Setdb1-mediated H3K9 methylation is enriched on the inactive X and plays a role in its epigenetic silencing

Over the past 15 years, it has been contentious whether H3K9 methylation is enriched on the mouse inactive X, and even in species where H3K9 methylation has been accepted to accumulate on the Xi, its role has remained unclear. Here we profiled the repressive histone marks H3K9me2, H3K9me3 and H3K27me3 on the mouse Xi. A striking feature of these profiles is the banded pattern they exhibit, with H3K27me3 marking gene dense regions and H3K9 methylation being predominant at gene poor regions, thereby creating distinct interspersed banding patterns along the chromosome and further enhancing our view of just how densely covered the Xi is with repressive histone modifications. Immunostaining of metaphase spreads has shown non-overlapping domains of facultative heterochromatin on the Xi defined by either H3K9me3 or H3K27me3 for human [24], cow [42] and vole [43]. More recently in mouse, H3K27me3 localisation was found to correlate with genes [22], but ours is the first description of discrete H3K9 methylated domains in mouse and indeed the first report of H3K9me3 on the murine Xi. Previous reports found no enrichment of H3K9me3 on the Xi by immunofluorescence [19]. Our data are consistent with these earlier findings, as we don’t find enrichment for H3K9me3 on the Xi compared with autosomes meaning enrichment on the Xi would be difficult to discern by immunofluorescence. Ours is also the first ChIP-seq analysis of H3K9me2 on the mouse inactive X. While early immunofluorescence studies identified H3K9me2 on the Xi, this work was later called into question due to antibody cross-reactivity with H3K27me3 [6]. Unfortunately, the paucity of SNPs at H3K9me2 peak regions on the X does not allow us to make a conclusion about the presence of this mark on the Xi; however, our comparison with the male X chromosome would suggest that it is. This combined with recent immunostaining evidence showing dense H3K9me2 regions colocalising with the dense H3K27me3 of the Xi [20], our own immunofluorescence study and an older study using an antibody against H3K9me2 without cross-reactivity issues [26], suggest further study of H3K9me2 on the Xi is warranted.

To functionally discern the role of H3K9 methylation on the Xi, we performed paired pooled and individual shRNA screens to identify which of the seven known H3K9 methyltransferases were involved in maintenance of XCI, using an X-linked GFP transgene. This approach revealed Setdb1 to be the most dose-dependent H3K9 methyltransferase for maintenance of XCI. Several other studies have investigated H3K9 methyltransferases in maintenance of XCI, including Ehmt2, Suv39h1 and Suv39h2 [20, 44, 45]. Consistent with our findings, only Setdb1 has been shown to play a role in XCI [20]. We additionally found no role for the remaining untested H3K9 methyltransferases Ehmt1 and Setdb2. As we performed our experiments by knockdown rather than knockout, we cannot exclude that the other H3K9 methyltransferases remain functional at depleted doses or that there is some compensation between various H3K9 methyltransferases such as those that are most closely related, e.g. Ehmt1 and Ehmt2. However, Setdb1 is clearly the most dose-sensitive H3K9 methyltransferase required for maintenance of XCI. Interestingly, while H3K9 methyltransferases are known to act as a multi-meric complex in some instances [46], here we find depletion of Setdb1 is sufficient to destabilise XCI. This is similar to what has been reported for silencing of LTR repeats by Setdb1 [40].

Using Setdb1 depletion as a tool to analyse the role of H3K9 methylation on the Xi, we analysed the epigenetic and transcriptional consequence of Setdb1 depletion in MEFs, where XCI is already completely established. Upon Setdb1 depletion, we observed a significant reduction in H3K9 methylation on the X; however, regions that lose methylation were often also marked by H3K27me3 or at the periphery of domains, meaning the distinct interspersed pattern of H3K9 and H3K27 methylation remained largely intact (Fig. 4c). This suggests these domains are stable once silencing is established, possibly protected by other H3K9 methyltransferases acting redundantly. Interestingly, a previous report identified the chromodomain protein Cdyl on the Xi [44]. They found that Cdyl’s recruitment to chromatin was dependent on both H3K9me2 and H3K27me3 and that it binds the H3K9 methyltransferase Ehmt2 (G9a), which is also enriched on the Xi. In combination with their data, our data suggest Setdb1 may establish H3K9 methylation on the Xi, providing a binding site for Cdyl, which could be involved in its maintenance through recruitment of Ehmt2 or other H3K9 methyltransferases such as Setdb1. However, there is likely redundancy between Ehmt2 and Setdb1, given that Ehmt2 has not been found to alter maintenance of the silent state of X-linked reporters here, or by others [20, 45], that Setdb1 itself has limited effects on the maintenance of endogenous X-linked and autosomal gene silencing, and that Setdb1 depletion does not completely ablate H3K9 methylation on the Xi. Likely this explains why upon Setdb1 depletion in MEFS we observed only a subtle effect on maintenance of X-linked gene silencing.

In addition to changes in H3K9 methylation upon Setdb1 depletion, we observed a decrease in DNA methylation at a subset of CpG islands on the Xi. This is consistent with the interplay between H3K9 methyltransferases, H3K9 methylation and DNA methyltransferases [47–50], and the synergism between Setdb1 and methyl binding domain protein 1 (Mbd1) that binds methylated CpGs for the maintenance of XCI [20]. Interestingly, depletion of Dnmt1 disrupts only the maintenance of random XCI, which was shown by normal expression of an X-linked reporter allele at E8.5 in both the embryo and placenta, but reactivation at E9.5 exclusively in the embryo [14]. This is in contrast to what we observed for Setdb1: firstly, we observed an effect for imprinted X inactivation; secondly, we observed an effect at E7.5. At this embryonic stage, cells are heterogeneously establishing or maintaining X inactivation, therefore raising the possibility that Setdb1-mediated H3K9 methylation could be involved in both stages of silencing. The earlier timing also places H3K9 methylation as a critical event in XCI that must occur prior to DNA methylation.

While we saw a striking effect on expression of the X-GFP transgene in Setdb1 null embryos, we saw only a very small reactivation of the same transgene in MEFs and saw even lower reactivation of endogenous X-linked genes in the same cells upon Setdb1 depletion. There are many possible contributors to these observations: knockout rather than knockdown of Setdb1; functional redundancy in maintenance of XCI and a requirement for Setdb1 at earlier stages of XCI. To address the latter two points and broaden our understanding of the role of Setdb1-mediated H3K9 methylation in inactivation of endogenous X-linked genes, we analysed the effect of depleting Setdb1 while XCI is first being set up in differentiating female ES cells. Using allele-specific RNA-seq, we show Setdb1-mediated H3K9 methylation has a striking role at this time, both on the Xi and at autosomal loci genome-wide. In each case hundreds of genes fail to become appropriately silenced, representing a central role for H3K9 methylation in setting up their silent state. Importantly, Setdb1-mediated H3K9 methylation has a larger effect when depleted while silencing is being established, than when it has already been completed and is only being maintained, both on the Xi and at autosomal genes. The functional role of Setdb1-mediated H3K9 methylation is in contrast to the role of PRC2-mediated H3K27me3. H3K27me3 is not required to establish or maintain random XCI, as determined by knockout of the PRC2 member Eed [12], nor is it sufficient to elicit gene silencing following expression of a silencing defective Xist [7, 51].