slightly less progressive motility and were a bit more static (Fig. 5E and Supplementary Table   1). While most histones are replaced by protamine in mature spermatozoa, a small minority carrying various histone modifications including H3K27me3 is retained [46][47][48] . To determine whether Ezhip deletion impacts this residual H3K27me3, we quantified this mark in epidydimal sperm. Western blot of sperm extracts isolated from Ezhip -/Y mice displayed higher H3K27me3 levels compared to sperm originating from WT animals (Fig. 5H). Whether this upregulation has any functional consequences remains to be investigated, nonetheless, these results confirm the inhibitory activity of EZHIP on H3K27me3 deposition in male germ cells.
Interestingly, they reveal that an excess of H3K27me3 is compatible with spermatogenesis and male fertility.

EZHIP controls H3K27me3 deposition in growing oocytes
In female, classical assembly of chromatin is conserved throughout oogenesis. While the genome-wide deposition of H3K27me3 in PGCs remains to be investigated, H3K27me3 was reported to be progressively restricted during oogenesis to "non-canonical" locations such as intergenic regions and gene deserts 28 . To assess whether EZHIP could play a role in the regulation of H3K27me3 during oogenesis, we first investigated its expression during mouse oocyte development. It is highly expressed at all developmental stages of oocyte maturation  Figure 6B). Finally, a strong increase in H3K27me3 deposition was also observed in mature MII oocytes from 4-month-old Ezhip -/-females (Fig. 5E). We conclude that EZHIP restrains the deposition of H3K27me3 during oocyte maturation.

Oocyte defects upon deletion of Ezhip.
To evaluate the consequences of this global gain in H3K27me3 on gene expression, we first analyzed the transcriptome of a pool of MII oocytes harvested after superovulation of prepubertal females (4-weeks-old). RNA-seq analysis revealed a very similar transcriptome for the mutant oocytes compared to WT (Fig. 7A). We next investigated whether transcriptomic alterations could appear with aging, as well as if there could be some variability in the transcriptome of individuals oocytes. To this end, we performed single-oocyte-RNA-seq (9 WT and 10 Ezhip -/-4-months-old oocytes). We first ran the comparison between wild type to mutant in aged oocytes by pooling the single cell results to mirror our analysis with younger females. We observed that in aged oocytes the number of significantly differentially expressed genes remains limited, although the comparative expression pattern appears more dispersed This revealed that while most of the oocytes (regardless of Ezhip expression) clustered together, two Ezhip -/-oocytes, originating from distinct mice, were clear outliers (Fig. 7C). One of the top genes differentially expressed comparing the outliers and the rest of the Ezhip -/-oocytes was Mos (figure 7D). Mos has been shown to be required for MAP kinase activation during oocyte maturation and its deletion impairs microtubules and chromatin organization during the MI to MII transition 50 . Next, we checked the chromosome metaphase plate in MII oocytes, and found that Ezhip mutant mice displayed a slight increase in the number of oocytes with lagging chromosomes compared to control (6-weeks-old; Fig. 7E). Our single oocytes RNA seq and staining of individual oocytes revealed some heterogeneity in oocyte maternal pool as well as in general competence for fertilization. Altogether, our results support a general role for EZHIP in oocyte fitness by regulating H3K27me3 deposition.

Impaired fertility of Ezhip knockout females
We next investigate whether EZHIP could be involved in the control of follicle maturation. We did not observe any significant differences in the number of primordial, primary and secondary/antral follicles of pre-pubertal females (P17) regardless of Ezhip expression status, indicating that the initial oocyte pool is apparently intact (Fig. 8A top panels). In contrast, sections from older females (16 weeks) showed a global reduction in follicle number in the absence of EZHIP (Fig. 8A bottom panels), although the low number of mature follicles (primary and secondary/antral) at this age was insufficient to reach statistical significance.
Collectively, these data suggest a progressive, age related, exhaustion of primordial follicle reserve, from which growing follicles develop. Incidentally, ovaries from Ezhip -/-females appeared smaller, with a weight that was reduced by about 30% compared to wild type and heterozygous counterparts (Fig 8B).
In line with these results, we observed that Ezhip -/-female mice give rise to fewer progeny.
We therefore monitored their fertility by comparing the size and number of litters of WT versus mutant females (Fig. 8C). Six-week-old WT and Ezhip -/-females were mated with a reliably fertile male in the same cage and monitored daily for 20 weeks (Fig. 8C). All litters were genotyped to assign them to the correct mother; the numbers of mice at birth and at 3 weeks of age (time of genotyping) were similar. However, the total number of pups obtained from Ezhip -/-mothers considerably decreased each month, as the females aged ( We conclude from these experiments that absence of Ezhip in oocytes leads to alterations of the epigenetic landscape and is associated to strong reduction in female fertility. Whether the impairment in oocyte pool and its fitness might impair subsequent development of the embryo around and after fertilization remains to be determined.

DISCUSSION
Gametogenesis entails significant reprogramming of the epigenome. While histone replacement in spermatogenesis and the progressive loss of DNA methylation during germ cell specification are well documented 51,27 , less is known about the regulation of histone post-translational modifications during this process. Here, we focus on the Polycomb complex PRC2 to investigate this question. We identify a new PRC2 interacting protein specific to the gonad and showed that it inhibits PRC2 enzymatic activity. Inactivation of this factor leads to a global increase of H3K27me3 during both spermatogenesis and oogenesis. Alteration of the epigenetic content of oocytes leads to a severely compromised fertility.
The PRC2 complex exists in several flavors depending on which enzymatic subunits it is formed around (EZH1 or EZH2) and depending on which set of cofactors it interacts with 7 . It is known that EZH1 and EZH2 exert redundant functions in spermatogenesis 25 , consistent with this redundancy both subunits have a similar interactome in adult mouse testis. Among it, EZHIP contrasts with most of the cofactors identified to date: (i) its expression seems mostly restricted to germ cells (ii) homologs have only been found in Eutherians and it is a fastevolving protein (iii) it is a robust inhibitor of PRC2 enzymatic activity and (iv) it pulls down the entire PRC2 interactome. These last two characteristics are likely linked: it is expected that effective inhibition of PRC2 requires all flavors of PRC2 to be regulated. The poor sequence conservation of EZHIP sequence and its rather disordered structure prediction are more surprising considering that PRC2 and its cofactors are, in contrast, very well conserved. This suggests that the specificity of action of EZHIP on PRC2 could be primarily conferred by the conserved stretch of 13 amino acids. Such a mechanism involving a short linear motif in direct contact with binding partners (including chromatin modifiers) is a common strategy for parasites such as toxoplasma to manipulate the host cellular machineries 52 . It will be particularly interesting to perform structural analyses in order to precisely determine how this interaction occurs, how it interferes with the binding of AEBP2, JARID2 or PALI1 to PRC2, and how this impairs the enzymatic activity of PRC2 without impacting its recruitment to chromatin.
Another interrogation raised by this study is the advantage of expressing an inhibitor of PRC2 to limit H3K27me3 deposition in the gonads rather than downregulating the enzyme itself. We speculate that an inhibitor enables a tighter control over the timing of the reduction in PRC2 activity. Consistent with this possibility, Ezhip was recently identified among a set of genes that is expressed in PGCs, in response to the developmental DNA demethylation of the germline genome 37 . Of note, its localization on the X chromosome, explains that it remains expressed in oocytes while it is silenced in spermatocytes due to meiotic sex chromosome inactivation. The link between the wave of DNA demethylation and expression of this inhibitor of PRC2 raises the question of whether both processes are functionally related (i.e. whether PRC2 has to be inhibited when DNA methylation is lowered). It will be interesting to map H3K27me3 deposition in Ezhip -/-oocytes in order to determine whether it maintains broad enrichment in intergenic regions or at gene deserts 30 and also, whether it could impact on the reestablishment of DNA methylation during oocyte growth.
Recent reports have shown that H3K27me3 on the maternal genome is important for the regulation of allele-specific gene expression 23 and therefore that disrupting PRC2 activity in oocyte through the deletion of Eed impairs, post-fertilization, the allelic expression of a subset of genes 53 . Conversely, it is tempting to speculate that PRC2 activity might be limited by EZHIP in order to prevent it from invading genomic regions and thus potentially promoting excessive imprinting. Our results also suggest that excessive H3K27me3 levels resulting from Ezhip deletion in testicular germ cells are partially retained in mature spermatozoa. Although it does not seem to impact on the fertilizing properties of the spermatozoa, it will be interesting to determine whether embryos derived from oocytes fertilized with Ezhip -/-sperm develop normally. If they do, it would be consistent with the report that paternally inherited H3K27me3 is rapidly erased in the zygote and carries limited intergenerational potential 28 .

Retroviral production
Production of pMSCV-Hygromycin retroviral vectors was performed in 293T cells.
Transduction and selection of targeted cells were performed according to the online Addgene procedure. Hygromycin B was added at 0.2 µg/ml.

Mass Spectrometry analysis
Affinity purifications and liquid chromatography coupled to tandem mass spectrometry (LC-MS) analysis was performed essentially as described in 59 starting from either testis nuclear extracts, HelaS3 nuclear extracts or U2OS nuclear extracts. In brief, nuclear extracts were subjected to a single step Flag-immunoprecipitation (IP) in triplicate (ipFLAG). Control IPs were performed on extracts not expressing the Flag-tagged protein 60 . Nuclear extracts from the Flag-tagged cell line were also incubated with beads lacking Flag antibody. Thus, nine pull-downs were performed in total, three specific pull-downs and six control pulldowns.
Precipitated proteins were subjected to on-bead trypsin (Promega) digestion, after which peptide mixtures were analyzed by LC-MS using an EASY-nLC 1000 from Thermo connected either to a Q Exactive mass spectrometer or an LTQ-Orbitrap Fusion Tribrid mass spectrometer (both from Thermo). For samples measured on the Q Exactive (U2OS and Testis), the 10 most These iBAQ intensities were therefore subtracted from the iBAQ intensity in the Flag pulldowns. The resulting corrected iBAQ intensity for the Flag-tagged protein was set to 1 and the iBAQ values of the interacting proteins with their SD were scaled accordingly. This enabled stoichiometry determination of all the interactors relative to the bait protein.

KMT assay
KMT assay with recombinant PRC2 and EZHIP proteins were performed as described previously 62 . Briefly, the reaction was performed with 200ng of PRC2 alone or in presence of EZHIP, 1ug of substrates, 4mM DTT in methylation reaction buffer (50mM Tris-HCl pH 8.5, 2.5mM MgCl2), 3 H-SAM, and incubated at 30°C for 30 min. For KMT assay with PRC2-Flag purified from U2OS WT and EZHIP -/-, nuclear extracts were first fractionated on High Trap Q (GE Healthcare) prior to Flag-IP. Nucleosomal substrate for the assay was assembled from 5S 12 repeats DNA 63 and purified HeLa cell histone octamers by salt dialysis through a linear gradient (2.2 M NaCl to 0.4 M NaCl) followed by dialysis against TE solution.

EZH1-and EZH2-Flag knock-in mice.
EZH2-Flag was previously described 33 , EZH1-Flag was generated by homologous recombination at the Institute Clinique de la Souris (ICS).

Ezhip knockout mouse
Mice were hosted in pathogen-free Animal Facility. Of the 13 pups generated, 8 carried at least one modified allele. Two founders (N0) carrying the expected 1.5-kb deletion were selected. The absence of in silico-predicted off-target mutations was verified by Sanger sequencing, the two founders were bred with C57B6N mice.
Two additional backcrosses were performed to segregate out undesired genetic events, following a systematic breeding scheme of crossing Ezhip heterozygous females with C57B6N males to promote transmission of the deletion. Cohorts of female and male mice were then mated to study complete knockout progeny.

Histological sections and immunostainings
For histological sections, testis and ovary from either human patients from Curie Institute Pathology Platform or mice were dissected, fixed for 6h in 4% paraformaldehyde (Sigma) and washed with 70% ethanol according to pathology platform standard protocols. Organs were paraffin-embedded, sectioned (8µm) and stained with Hematoxylin using standard protocols. Slides were washed in PBS three times for 5 min, before incubating with ABC substrate for 30 min at RT. After washing again with PBS, DAB was prepared according to manufacturer instruction (Vector Laboratories) and the staining reaction monitored from 1 to 5 minutes.
Slides were stained with H/E following standard methods, dehydration steps from 90% Ethanol solution to Xylene is performed and slides were mounted in VectaMount permanent Mounting Media (Vector Laboratories).

Immunomagnetic, flow cell sorting and flow cytometry analysis of mice testis cell populations
Testicular single-cell suspensions were prepared from 2-3-months-old from WT and Ezhip -/mice as described previously 41 . The albuginea was removed and the seminiferous tubules were dissociated using enzymatic digestion by collagenase type I at 100 U/ml for 25 minutes at 32°C in Hanks' balanced salt solution (HBSS) supplemented with 20 mM HEPES pH 7.2, 1.2 mM MgSO4, 1.3 mM CaCl2, 6.6 mM sodium pyruvate, 0.05% lactate. Next, a filtration step was performed with a 40 µm nylon mesh to discard the interstitial cells. After HBSS wash, tubules were further incubated in Cell Dissociation Buffer (Invitrogen) for 25 minutes at 32°C. The resulting whole cell suspension was successively filtered through a 40 µm nylon mesh and through a 20 µm nylon mesh to remove cell clumps. After an HBSS wash, the cell pellet was resuspended in incubation buffer (same as previously plus glutamine and 1% fetal calf serum).

Mouse sperm quality test
Adult male mice were euthanized by dislocation and cauda epididymis was collected postmortem after carefully removing fat pad. Epididymis was opened and sperm released in IVF media (Vitrolife). 1:100 sperm dilution was loaded on Ivos (Hamilton Thorne machine) and sperm parameters were evaluated by Remote Capture software.

Mouse fertility evaluation
6-week-old WT and Ezhip -/-females (N=13 each genotype) were crossed and monitored for 20 weeks. One WT female and one Ezhip -/-female mouse were housed with an adult breeder male tested previously. Cages were monitored daily and pup numbers and litters were constantly registered. Adult females were euthanized at the end of the study and gonad morphology analyzed.

Organ Phenotypic Analysis
Adult males and females starting from 3/4 months old have been euthanized, ovaries/testis collected and individually weighted. The whole organ weight has been considered for comparison among the three genotypes for females and testis weight ratio for males.

Follicle counting
Sections were prepared as described above: For p17 mice, follicles were counted from at least 2 sections each organ/genotype (N=4 each genotype). Primary antibody against DPPA3 (Stella) was used to stain germ cells and DAPI staining to stain nuclei. Different types of follicles were classified by surrounding follicular cells shape.

GV and MII oocytes isolation from female mice
Germinal vesicle (GV) stage oocytes were obtained from 12-weeks-old females. The ovaries were removed, passed in pre-warmed PBS and transferred to M2 medium supplemented with 100ug/mL of dibutyryl cyclic AMP (dbcAMP; Sigma-Aldrich) at 38°C. The ovarian follicles were punctured with a 21-gauge needle and GV oocytes (fully grown oocytes exhibiting a centrally located GV) have been washed five times through M2 droplets in order to ensure that dbcAMP is removed. Zona pellucida has been removed by 3 passages in tyroide acid solution, followed by 3 washes in M2 medium. GV oocytes were further washed in PBS and processed for Immunostaining as described below. increasing times (2,5% for 5 min-5% for 5 min-10% for 10 min-20% for 5 min-50% for 15 min-DTG for 15 min) were then mounted on glass slides with ProLong Gold mounting medium (Life Technologies) for sequential Z-stack imaging. Fluorescence was detected using an Inverted Laser Scanning Confocal LSM700 UV Zeiss microscope with a 63x objective and Zstack scanning. More than 8 oocytes were examined for each condition unless otherwise specified.

RT-qPCR from mouse tissues
Total RNA was isolated using the Rneasy Mini Kit (Qiagen). cDNA was synthetized using High Capacity cDNA RT kit (4368814-Applied Biosystems) and quantitative PCR was performed with technical triplicate using SYBR green reagent (Roche) on a ViiA7 equipment (Applied Biosystems). At least three biological independent experiments were performed for each assay. Primers sequences are provided in Supplementary Table 2.

RNA extraction from Mature MII oocytes
Oocytes were incubated in M2 containing tyroide acid's solution for 2-3 min to remove their ZP (zona pellucida). ZP-free oocytes were carefully washed several times with M2 and were pooled prior to lysis in XB buffer from Arcturus PicoPure RNA isolation Kit (Applied Biosystems). We then added the spike-in control External RNA Control Consortium (ERCC) molecules (Invitrogen). Normalization was performed using ERCC spike in at 1:1000000. The purified total RNA concentration was measured using Agilent High Sensitivity RNA ScreenTape on Agilent 2200 TapeStation. First-strand cDNA (from total RNA) was synthesized according to the SMART-Seq™ v4 Ultra™ Low Input RNA Kit protocol (Clontech Laboratories). The PCR-amplified cDNA was purified using SPRI beads (Beckmann Coulter).

RNA sequencing from Mature MII oocytes
For sequencing, 75 bp paired-end reads were generated using the Illumina MiSeq. Raw reads were trimmed for adapters with cutadapt (1.12) using the Trim Galore! (0.4.4) wrapper (default settings) and subsequently mapped to the complete mouse rRNA sequence with Bowtie2 For differential expression analysis, genes were filtered to include those with CPM > 0.2 in at least 2 samples and filtered counts were transformed to log2-CPM and normalized with the TMM method. A linear model was fit to the normalized data and empirical Bayes statistics were computed. Differentially expressed genes for the KO versus WT were identified from the linear fit after adjusting for multiple testing and filtered to include those with FDR < 0.05 and absolute log2fold-change > 1.

RNA-seq in Spermatogonial stem cells
Kit-spermatogonial population was isolated by FACS as specified above from adult mice testis WT and Ezhip -/Y (pool of three different mice per experiment for each genotype). Sorted cells were resuspended directly in XB lysis buffer from Arcturus PicoPure RNA isolation Kit (Applied Biosystems). The purified total RNA was stored in nuclease-free water and RNA concentration was measured using Agilent High Sensitivity RNA ScreenTape. cDNA synthesis and library preparation were performed using SMARTer Stranded Total RNA-Seq Kit-Pico Input Mammalian. 100bp paired-end reads were generated using the HiSeq 2500 platform. Raw reads were trimmed for adapters with cutadapt (1.12) using the Trim Galore! (0.4.4) wrapper (default settings) and subsequently mapped to the complete mouse rRNA sequence with Bowtie2 (2.2.9). Unmapped reads were then mapped with STAR (2.5.3a) to the full reference genome (GRCm38/mm10). Libraries were confirmed to be stranded according to RSeQC after sampling 200000 reads with MAPQ>30. Gene counts were generated with STAR --quant_mode (uniquely mapped, properly paired reads that overlap the exon boundaries of each gene) using the Ensembl GTF annotation (vM13).
For differential expression analysis, genes were filtered to include those with CPM > 1 in at least 2 samples. Raw count data was normalized with the TMM method and transformed to log2-CPM. A linear model was fit to the normalized data, adjusting for batch effects, and empirical Bayes statistics were computed. Differentially expressed genes for each KO versus WT were identified from the linear fit after adjusting for multiple testing and filtered to include those with FDR < 0.05.

Single-cell RNA amplification
RNAs from individual oocytes were reverse transcribed from the 3'UTR and amplified as described 68 . ERCC spike-in (Invitrogen) were added in the lysis buffer (1:1000000) to address technical variation and further normalization. Only cells with high quality, based on morphology and amplification yield of housekeeping genes and ERCC sequences, were submitted to sequencing. Single-cell libraries were carried out according to manufacturer's protocol (Nextera XT, Illumina) and sequencing was performed on an Illumina HiSeq instrument in pair-end 100-bp reads

RNA sequencing in U2OS cells
Total RNA from U2OS cells was extracted with TRIzol. cDNA were generated according to manufacturer protocols (Illumina). 50bp single-end reads were generated using the HiSeq 2500 platform. Reads were first mapped to the complete human rRNA sequence with Bowtie2 (2.2.9). Unmapped reads were then mapped with STAR (2.5.2b) to the complete human reference genome (GRCh37/hg19). Gene counts were generated using STAR --quant_mode.
Libraries were confirmed to be strand-specific according to RSeQC (2.6.4) after sampling 200000 reads with MAPQ>30.
For differential expression analysis, genes were filtered to include those with CPM > 1 in 2 or more samples. Raw count data was normalized with the TMM method and converted to log2-CPM. A linear model was fit to the normalized data and empirical Bayes statistics were computed for each comparison. Differentially expressed genes for each comparison were identified from the linear fit after adjusting for multiple testing and filtered to include those with FDR < 0.05.

ChIP-seq, CUT&RUN
ChIP-seq was performed as described in 67 . CUT&RUN was performed as described 48 with some modifications described below. We started from 1.10 6 cells of interest (here U2OS) mixed with 50.10 3 drosophila S2 cells used for normalization (spike-in). The permeabilization and binding of primary antibody was performed for 1 hour at room temperature. Binding of the protein A-MNase fusion protein (pK19pA-MN, addgene 86973, produced in E. Coli) was done for 10 mn at room temperature. Targeted DNA digestion lasted 30 mn on ice. DNA extraction was performed using Nucleospin Gel and PCR clean up (Macherey Nagel).

Sequencing, alignment peak calling.
100 bp single-end reads were generated using the HiSeq2500 sequencer for H3K27me3 ChIPseq (WT, dEED, and dCxorf67), and 50 bp paired-end reads for CUT&RUN samples (Suz12, H3K27ac, H2AK119ub, H3K27me3, and IgG). Reads were simultaneously mapped to the human (GRCh37/hg19) and drosophila (dm6) reference genomes with Bowtie2 (2.2.9) using end-to-end alignment with the preset --very-sensitive. PCR duplicates were removed with Picard Tools MarkDuplicates (1.97) and BAM files were filtered to exclude common artifact regions (http://mitra.stanford.edu/kundaje/akundaje/release/blacklists/hg19-human). Peaks were called with MACS2 on combined replicates using the EED KO as a control for the K27me3 and SUZ12 ChIP and IgG control for CUT&RUN samples with the following parameters: -f BAM --gsize hs --broad --broad-cutoff 0.1 --bdg. Reads were counted in bins of length 50 and RPKM normalized and converted to bigWig format using DeepTools bamCoverage (2.4.1). Spike-in normalization: reads mapping to the drosophila genome were counted into 10kb bins and scale factors were calculated using DESeq2 estimateSizeFactors.