Chromatin-mediated silencing on the inactive X chromosome Free

X-chromosome inactivation (XCI) is a fascinating epigenetic process, whereby one of the two X chromosomes in female mammals is inactivated to achieve dosage compensation with males that have one active X chromosome. XCI achieves near chromosome-wide silencing that, once established, is faithfully maintained through mitosis for the lifetime of the mammal, which can be over 100 years. This incredible stability is one of the hallmarks of the inactive X (Xi), alongside a series of chromatin features whose role in the silencing process form the basis of this Review.

The dense chromatin state of the Xi was the key feature that catalysed its discovery. In 1949, Barr and Bertram discovered a densely staining nuclear body found only in samples from female cats but absent in their male counterparts (Barr and Bertram, 1949). A decade later, Ohno and Hauschka proposed that this nuclear body was an X chromosome (Ohno and Hauschka, 1960). Famously, Mary Lyon then proposed that the second densely staining X chromosome was genetically inactive and that a female mammal consists of cells in which the paternal X is inactive in some cells and the maternal X in others. She proposed that the inactive state is established early in embryonic development and heritable in all daughter cells, meaning females are essentially mosaics of cells that express either the paternal or maternal X (Lyon, 1961). These features predicted by Mary Lyon, based on mouse genetic studies, have all borne out and formed the basis of more than 60 years of exciting research in the XCI field.

The Lyon hypothesis also nicely encapsulates key features of XCI related to the presentation of X-linked diseases. Because females are mosaics of cells expressing the maternal or paternal X, often females heterozygous for an X-linked pathogenic variant have sufficient cells expressing the wild-type copy of the gene to protect them from disease presentation (Blewitt, 2023). However, in some cases, particularly neurodevelopmental disorders, the level of wild-type expressing cells is insufficient, either owing to skewing of which allele is the Xi or the crucial role the gene plays in the brain (Brand et al., 2021). One example of such a disease is Rett syndrome (Ip et al., 2018).

That female carriers of X-linked variants always have one wild-type allele presents the enticing prospect that the inactive wild-type allele could be switched back on from the Xi as a potential disease treatment. Here, the incredibly stable nature of the Xi poses a challenge. Based on an impressive series of studies from the field, enabled by the latest genomic methods partnered with imaging, there is an excellent understanding of the chromatin features of the Xi, when these features are acquired during the process of XCI (Fig. 1), and frequently a good understanding of the contribution each makes to gene silencing on the Xi. However, because of the inherent stability of the Xi, our understanding of how to reactivate gene expression has lagged behind. The latest epigenomic engineering methods have paved the way for locus-targeted reactivation, but, still, for the Xi it will be crucial to understand which chromatin features need to be removed and which added to enable long-lived activation from specific loci.

Mouse embryonic stem cells (mESCs) bi-allelically express both X chromosomes and establish XCI upon either differentiation or artificial induction of Xist expression. These in vitro models have proven to be the most tractable for the study of XCI, although they do have their limitations. In vivo mouse studies have also been invaluable (Box 1; Fig. 2). Occasionally, factors required for X-chromosome silencing in vitro are not required in vivo, likely reflecting a more stable and redundant repressive state in vivo. We discuss some of these occasions below. The lack of human pluripotent stem cell models of XCI means information on the establishment of human XCI is limited, although models are now emerging (Khan and Theunissen, 2023; Patrat et al., 2020). For this reason, we focus primarily on mouse, pointing out the human situation when available. We focus on chromatin changes downstream of Xist upregulation, centring on their role in gene silencing. Xist itself has been recently reviewed by others (Boeren and Gribnau, 2021; Brockdorff et al., 2020; Loda et al., 2022).

The initiating event for XCI was discovered in 1991 – the expression of the long non-coding RNA Xist/XIST, which is exclusively expressed from and spreads to coat the Xi (Borsani et al., 1991; Brockdorff et al., 1991, 1992; Brown et al., 1991a,b, 1992). For many years, the mechanism by which Xist could facilitate chromosome-wide silencing was elusive. Despite being generally poorly conserved, Xist contains a series of secondary structure-forming repetitive domains (Repeats A-F) that are conspicuously highly conserved and were, therefore, proposed to be crucial for the silencing function of Xist (Brockdorff et al., 1991 ;Brown et al., 1992 ;Nesterova et al., 2001).

Following initial observations that early-stage silencing of the Xi required Repeat-A (Wutz et al., 2002), a series of breakthrough studies utilised proteomic and functional genomic approaches to identify Xist RNA-interacting proteins. They found Xist Repeat-A recruits a host of repressive proteins, identifying the likely functional link between the non-coding RNA and gene silencing (Chu et al., 2015; McHugh et al., 2015; Minajigi et al., 2015; Moindrot et al., 2015; Monfort et al., 2015). Follow-up studies confirmed that Repeat A is crucial for the establishment of Xi silencing (Colognori et al., 2020; Dossin et al., 2020; Lu et al., 2016; Nesterova et al., 2019; Wang et al., 2019b).

The core Repeat-A binding protein appears to be SPEN, a large RNA-binding transcriptional repressor, which recruits co-repressors (Chu et al., 2015; McHugh et al., 2015; Minajigi et al., 2015; Moindrot et al., 2015; Monfort et al., 2015). SPEN also regulates Xist spreading and stability (Rodermund et al., 2021). Acute depletion of SPEN in mESCs and knockout in mice results in failed gene silencing (Dossin et al., 2020; Nesterova et al., 2019) (Fig. 1), demonstrating its crucial importance for the establishment of the Xi. Indeed, it is the recruitment of SPEN to the Xist-coated future Xi that heralds the earliest chromatin modifications required for XCI.

Chromatin of the future Xi undergoes progressive modifications to histone tails during establishment of silencing (Fig. 1). Live imaging shows that SPEN localises to Xist RNA immediately upon Xist induction, suggesting that SPEN-mediated repression occurs concurrently with Xist coating (Dossin et al., 2020). SPEN recruits multiple histone deacetylation complexes required for gene silencing. Indeed, histone deacetylation is the first chromatin modification in the hierarchy of XCI, occurring within 4-8 h of induction of Xist in an inducible female mESC model (Zylicz et al., 2019) (Fig. 2). Histone acetylation is associated with gene activity and its removal is a necessary first step in the silencing process, with chromosome-wide loss of H3 and H4 acetylation at promoters and enhancers (Dossin et al., 2020; Heard et al., 2001; Keohane et al., 1996; Zylicz et al., 2019). SPEN recruits the histone-deacetylating complexes NURD and NCOR/SMRT, which also includes HDAC3 (Dossin et al., 2020; McHugh et al., 2015). However, only HDAC3 as part of the NCOR/SMRT complex is required for gene silencing in XCI (McHugh et al., 2015; Zylicz et al., 2019). As HDAC3 is prebound to the future Xi, an interesting proposed mechanism for HDAC3-mediated deacetylation is that SPEN activates catalytically inactive pre-positioned HDAC3, presumably by recruiting other necessary members of the NCOR/SMRT complex to the Xi (Dossin et al., 2020; McHugh et al., 2015; Zylicz et al., 2019). Histone deacetylation seems to facilitate loss of activity-associated histone methylation, as deacetylation precedes loss of H3K4 trimethylation (H3K4me3) at promoters and H3K4me1 at enhancers (Zylicz et al., 2019).

Also early in the hierarchy of XCI, prior to gene silencing, nucleosome spacing at promoters becomes briefly relaxed (Keniry et al., 2022). This requires the nucleosome remodelling BAF complex; depletion of the BAF complex members Smarcc1 or Smarca4 results in failure of Xist to spread, failure to acquire repressive polycomb-associated histone marks and failure of gene silencing. Several subunits of the mESC-specific BAF complex (esBAF), including SMARCC1 and SMARCA4, interact with Xist, likely through Repeats A, C, E and F (Jegu et al., 2019; Minajigi et al., 2015). It is unclear why promoter relaxation is required. Potentially, it is a pioneering function that allows repressive complexes access to promoters of the future Xi. Interestingly, the association of SPEN with chromatin broadly correlates with BAF-mediated nucleosome dynamics at promoters. SPEN is present at promoters at a similar time to promoter relaxation and its association with chromatin seems dependent on active transcription, relating to promoter chromatin relaxation. Following gene silencing, promoter nucleosomes are condensed and SPEN disengages from chromatin (Dossin et al., 2020; Keniry et al., 2022). At this point, SPEN is not required to maintain silencing, suggesting that SPEN only facilitates establishment of XCI (Dossin et al., 2020; Zylicz et al., 2019).

The polycomb repressive complexes deposit histone tail modifications associated with repression, with polycomb repressive complex 1 (PRC1) catalysing monoubiquitylation of H2AK119 (H2AK119ub) and PRC2 mediating H3K27me3. Participation of PRC1 (de Napoles et al., 2004; Fang et al., 2004; Plath et al., 2004; Schoeftner et al., 2006) and PRC2 (Mak et al., 2002; Plath et al., 2003; Silva et al., 2003) in XCI is well known; however, their dynamics and requirement for gene silencing are only now becoming clear. On the future Xi, H2AK119ub deposition precedes H3K27me3, with the marks accumulating approximately 4 and 8 h post Xist induction, respectively, in an inducible female mESC model (Zylicz et al., 2019). H2AK119ub enrichment on the Xi occurs with similar timing to histone deacetylation, making it one of the first chromatin modifications in XCI (Nesterova et al., 2019; Zylicz et al., 2019).

Variant PRC1, a non-canonical form of the complex containing either polycomb group ring finger 3 (PCGF3) or PCGF5, is recruited to the future Xi by Xist, mediated predominantly through Repeat B/C, with the heterogenous ribonuclear ribonucleoprotein K (HnRNPK) acting as a molecular scaffold (Almeida et al., 2017; Bousard et al., 2019; Nakamoto et al., 2020; Pintacuda et al., 2017; Zylicz et al., 2019). Accumulation of H2AK119ub follows a similar pattern to Xist spreading, beginning in close proximity to the Xist locus and spreading to gene-dense regions that are in topological contact (Xist entry sites), and finally into proximal intergenic regions (Almeida et al., 2017; Engreitz et al., 2013; Pintacuda et al., 2017; Pinter et al., 2012; Schertzer et al., 2019; Simon et al., 2013; Zylicz et al., 2019). The JARID2 subunit of PRC2 recognises H2AK119ub, directing PRC2 to sites pre-marked by PRC1, with enrichment of PRC2-mediated H3K27me3 then following a similar pattern to H2AK119ub (Blackledge et al., 2014; Cooper et al., 2016; da Rocha et al., 2014; Kalb et al., 2014; Zylicz et al., 2019). Cyclin-dependent kinase 8 (Cdk8) is also required for recruitment of PRC2, via an unknown mechanism (Postlmayr et al., 2020), demonstrating that there is more to learn about polycomb recruitment to the Xi.

The role of Polycomb in transcriptional silencing is not entirely clear, likely owing to redundant functions within and between polycomb complexes and cell type-specific mechanisms of XCI. Polycomb appears to play a major, or perhaps less redundant, role in extra-embryonic tissues with loss of the PRC2 subunit EED being required for extra-embryonic but not embryonic XCI (Kalantry and Magnuson, 2006; Wang et al., 2001). Interestingly, PRC1 and PRC2 silence overlapping but different sets of X-linked genes in early post-implantation extra-embryonic tissues (Masui et al., 2023).

Studies of the role of polycomb in XCI in embryonic tissues are less conclusive. In both mESCs and mouse embryos, gene silencing fails in cells carrying an inducible Xist transgene with deleted Repeat A, despite apparently normal accumulation of polycomb marks (Cooper et al., 2016; da Rocha et al., 2014; Kohlmaier et al., 2004; Sakata et al., 2017; Wutz et al., 2002), meaning that polycomb is not sufficient to enable silencing in the absence of SPEN recruitment. Moreover, loss of RING1B, the main catalytic subunit of PRC1, does not affect XCI in the embryo (Leeb and Wutz, 2007). The apparently minimal role of polycomb for gene silencing may be complicated by redundancy between H2AK119ub and H3K27me3 and a need to block recruitment of both PRC1 and PRC2 to the Xi (Brockdorff, 2017). In support of this, variant PRC1 mutant embryos, in which recruitment of PRC1 and PRC2 is abolished, display female-specific lethality at embryonic day (E) 7.5-E8.5 (Almeida et al., 2017). Recent studies using Xist Repeat B/C deletion models, also attribute varying degrees of silencing potential to polycomb in embryonic tissue, with studies finding both large (Colognori et al., 2019; Nesterova et al., 2019; Pintacuda et al., 2017) and small (Bousard et al., 2019) effects on silencing. The study of polycomb in XCI is complex and has been reviewed in detail (Almeida et al., 2020).

Based on observations that both SPEN and polycomb are involved in the earliest stages of XCI, and are recruited by different Xist repeats, a recent report by Bowness et al. performed a head-to-head comparison of the two, showing that SPEN-dependent gene silencing and PRC1-dependent silencing occur in parallel. Removal of both pathways results in complete failure of XCI in a female mESC differentiation model with inducible endogenous Xist, which cannot be attributed to a simple failure of Xist spreading. These data suggest that SPEN and PRC1 are the main players required early during XCI to trigger establishment of gene silencing (Bowness et al., 2022).

Although it is clear that SPEN mediates silencing via histone deacetylation, the mechanism by which polycomb facilitates silencing is not understood; however, it likely in part involves polycomb reader proteins. One potential reader protein is structural maintenance of chromosomes flexible hinge domain containing 1 (SMCHD1), which becomes enriched on the Xi relatively late in the ontogeny of XCI, 3-4 days after Xist induction and differentiation in vitro (Bowness et al., 2022; Gendrel et al., 2012). Recruitment of SMCHD1 is dependent on H2AK119ub, downstream of the Xist B/C repeat and HnRNPK. Curiously, SMCHD1 does not require H3K27me3 for Xi localisation (Jansz et al., 2018b; Wang et al., 2019a). Zygotic loss of SMCHD1 leads to loss of silencing at many (although not all) Xi genes in vivo, resulting in female-specific embryonic lethality around E10 (Blewitt et al., 2008; Gdula et al., 2019; Gendrel et al., 2012; Mould et al., 2013; Sakakibara et al., 2018; Wang et al., 2018), a little after that of PRC1 mutant embryos potentially owing to major roles for PRC1 at autosomal targets. It is tempting to speculate that SMCHD1 is a downstream factor required for PRC1-dependent silencing, but this has not been formally tested.

Although SMCHD1 is recruited relatively late in XCI (Bowness et al., 2022; Gendrel et al., 2012), it does not necessarily play a role in maintenance of gene silencing. Removal of SMCHD1 after XCI is already established does not result in gene reactivation, in stark contrast to what was observed in Smchd1 null embryos. This suggests that the primary role of SMCHD1 is during establishment of XCI (Bowness et al., 2022; Gdula et al., 2019; Jansz et al., 2018a; Sakakibara et al., 2018; Wang et al., 2019a). SMCHD1 predominantly silences genes that are repressed later in the time course of XCI, and, intriguingly, these genes require differentiation to be silenced. The SMCHD1-dependent silencing cannot be observed in undifferentiated mESC models with inducible Xist, unless differentiation is induced (Bowness et al., 2022). These data suggest that differentiation is required for full XCI. Exactly why SMCHD1 is not recruited to the Xi in the pluripotent state, given that the Xi is H2AK119ub enriched, is unknown (Box 2; Fig. 3).

Additional less well-characterised chromatin marks also accumulate on the Xi, including H3K9 methylation and H4K20me1. It is unknown how early H3K9 methylation accumulates on the Xi; however, it is clear that H3K9me2/3 forms domains unoccupied by H3K27me3 in both mice (Ichihara et al., 2022; Keniry et al., 2016) and humans (Chadwick and Willard, 2003, 2004; Nozawa et al., 2013). Whereas polycomb accumulates in gene-rich regions, where Xist tends to be located (Engreitz et al., 2013; Simon et al., 2013), H3K9 methylation is enriched in intergenic regions, repetitive regions and gene deserts. That H3K9 methylation is enriched where Xist is not suggests that another mechanism results in its enrichment domains.

Although the H3K9 methylation targeting mechanism is unknown, two studies show SETDB1 is the main dose-dependent H3K9 methyltransferase in silencing of the Xi in mice (Keniry et al., 2016; Minkovsky et al., 2014), both using screen-based approaches to identify a role for SETDB1 in maintenance of the silent state. SETDB1 deletion also impacted XCI in vivo early post-implantation and during the establishment of XCI in differentiating female mESCs, suggesting that it also plays a role earlier in XCI. This role is more pronounced than its role in maintenance of silencing (Keniry et al., 2016). Consistent with SETDB1 being the predominant player, Ohhata and colleagues found no role for the facultative H3K9 demethylase G9A in XCI in vivo (Ohhata et al., 2004). However, a recent study in female mESCs and their differentiated progeny suggests that G9A is required for Tsix repression of Xist, and that G9A binds Xist RNA to help mediate silencing (Szanto et al., 2021). These results are challenging to reconcile with the in vivo studies. The authors suggest potential in vivo functional redundancy with other H3K9 methyltransferases, so whether there is a role of G9A in H3K9me2 on the Xi remains of interest.

One of many open questions is how H3K9 methylation is targeted to the Xi. H3K9me3 is also found on the active X chromosome, so perhaps the question is really how it accumulates on the Xi beyond what is found on the Xa. Ichihara and colleagues recently showed that SMCHD1 is required for appropriate H3K9me3 distribution, as SMCHD1-deficient epiblast cells and fibroblasts did not possess expanded H3K9me3 domains on the Xi. H3K27me3 and H2AK119ub accumulate on the Xi in the absence of SMCHD1 beyond their usual levels and spread into regions not normally marked by polycomb (Gdula et al., 2019; Ichihara et al., 2022; Jansz et al., 2018a). SMCHD1 is known to bind the HP1-binding protein LRIF1, providing a link with H3K9me3 (Brideau et al., 2015; Nozawa et al., 2013). Smchd1 null cells fail to accumulate LRIF1 on the Xi, whereas Lrif1 null cells retain SMCHD1 enrichment on the Xi (Brideau et al., 2015; Ichihara et al., 2022). Whether SMCHD1 directly enables H3K9me3 spreading or the direct effect of SMCHD1 is in limiting H3K27me3/H2AK119ub is unknown. In either case, SMCHD1 appears to allow establishment of a chromatin state that enables appropriate enrichment of H3K9me3 on the Xi (Fig. 3).

CDYL, a chromodomain protein that binds H3K9me2/3 and H3K27me3 peptides and reconstituted nucleosomes, is also enriched on the Xi (Bartke et al., 2010; Escamilla-Del-Arenal et al., 2013; Franz et al., 2009; Vermeulen et al., 2010). Its enrichment occurs downstream of Xist, H3K27me3 and H3K9me2, and, just as for polycomb and SMCHD1, is not dependent on Xist Repeat A. Similarly to SMCHD1, CDYL recruitment and H3K9me2 enrichment on the Xist-expressing chromosome is dependent on differentiation (Escamilla-Del-Arenal et al., 2013). It remains to be determined how these factors work to lay down, read and spread H3K9 methylation during differentiation.

H4K20me1 is another Xi mark about which rather little is known (Kohlmaier et al., 2004). SET8 (KMT5A) is the sole enzyme responsible for H4K20me1, and this mark is known to have a role in chromatin compaction, both during mitosis and interphase (Xiao et al., 2005). Set8^−/− mouse embryos do not survive past the 8-cell stage, indicating a vital role in normal development in males and females, but precluding further study of XCI in the females (Oda et al., 2009). Recent live imaging has shown that H4K20me1 becomes enriched on the Xi with similar dynamics to H3K27me3, although the level of H4K20me1 reaches its maximum much earlier than H3K27me3 in the inducible Xist model used. Just like polycomb and SMCHD1, H4K20me1 is dependent on Xist Repeat B/C (Tjalsma et al., 2021). Given the Xist Repeat A deletion mutant still accumulates each of these marks yet cannot effect silencing, these data suggest that SMCHD1, polycomb and H4K20me1 are not sufficient for establishing early-stage silencing of the Xi. Rather, H4K20me1 may have a role later in XCI.

The Barr body, representing a densely compacted Xi, was first identified in cats (Barr and Bertram, 1949). Although the Barr body is present in most mammals, no obvious Barr body is visible in rodents, bringing into question the necessity of chromatin compaction for gene silencing (Moore and Barr, 1953). The Xi is approximately 1.2-fold more compressed than the active X chromosome, in mouse and human (Collombet et al., 2020; Giorgetti et al., 2016; Teller et al., 2011), so, although a compacted rodent Xi was not apparent in early microscopy, it is visibly compacted by modern methods. The Xi forms a repressive nuclear compartment that excludes RNA polymerase II (PolII) (Chaumeil et al., 2006); however, exclusion of PolII is not by physical exclusion, but simply due to absence of transcription (Collombet et al., 2023). How compaction occurs is unclear. Studies have found that PRC1 may (Markaki et al., 2021) or may not (Colognori et al., 2019; Nozawa et al., 2013) be required for compaction. Likewise, SMCHD1 may (Nozawa et al., 2013) or may not (Jansz et al., 2018a) play a role. Potentially, H4K20me1 also plays some role in compaction (Tjalsma et al., 2021). Differences in techniques, cell types, cell models and species likely explain the reported discrepancies, but for now the role of compaction in silencing and the factors required remain open questions.

The Barr body was first noticed localised to the peri-nucleolar region within the nucleus (Barr and Bertram, 1949; Bourgeois et al., 1985; Petersen and Therelsen, 1962), with subsequent studies finding the Xi more often at the nuclear periphery (Barton et al., 1964; Belmont et al., 1986; Borden and Manuelidis, 1988; Dyer et al., 1989; Rego et al., 2008; Zhang et al., 2007). Whether this nuclear localisation is important for silencing is unclear as the active X is also regularly at the nuclear periphery (Dyer et al., 1989; Eils et al., 1996). More recently, the requirement of Xi localisation for gene silencing has again come under debate owing to a series of seemingly contradictory studies. Proteomic experiments find that Xist binds the peripherally located lamin B receptor (LBR) (McHugh et al., 2015; Minajigi et al., 2015), with depletion of LBR or deletion of the LBR-binding site within Xist Repeat F resulting in delocalisation of the Xi from the nuclear lamina and failure of gene silencing (Chen et al., 2016a). Other studies, however, found only limited loss of silencing in the absence of LBR in mESCs despite defective localisation of the Xi (Nesterova et al., 2019; Young et al., 2021) and artificial tethering of the X inactivation centre (XIC) to the lamina does not induce stable gene silencing (Pollex and Heard, 2019). Furthermore, mice lacking LBR do not show a female-specific defect (Cohen et al., 2008; Shultz et al., 2003; Young et al., 2021).

The XIC is a section of the X chromosome defined as containing the region required to establish monoallelic X-chromosome silencing. It was first narrowed down to a region of several hundred kilobases using BAC transgenes (Heard et al., 1999). Modern chromosome capture techniques (HiC) precisely mapped the mouse XIC to an 800-kilobase region split into two topologically associated domains (TADs), where genes that promote the inactive state (including Xist) and those that promote the active state (including the Tsix promoter) separate into different TADs (Nora et al., 2012) (Fig. 4), explaining the original BAC boundaries. The Xist-containing TAD is highly conserved in human; however, the Tsix TAD, like the gene itself, is less conserved (Bonev et al., 2017; Galupa and Heard, 2018; Rao et al., 2014). Topological partitioning of positive and negative controllers of XCI helps facilitate their differential regulation, with an inversion of the TAD boundary that places the Xist and Tsix promoters in their opposing TADs, leading to a switch in promoter interactions and mis-regulation of their expression (van Bemmel et al., 2019).

Like the XIC, the Xi is also split into two domains. This separation was apparent in early microscopy studies and, much like other molecular features of the Xi, has been brought into focus by HiC experiments, which show that the Xi forms two large bipartite megadomains hinged at the Dxz4 macrosatellite repeat in both human and mouse (Deng et al., 2015; Giorgetti et al., 2016; Rao et al., 2014). In humans, the DXZ4 loci on the active and inactive X are conspicuous in their opposing chromatin states compared with the rest of the chromosome: H3K9me3 and DNA methylation on the active X and H3K4me2 and DNA hypomethylation on the Xi (Chadwick, 2008). Similarly, the structural protein CTCF is enriched at DXZ4 on the Xi, despite being depleted from the rest of the chromosome (Chadwick, 2008; Moseley et al., 2012; Yang et al., 2015). In mouse, CTCF is also bound to the Dxz4 locus on the Xi; however, the epigenetic state of the locus is unclear (Horakova et al., 2012).

Within the Xi megadomains there is a reported depletion of both compartmental structure (A and B compartments, broadly representing active and inactive regions) and TAD structures, which are typical of both the active X and autosomes. TADs often suggest active promoter and enhancer interactions and can be seen around escapee genes (Deng et al., 2015; Giorgetti et al., 2016; Rao et al., 2014; Splinter et al., 2011). More recent work shows that compartments and TADs can be detected on the Xi using higher resolution analyses, although they are still depleted compared with other chromosomes (Bauer et al., 2021).

How are TADs blocked on the Xi? It seems partly an active process requiring SMCHD1, which antagonises binding of the TAD-forming protein CTCF (Chen et al., 2015). Without SMCHD1, CTCF accumulates on the Xi (Gdula et al., 2019; Wang et al., 2018), compartments become detectable, short-range interactions are strengthened and TADs detected. Therefore, SMCHD1 is thought to enable ultra-long-range chromatin interactions beyond the scale of TADs, which may antagonise TADs (Gdula et al., 2019; Jansz et al., 2018a; Wang et al., 2019a, 2018). More detail has recently been revealed, as Smchd1 null neural stem cells differentiated from their null counterpart mESCs show relaxation of some peripheral domains on the Xi, such that they are outside the Xist domain. These domains correlate with the chromosomal regions housing genes that are reactivated without SMCHD1 (Poonperm et al., 2023). However, SMCHD1 is not required to maintain XCI, providing an opportunity to discriminate architectural changes with and without altered transcription (Gdula et al., 2019; Jansz et al., 2018a; Wang et al., 2019a). Upon SMCHD1 removal during maintenance of XCI, with no changes in gene expression, similar changes in Xi chromatin architecture are observed, although potential relaxation of peripheral regions of the Xi has not been tested in this scenario. These data suggest that the changes observed are not due to transcription at reactivated genes. One proposed hypothesis is that SMCHD1, as a non-canonical SMC protein, may act much like the SMC2/4 condensin complex and dissolve TAD structure (Gdula et al., 2019; Gibcus et al., 2018). The opposition of CTCF binding likely contributes to SMCHD1 preventing TAD formation, although this cannot only be due to hypomethylation of CTCF sites leading to enhanced CTCF binding, as deletion of SMCHD1 during maintenance of XCI does not alter DNA methylation, but can alter CTCF binding (Gdula et al., 2019; Jansz et al., 2018a). It is tempting to speculate that the structures SMCHD1 enables on the Xi facilitate spreading of H3K9me2/3 or prevention of H3K27me3 spreading. This remains to be determined, but our neomorphic SMCHD1 model, which has a hypomorphic effect on Xi structure (similar to the null) alongside decreased H3K27me3 enrichment on the Xi, suggests that long-range interactions and the effects on chromatin structure may not be related functionalities of SMCHD1 (Tapia Del Fierro et al., 2023) (Box 2; Fig. 3).

That deletion of SMCHD1 alters chromosome structure but not maintenance of Xi gene silencing suggests that the structure of the Xi is not crucial for silencing: deletion of Dxz4 in mouse during establishment of XCI or in vivo results in loss of the bipartite structure but no loss of gene silencing or change in viability (Andergassen et al., 2019; Bonora et al., 2018; Fang et al., 2020; Froberg et al., 2018; Giorgetti et al., 2016); silencing remains stable when DXZ4 is deleted in human cells during maintenance of XCI (Darrow et al., 2016); and in vivo the two megadomains form in post-implantation embryos after gene silencing (Collombet et al., 2020). These data are consistent with the megadomain structure not being directly related to gene silencing.

The role of chromatin modification in transcriptional silencing of the X chromosome is complex, with factors likely playing overlapping and redundant roles to enable a life-long stable and mitotically heritable inactive chromosome. This is crucial for mammalian life; however, it impedes the laboratory study of XCI. Following several days of Xist expression, there is a developmental switch whereby XCI transitions from Xist dependent and reversible to Xist independent and irreversible (Wutz and Jaenisch, 2000) (Fig. 1). Indeed, deletion of Xist during maintenance of XCI leads to very limited gene reactivation in somatic cells (Adrianse et al., 2018; Brown and Willard, 1994; Csankovszki et al., 1999; Hong et al., 2017; Splinter et al., 2011). Likewise for deletion of Spen (Dossin et al., 2020). However, even the limited reactivation observed in these models would suggest some role for Xist in maintenance of XCI and warrants further investigation.

Several chromatin modifications appearing late in XCI contribute to maintenance of the silent state, creating a chromosome refractory to reactivation. These include replacement of histone 2A (H2A) with the variant macroH2A (Chadwick and Willard, 2002; Chaumeil et al., 2002; Costanzi and Pehrson, 1998; Mermoud et al., 1999; Pasque et al., 2011; Rasmussen et al., 2000). However, ablation of macroH2A late in differentiation of mESCs (Tanasijevic and Rasmussen, 2011) and in vivo in embryos (Changolkar et al., 2007; Pehrson et al., 2014) showed no obvious defect in XCI, with female null mice viable and fertile. MacroH2A can be monoubiquitylated at K115/K116, but the role of ubiquitylated macroH2A in XCI is unclear (Hernandez-Munoz et al., 2005; Zhuang et al., 2009).

DNA methylation in the CpG context by DNMT3B (Yagi et al., 2020) also occurs late in XCI, with promoter-associated CpG islands becoming hypermethylated (Gendrel et al., 2012; Norris et al., 1991; Sado et al., 2000; Weber et al., 2007). Curiously, hypermethylation at promoter elements contrasts with the rest of the Xi, with gene bodies and gene-poor intergenic regions hypomethylated (Hellman and Chess, 2007; Weber et al., 2007). Gene body methylation appears to correlate passively with transcription, whereas promoter hypermethylation relates to monoallelic X expression. The deposition of DNA methylation at CpG islands of the Xi is developmentally regulated in an SMCHD1-dependant manner, with some islands becoming methylated early independently of SMCHD1 and others that require SMCHD1 to acquire methylation late (Gendrel et al., 2012). Studies of mice unable to maintain DNA methylation via knockout of Dnmt1 confirm the crucial requirement of DNA methylation during XCI in the mammalian embryo (Sado et al., 2000).

It is likely we need to relieve maintenance of XCI before we can truly appreciate the roles of each unique factor in the maintenance phase. Current single factor depletion or deletion studies miss redundancies and networks, making studies that disrupt networks of silencing factors a crucial next step in deciphering the hierarchy of epigenetic silencing. Inhibition of DNMT1, in combination with depletion of a gene of interest, has allowed researchers to observe that reactivation of silencing not possible with gene depletion alone (Carrette et al., 2018a; Keniry et al., 2016; Lessing et al., 2016; Minajigi et al., 2015; Minkovsky et al., 2014; Wang et al., 2019a). Moreover, deletion of Xist in B cells causes loss of silencing at genes low in promoter DNA methylation, but not those with high DNA methylation (Yu et al., 2021). Although DNA methylation is clearly a key player in XCI maintenance, the switch to Xist-independent XCI occurs early in mESC differentiation (Wutz and Jaenisch, 2000), prior to DNA hypermethylation, suggesting additional underlying and redundant maintenance mechanisms. Recently, it has been revealed that Xist nucleates a locally dense protein gradient around the Xi, explaining how chromosome-wide silencing at hundreds of target genes is achieved from relatively few Xist RNA molecules (Jachowicz et al., 2022; Markaki et al., 2021; Pandya-Jones et al., 2020). Pertinent to maintenance of XCI, several RNA-binding proteins are recruited to the Xi by Xist Repeat E, where they form a condensate through self-aggregation and protein–protein interactions. This condensate is required for gene silencing at the time of the switch to Xist independence and for heritability of the silent state in an Xist-independent manner (Pandya-Jones et al., 2020). The mechanism by which a condensate of ubiquitously expressed RNA-binding proteins elicits epigenetic memory of repression is unclear, but potentially crucial to understand for maintenance of XCI.

Understanding the full repertoire of X chromosome silencing factors may enable modulation of X-linked genes, both reactivation of genes subject to inactivation and silencing of genes that escape, for the treatment of X-linked genetic disorders. This is particularly enticing in immune and neural cells. ‘XX’ individuals are substantially over-represented in many autoimmune disorders, with a potential cause being bi-allelically expressed genes that escape XCI, suggesting that these individuals may benefit from monoallelic silencing of escape genes in immune cells (Sierra and Anguera, 2019). Of all X-linked genes, some 20% have been linked to neurodevelopmental disorders and the XCI status of a particular gene, either inactive or escaping, can influence neurodevelopmental disease severity, suggesting that both reactivation and silencing of X-linked genes or influencing skewing of XCI may be beneficial in neural cells (Brand et al., 2021). Indeed, skewing of XCI can account for previously undiagnosed neurodevelopmental disorders (Giovenino et al., 2023). The most extreme case of manipulating dosage compensation has been in Down syndrome, where it is possible to enforce chromosome 21 silencing using XIST (Jiang et al., 2013), providing excitement for the future therapeutic manipulation of XCI mechanisms in aneuploidy.

There has been progress towards reactivating the X chromosome for therapeutic gain in mouse. Particularly, reactivation of Mecp2 in the neurodevelopmental disorder Rett syndrome, with depletion of Xist in combination with small molecule inhibitors of DNA methylation, or an inhibitor of the JAK/STAT pathway alone, can reactivate several genes, including Mecp2 (Carrette et al., 2018a,b; Lee et al., 2020). Excitingly, Xist RNA itself is targetable by small molecule inhibitors, with a study demonstrating that the compound ‘X1’ blocks initiation of XCI by binding to Xist, thereby greatly improving the possibility of XIST modulation therapeutically (Aguilar et al., 2022). Mechanisms to enhance global Xi reactivation during maintenance may also exist. For example, depletion of SPEN increases the level of escape gene expression (Dossin et al., 2020), and the BAF complex member SMARCA4 promotes pharmacological reactivation of the Xi (Jegu et al., 2019).

In many cases, reactivation of the entire X chromosome may be too blunt an approach. Instead, well-informed and precise locus-specific modifications may be more appropriate. Most of the knowledge we have about XCI comes from mice and a limited set of tissues, yet XCI differs between species and to some degree cell types. Species are especially divergent with regard to genes that escape XCI, with substantially more genes escaping silencing in humans than mice (Jacobson et al., 2022). Broadly, approximately 3% of genes in mouse and 20% in human escape silencing, with some genes constitutively escaping in all cells and others variably escaping between cell types and individuals (Balaton and Brown, 2016; Berletch et al., 2015; Carrel and Willard, 2005; Marks et al., 2015; Tukiainen et al., 2017). There are mechanistic differences too as loss of XIST impairs differentiation of human mammary stem cells (Richart et al., 2022), whereas mice are thought to largely tolerate Xist loss (Yang et al., 2020). XCI mechanisms also have some cell-type specificity, for example XCI is maintained in B cells despite Xist delocalisation (Syrett et al., 2017, 2019). Recently, CRISPR/Cas9 targeting of the gene activator VP64 together with the demethylater Tet1 switched on the inactive copy to 60% of the active X copy for an X-linked gene associated with infantile epilepsy, CDKL5, in human neuronal-like cells in vitro (Halmai et al., 2020). Together, these data suggest that with targeted, and likely lineage-specific, approaches there is potential for targeting dosage compensation to treat disease if we understand the human process.

Interestingly, the Xi provides its own natural template for the specific activation of genes. Escape genes reveal the chromatin modifications required for expression in the otherwise repressive landscape of the Xi. Escape genes have low Xist enrichment (Engreitz et al., 2013; Simon et al., 2013), avoid the chromatin modifications associated with repression and display chromatin modifications associated with activity (Carrel and Brown, 2017). Chromatin at escape genes is accessible by ATAC-seq (assay for transposase-accessible chromatin with sequencing) and DNAseI (Calabrese et al., 2012; Giorgetti et al., 2016; Qu et al., 2015), is enriched in CTCF binding and shows local TAD formation (Bonora et al., 2018; Giorgetti et al., 2016). Whether these local structures and chromatin modifications could aid activation, or are simply a by-product of transcription, is an open question. By-product or not, it is tempting to speculate that the absence of transcription itself may be a barrier that reinforces heterochromatic pathways, and the act of transcription provides a positive feedback loop to maintain gene activity. Multi-factor epigenome editing will be required to resolve these questions.

More than 60 years on from the Lyon hypothesis, XCI has become a paradigm for epigenetic and RNA-directed silencing, and rightly remains intensely studied with previously unknown facets of the silencing mechanism reported frequently. The chromatin of the future Xi becomes heavily modified in a highly organised hierarchy of overlapping and redundant features, which results in a repressive state that is very hard to overcome experimentally. The challenge of the next 60 years (ideally fewer) will be to take our current knowledge and use it to reverse X-chromosome repression at specific gene loci and in specific cell types, to cure disease by unlocking the suite of genes that reside silently in the female genome.

We thank Natasha Jansz for valuable feedback on the manuscript. Figures created with BioRender.com.