The Implication of Early Chromatin Changes in X Chromosome Inactivation

1Successful development requires the establishment and maintenance of euchromatin and heterochromatin in different parts of the genome. Progressively stable silencing of some genic regions occurs as epigenetic memory continues to accrue, finally leading to facultative heterochromatin formation. Multiple layers of chromatin modifications are believed to enable stable transcriptional silencing. However, little is known about how facultative heterochromatin is dynamically formed and to what extent chromatin changes are involved in the establishment of gene silencing. A powerful model for developmentally induced gene silencing and formation of facultative heterochromatin is X chromosome inactivation (XCI) in female mammals. Although the role for chromatin changes in maintenance of the inactive state has been extensively studied, almost nothing is known about their role in the initiation of gene silencing.

2In female mouse embryos, one of the two X chromosomes is randomly chosen for inactivation around the time of implantation. This phenomenon is dependent on the coating of the future inactive X chromosome (Xi) by the long non-coding RNA Xist. The conserved A-repeat region of Xist mediates transcriptional silencing, whereas other parts of Xist ensure chromosome coating. Assays coupling immunofluorescence with RNA fluorescence in situ hybridization (IF/RNA FISH) have revealed that, upon Xist RNA coating, a program of striking chromatin rearrangements ensues. These include rapid loss of histone modifications associated with active promoters (H3K4me3, H4ac, H3K9ac, and H3K27ac) and enhancers (H3K27ac and H3K4me1). Furthermore, there is accumulation of H2AK119Ub and H3K27me3, two repressive histone marks dependent on the activity of Polycomb repressive complex (PRC) 1 and 2, respectively (for a review, see ). These progressive alterations of chromatin states are associated with stable repression of the majority of genes on the Xi. Not all X-linked loci are affected in the same way, however, with some genes being silenced much faster than others or even entirely resisting repression (“escapees”). The reason for this striking diversity in gene inactivation dynamics remains unclear. It is likely that the susceptibility of loci to Xist spreading plays a role in this process; however, differences in their chromatin status could also underpin transcriptional silencing dynamics.

3Although some chromatin marks have previously been mapped on the X chromosome in female embryonic stem cells (ESCs) and differentiated cells, nothing is known about the dynamics of the XCI process, especially during early stages when gene silencing actually takes place. Specifically, the order of chromatin changes and their possible role(s) in mediating transcriptional silencing remain largely unknown. Previous studies have relied on ESC differentiation for XCI initiation, which results in highly asynchronous Xist induction and elevated levels of heterogeneity in the population. These complications entirely mask primary chromatin events occurring in a small subset of cells embarking on the process of XCI. Recent advances in the identification of Xist binding proteins revealed that most histone-modifying activities are not directly recruited by Xist RNA but, rather, by its binding partners, like SPEN and HNRNPK. A major question is whether the chromatin changes that occur early on in XCI are actually involved in gene silencing during XCI and, if so, how.

4Here we investigate the earliest events accompanying transcriptional silencing of the X chromosome. Using a female ESC line in which one X chromosome can be specifically inactivated via an inducible Xist gene, we performed allele-specific native chromatin immunoprecipitation sequencing (ChIP-seq) as well as nascent transcript profiling at the initiation stages of XCI. Using 4-hr time resolution, we were able to reveal the precise order of chromosome-wide epigenetic events that are intricately coupled to gene repression during XCI. We find that loss of histone acetylation and, in particular H3K27ac, is one of the first events following Xist RNA accumulation during initiation of XCI. We also show that histone deacetylation, via HDAC3, is vital for efficient silencing of most genes on the Xist-coated X chromosome. Contrary to the prevailing hypothesis, HDAC3 is not acutely recruited to the X chromosome by Xist; rather, it is pre-loaded, mainly at putative enhancers. Upon Xist coating, pre-loaded HDAC3 likely mediates histone deacetylation and facilitates transcriptional silencing. We also uncover a surprisingly rapid accumulation of the PRC1-dependent H2AK119Ub mark on the X chromosome, particularly at intergenic regions lying in proximity to Xist RNA entry sites, which are pre-marked by Polycomb (PcG) marks. Accumulation of H3K27me3 appears slightly later and is delayed compared with gene silencing. Using Hdac3 and Xist mutant cells, we also show that spread of Polycomb marks into gene bodies can only occur when transcriptional repression begins. Our study reveals the chromatin choreography across the X chromosome during the early steps of XCI initiation and uncovers a role for histone deacetylation in the establishment of gene silencing.

5A molecular roadmap of the earliest chromatin changes during the initiation of XCI has not yet been established. To reduce heterogeneity and obtain a highly resolved time series of chromatin events upon Xist RNA coating, we used the TX1072 ESC line (Figure 1A; ). This hybrid (Mus musculus castaneus x C57BL/6) line harbors a doxycycline (DOX) inducible promoter upstream of the endogenous Xist at the C57BL/6 (B6) allele. By adding DOX, we synchronously induce Xist expression and can thus decouple the onset of XCI from differentiation-dependent chromatin changes (Figure S1C). We performed allele-specific native ChIP-seq (nChIP-seq) for seven histone modifications across five time points at up to 4-hr resolution on biological duplicates (Figure 1A; Figures S1A and S1B). We focused on histone modifications associated with active promoters (H3K4me3, H3K9ac, H4ac, and H3K27ac) and active enhancers (H3K4me1 and H3K27ac) as well as repressive Polycomb-dependent marks (H3K27me3 and H2AK119Ub). All nChIP-seq datasets were sequenced to a depth of at least 40 million reads and showed a high signal-to-noise ratio (Figures S1B and S1E). The reads were split according to content of allele-specific SNPs (B6, mapping to Xi; Cast, mapping to active X); this allelic information was then analyzed in detail to define XCI-specific changes.

6Upon Xist coating, the Xi becomes rapidly depleted of active histone modifications. Although this global loss of euchromatin occurs in a similar time window as the changes in X-linked transcriptional activity, the precise chronology and possible mechanistic relationship of these two events have remained unclear. To address this, we first analyzed nChIP-seq datasets for a panel of active histone modifications (H3K27ac, H3K4me1, H3K4me3, H3K9ac, and H4ac; Figure S1E). This revealed progressive depletion of B6-specific reads (originating from Xi) upon Xist induction, with different marks showing different kinetics of loss (Figure S1D). To analyze this in detail, we performed peak calling on all the reads (STAR Methods). For each peak with at least 50 allele-specific reads, we calculated its d-score (readsB6/[readsCast + readsB6]). This parameter of allelic skewing uses the active X chromosome as a powerful internal control, rendering the analysis less sensitive to variability in nChIP-seq quality and, thus, more stable between replicates.

7We first set out to track the relative dynamics of active chromatin marks by comparing d-score dynamics of all peaks. We found that loss of H3K27ac and H4ac is among the earliest and most robust chromatin events on the X chromosome following Xist induction (Figure 1B). Furthermore, we found that not only did different histone modifications show distinct behaviors but, also, that specific regions of the X chromosome followed rather different dynamics of inactivation (Figure S1G). Thus, there is a temporal hierarchy of chromatin events both at the level of different histone marks and genomic loci.

8To better understand in which context chromatin changes might affect gene silencing, we separately focused on promoters and putative enhancers. For each window, we plotted a d-score (normalized to t = 0 hr) as a function of time and fitted a sigmoidal curve (STAR Methods). We found that the maximal dynamic range of inactivation was from 0.5 to ∼0.2 and, thus, set a threshold for when the curve reached 50% of this dynamic range; i.e., d = 0.35 (STAR Methods). This IC35 (inactivation criterium 0.35) parameter indicates how quickly a given peak becomes significantly decreased or lost (Figure S1F). Pairwise comparisons of the IC35 at X-linked gene promoters revealed that H3K27ac becomes efficiently deacetylated prior to demethylation of H3K4me3 (Figure 1C). This trend holds true for ∼88% of measured promoters (Figure S1H), including a representative example of the Pdk3 promoter (Figure 1D). Similar analysis of the enhancer inactivation dynamics revealed that H3K27ac is efficiently depleted prior to loss of H3K4me1 (Figure 1C). This robust difference is observed in ∼95% of the cases; e.g., at a putative Pdk3 enhancer element (Figure 1E; Figure S1I). Thus, histone deacetylation is an early and efficient event during XCI.

9Next, to extract information about relative timing rather than the efficiency threshold, we obtained the time when the curve reaches its maximum slope (effective dose 50%, ED50; Figure S1F). ED50 analysis revealed that both H3K27ac and H3K4me3 become depleted concomitantly (Figure S1J). Together with the IC35 data, this indicates that, even though demethylation of promoters and their deacetylation are concurrent, the latter process is significantly more efficient. Furthermore, unlike promoters, enhancers showed significant differences in the ED50 parameter between H3K27ac and H3K4me1, indicating that enhancer demethylation is not only less efficient but also significantly delayed (Figure S1J).

10To assess the relative deacetylation dynamics of enhancers and promoters, we compared the IC35 parameter for H3K27ac at both types of genomic regulatory elements. Intriguingly, enhancer deacetylation was slightly more efficient than that of promoters (Figure S1K). After pairing active promoters to their putative enhancers (STAR Methods), ∼76% (37 of 49) of them showed more efficient deacetylation at enhancers than at promoters (Figure S1M). On the other hand, loss of H3K4me1 on the Xi, which is indicative of enhancer decommissioning, was found to be significantly less efficient compared with the loss of H3K4me3 at promoters (Figure S1L). Thus, inactivation (loss of H3K27ac) of enhancers proceeds at a slightly higher efficiency than of promoters, although stable decommissioning (loss of H3K4me) of promoters precedes that of enhancers. Intriguingly, upon H3K4me3 loss following Xist induction, we observe a slight increase in H3K4me1 levels around transcriptional start sites (TSSs), which probably reflects an intermediate of stepwise enzymatic H3K4me3 demethylation (Figures S1N and S1O).

11In summary, our data show that efficient histone deacetylation at promoters and enhancers is one of the first events during XCI, indicating that it might contribute to initiation of transcriptional silencing. In comparison, demethylation of H3K4me3 and H3K4me1 is less efficient and delayed, respectively. Finally, deacetylation of promoters and enhancers seems to follow broadly comparable dynamics, whereas their decommissioning differs significantly.

12To gain insight into the relationship between loss of active chromatin marks and transcriptional silencing of X-linked genes, we compared our nChIP-seq dataset with transient transcriptome sequencing (TT-seq; ). This allows a direct measure of newly synthesized RNA and, thus, enabled us to measure transcriptional activity as the X chromosome becomes inactivated. Comparison of the IC35 for TT-seq and individual histone modifications at associated promoters revealed that transcriptional silencing dynamics are most tightly linked to promoter deacetylation at H3K27 and, to a lesser extent, to loss of H4ac, H3K9ac, and H3K4me3 (Figures 2A and 2B; Figure S2A). A similar pattern of gene silencing and promoter deacetylation dynamics further exemplifies this correlation (Figure 2C).

13The striking variability in dynamics of promoter chromatin changes begged the question of whether these differences could be due to transcriptional status or genomic context; for example, proximity to Xist (the locus and/or Xist RNAs entry sites; ), gene density, or LINE (long interspersed nuclear elements) density. Indeed, the latter feature has been proposed previously to influence the efficiency of XCI (,; Figure S2B and S2C). We found that gene promoters that are rapidly H3K27-deacetylated preferentially reside in gene poor, LINE-dense regions in proximity to lamina-associated domains (LADs) and the Xist locus (Figure 2D).

14Given that the histone deacetylation is the primary chromatin change observed upon Xist induction and that its kinetics are tightly linked to transcriptional silencing, we decided to investigate whether the loss of histone acetylation might play a functional role in mediating the initiation of gene silencing during XCI.

15To investigate whether this rapid histone deacetylation is mediated by histone deacetylases (HDACs), we first tested selective pharmacological inhibitors against all HDACs expressed in ESCs (Figure S3A). Importantly, after 24 hr of DOX induction, only HDAC3 catalytic inhibition resulted in a significant silencing defect compared with the control (DMSO) (Figure 3A). We further validated these findings using nascent RNA FISH and, unexpectedly, found that HDAC1 and 2 inhibition promoted more efficient Xist upregulation, possibly because of rtTA upregulation (Figure 3B). The finding that HDAC3 is involved in XCI is in line with previous reports that it might interact with SPEN, which is one of the key Xist-silencing factors. Also, Hdac3 knockdown was recently shown to result in defective silencing of another X-linked gene: Gpc4.

16To better interrogate the specific functions of HDAC3 in XCI, we generated two independent Hdac3−/− TX1072 deletion clones using CRISPR/Cas9 technology (STAR Methods; Figure 3C). We performed RNA sequencing (RNA-seq) on these, both prior to and 24 hr following DOX addition. Differential gene expression analysis revealed relatively mild changes apart from some meiosis-associated genes (Figures S3B–S3D). Consistent with our inhibitor treatment results, allele-specific analysis revealed that both Hdac3−/− clones are unable to efficiently silence X-linked genes (Figure S3E). Although the majority of genes across the X chromosome were affected, ∼24% of genes (43 of 182) were not, showing, instead, almost normal silencing (Figure 3D; Figures S3F and S3G). We further validated our findings in the context of EpiLC (epiblast like-cells) differentiation both with and without DOX treatment (Figures 3E–3G; Figure S3H). In our pyrosequencing experiments, we monitored the allelic bias in expression of genes affected (AtrX, G6pdX, and Pgk1) in their silencing by the HDAC3 loss and three less affected genes (Ints6l, Rnf12, and Elf4) (Figure 3E).

17Our results indicate that loss of HDAC3 leads to a delay in transcriptional silencing of most X-linked genes. We noted that, over time, XCI is not fully prevented because gene silencing progressively ensues, although more slowly than in control cells (Figure 3E), indicating that mechanisms parallel to HDAC3 must also contribute to XCI. To address this further, we tested whether other HDACs might compensate for HDAC3 loss. To this end, we treated Hdac3−/− ESCs with selective inhibitors against all expressed HDACs (1,2,3,4,5,6, and 8). We found that allelic skewing (i.e., XCI) of AtrX, G6pdX, and Rnf12 were not affected in any of the treated cells (Figure 3H), arguing against the involvement of other HDACs in XCI. Thus, our data reveal that HDAC3 is the main histone deacetylase important for the initiation of silencing of most genes on the X chromosome. Because HDAC3 can mediate deacetylation of many substrates, we next investigated whether its effect on gene silencing is mediated by changes in chromatin status.

18Although our experiments using inhibitors and knockout ESCs indicated the importance of HDAC3 catalytic activity in mediating efficient XCI, it remained unclear whether histone deacetylation dynamics were also affected by loss of HDAC3. To address this question, we first monitored the global distribution of histone modifications by IF/RNA FISH. We observed that the Xi silent compartment formed much less efficiently in Hdac3−/− ESCs, with increased levels of H3K27ac and H4ac remaining within the Xist cloud (Figures 4A and 4B; Figures S4A–S4D). To further explore the chromatin phenotype of Hdac3−/− ESCs, we performed nChIP-seq for H3K27ac and H4ac at 0, 8, and 24 hr of DOX treatment. Differential peak analysis has revealed that less than 9% of peaks significantly change in Hdac3−/− compared with WT cells at t = 0 hr (Figure S4E). Consistent with the IF/RNA FISH results, allele-specific analysis revealed delayed histone deacetylation in the Hdac3−/− clones compared with the parental line (Figure 4C; Figures S4F and S4G). As with transcriptional silencing, the deacetylation defects affected most of the X chromosome but were not uniform, with some peaks being almost entirely resistant to HDAC3 loss (∼29%, 38 of 133). Indeed, protracted loss of H3K27ac at promoters was linked to a severe defect in transcriptional silencing in mutant cells (Figures 4C and 4D). Together, our data show that HDAC3 mediates efficient transcriptional silencing of the majority of genes through deacetylating histones. However, it remains unclear how HDAC3 is targeted to the X chromosome to help silencing.

19It has been proposed that HDAC3 might be either de novo recruited to the Xi by Xist through interaction with the SPEN protein or, alternatively, that Xist-coating might lead to the activation of already pre-bound HDAC3. To test these two hypotheses, we measured sub-nuclear localization of HDAC3 after 24 hr of DOX treatment. IF analysis revealed no HDAC3 enrichment on the Xi and even a slight depletion (Figure 5A). This surprising result prompted us to further explore the question of HDAC3 recruitment. To circumvent problems with antibody quality, we knocked in a 3xFLAG tag into the 3′ region of the endogenous Hdac3 gene (Figure 5B). This novel ESC line (Hdac3FLAG/FLAG) expressed nearly normal levels of HDAC3 (Figure S5A) and robustly silenced the X chromosome, indicating that HDAC3 remained functional (Figure S5B).

20We then set out to identify potential common interactors of Xist and HDAC3 in DOX-treated cells. To this end, we performed anti-FLAG immunoprecipitation followed by quantitative label-free mass spectrometry analysis (Hdac3FLAG/FLAG versus Hdac3+/+ in five replicates) (Figure 5C). Among significant interactors, we identified nearly all components of the NCOR and SMRT complexes as well as their chaperone complex CCT (Figures 5D and 5E; ). Intriguingly, of all Xist binding partners, only RYBP (a PRC1 factor) showed a weak binding to HDAC3 (Figure 5C). Our results confirm the previously reported interactions between HDAC3 and NCOR and SMRT but argue against strong HDAC3 recruitment by Xist RNA binding partners like SPEN.

21To directly measure HDAC3 recruitment to the Xi, we performed anti-FLAG ChIP-seq prior to and during XCI. We found robust binding of HDAC3 to multiple H3K27ac and H4ac peaks throughout the genome (∼65% of HDAC3 peaks are enriched for H3K27ac [∼65% on the X chromosome] and ∼61% for H4ac [∼58% on the X chromosome]) (Figure 5F). The vast majority of HDAC3 peaks were intriguingly found to reside within putative enhancers rather than at gene promoters (Figure 5F; Figure S5C). Allele-specific analysis revealed a transient and modest enrichment after 4 hr of induction on the Xi (B6 reads) at the Xist locus and at some HDAC3 peaks (Figure 5G; Figures S5D and S5E). Together, this indicates that HDAC3 indirectly and unstably interacts with some of the Xist binding partners to allow for low-level and transient recruitment to the X chromosome. However, no significant correlation between HDAC3 de novo recruitment and timing of deacetylation was found (Figure S5F). On the other hand, the histone acetylation peaks that are lost early during XCI did show higher levels of HDAC3 pre-binding (Figure S5G). In fact, HDAC3-bound regions show more efficient (low IC35) deacetylation of H4ac and a similar but statistically not significant trend for H3K27ac (Figure 5H; Figure S5H). Finally, we observed a mild but statistically insignificant trend for HDAC3 pre-bound enhancers to surround rapidly silenced genes (Figure S5I). In summary, we show that HDAC3 does not appear to be robustly recruited de novo to the X chromosome via Xist RNA coating and does not strongly bind known partners of Xist. Rather, the majority of HDAC3 is pre-bound at putative enhancers, and, upon Xist induction, this results in efficient histone deacetylation, promoting transcriptional silencing.

22Repressive chromatin modifications are also known to appear shortly after Xist RNA coating, in particular H2AK119Ub (PRC1-dependent) and H3K27me3 (PRC2-dependent). However, the early dynamics of this process, as well as the relative timing of H3K27me3 and H2AK119Ub deposition, have remained largely unexplored. To address this question, we have performed nChIP-seq for both PcG-dependent marks. These datasets showed a reproducible and stable pattern of enrichment on autosomes, both between time points and biological replicates (Figure S6A). In contrast, the X chromosome showed a bias in H2AK119Ub enrichment toward the B6 allele (Xi) after 4 hr of DOX induction and after 8 hr in the case of H3K27me3 (Figure S6B). First, we analyzed relative B6-read enrichment within 10-kb windows across the whole X chromosome normalized to t = 0 hr and observed rapid deposition of H2AK119Ub preceding that of H3K27me3 (Figure 6A). To quantify the dynamics at individual windows, we extracted the ED50 parameter, as was done for active marks, but using B6-read accumulation (STAR Methods; Figure S1F). Pairwise comparison of ED50 at all 10-kb windows across the X chromosome showed that H2AK119Ub accumulates most efficiently prior to H3K27me3 (Figure 6B), with very few exceptions (∼1%) (Figure S6C).We also validated this finding using IF/RNA FISH during ESC differentiation toward EpiLCs (Figure S6D). Evidently, the Xi becomes enriched for H2AK119Ub prior to the PRC2-dependent H3K27me3 across the whole chromosome, although there may be local differences in the dynamics of this process. Indeed, certain regions preferentially accumulate Polycomb marks already at early time points (i.e., by 4 hr for H2AK119Ub or 8 hr for H3K27me3; Figure 6D). This early accumulation was found to be prominent around the Xist locus (green line) as well as the first regions with which it associates (“entry sites,” blue lines). There is also significant overlap between H2AK119Ub and H3K27me3 patterns of accumulation, and, interestingly, these primary deposition sites are typically marked by PcG prior to Xist upregulation (Figure 6C, Figure S6E). Thus, most H2AK119Ub accumulation occurs at sites pre-marked by PcG and rapidly coated by Xist. Subsequent H3K27me3 deposition closely follows this early deposition pattern.

23We next investigated the accumulation dynamics of both marks at specific types of genomic windows. ED50 analysis revealed that H2AK119Ub accumulates first intergenically and is deposited only later at active gene bodies, promoters, and putative enhancers (Figure S6F). This pattern of accumulation is tightly followed by H3K27me3 deposition and spread. What is more, PcG-dependent mark accumulation seems to be closely linked to gene silencing kinetics (Figure 6E). This is exemplified by tracks for genes that are either silenced rapidly (Rnf12) or slowly (Huwe1 and Alg13) or escape the process altogether (Kdm6a) (Figures 6F and 6G). Direct comparison of gene silencing and promoter deacetylation dynamics with PcG mark accumulation at gene bodies revealed a significantly lower ED50 parameter for H2AK119Ub (Figure S6G). This indicates that the H2AK119Ub deposition reaches its maximum rate earlier than gene silencing and promoter deacetylation. Next, we measured the time when the sigmoid’s second derivative reaches its minimum or maximum. We use this parameter as a proxy for when the curve starts to increase or decrease (Figure S6I). This revealed that gene silencing, promoter deacetylation (H3K27ac loss), and H2AK119Ub accumulation at gene bodies are all initiated at the same time, whereas H3K27me3 is significantly delayed (Figure S6H). Thus, we see dynamic spreading of H2AK119Ub from intergenic regions to gene bodies coinciding with transcriptional silencing.

24We also investigated genes that normally escape from XCI (constitutive) or escape more variably and in specific tissues (facultative escapees). We examined a highly expressed constitutive escapee (Kdm6a; Figures 6F and 6G) and found no PcG-dependent mark accumulation at its promoter or gene body in our dataset. In line with this, both constitutive and facultative escapees show no or lower accumulation of H2AK119Ub and H3K27me3 at their gene bodies (Figures S6J and S6K). This is consistent with significantly delayed H3K27ac loss at their promoters (Figure S6L). Thus, genes escaping transcriptional silencing do not undergo promoter deacetylation and do not accumulate PcG-associated marks.

25All in all, our analysis reveals the hierarchy of PcG marks during XCI, with H2AK119Ub accumulating significantly prior to H3K27me3. This process is first initiated at large chromatin domains that are pre-marked by PcG (megabase scale) and that correspond to some of the first regions with which Xist RNA interacts. We demonstrate that PcG marks accumulate first intergenically and subsequently spread into regulatory elements and gene bodies. This spread into genes coincides with transcriptional silencing, raising the question of whether the two processes are functionally linked.

26To address whether spreading of these repressive histone modifications from intergenic regions into genes is a cause or consequence of transcriptional silencing, we decided to investigate PcG mark distribution in a Xist mutant lacking the A-repeat element (Xist:ΔA). Xist:ΔA RNA is able to coat the X chromosome and lead to PcG mark accumulation (based on IF) but cannot initiate gene silencing. Xist:ΔA thus allows the relationship between repressive chromatin changes and transcriptional repression to be addressed. For this, we used previously published male ESCs harboring wild-type (TXY:Xist) or mutant Xist (TXY:XistΔA) under a DOX-inducible promoter at the endogenous locus (; Figure 7A). These ESCs do not require allele-specific analysis, as all X-mapping reads will originate from the single X chromosome. After collecting DOX-treated and untreated differentiated cells (48 hr), we performed nChIP-seq for three histone modifications (H3K27ac, H3K27me3, and H2AK119Ub) in biological duplicates. Samples were normalized for ChIP quality prior to further analysis (STAR Methods).

27First, we investigated the relationship between PcG-associated mark spread and transcriptional silencing in WT and Xist:ΔA cells. We observed that most regions of the X chromosome recruit nearly normal levels of H3K27me3 in the Xist:ΔA cells. However, clusters of active genes are a notable exception to this (Figure 7B; Figure S7A). The same regions are also found not to accumulate H2AK119Ub. Intriguingly, reduced recruitment of H2AK119Ub was observed throughout the chromosome in Xist:ΔA cells compared with WT Xist (Figure 7B). We next quantitated the deposition of both Polycomb marks in relation to regions with actively transcribed genes (Figure 7C). Indeed, both marks show negligible accumulation in the Xist:ΔA mutant when within 100 kb from active genes. On the other hand, regions more distal (a minimum of 200 kb away) to transcribed loci gain normal levels of H3K27me3 and intermediate levels of H2AK119Ub, as exemplified in genome browser tracks (Figures 7F and 7G). Thus, Xist:ΔA leads to PRC2 and PRC1 recruitment to regions without active transcription; however, both complexes seem impeded from spreading into domains of active genes. In line with this, we observed no H3K27me3 or H2AK119Ub enrichment within promoters and bodies of active genes (Figure 7D; Figures S7B–S7E).

28Our finding that H3K27 deacetylation promotes gene silencing during XCI prompted us to investigate H3K27ac loss at gene regulatory elements upon coating by Xist:ΔA RNA (Figure 7E; Figures S7C and S7E). Importantly, we did not observe significant deacetylation of either promoters or enhancers in the Xist:ΔA mutant. This was in stark contrast to robust inactivation of such regulatory elements when using full-length Xist. Thus, promoter and enhancer deacetylation are tightly linked to transcriptional silencing and dependent on the Xist A-repeat. This is in line with our HDAC3 experiments indicating a functional role for histone deacetylation in initiating gene repression.

29To further assess how PcG accumulation relates to histone deacetylation and gene silencing, we performed nChIP-seq in Hdac3−/− cells. We mapped H2AK119Ub and H3K27me3 in two independent Hdac3−/− clones at 0 and 24 hr of DOX treatment. Although HDAC3 loss results in inefficient transcriptional silencing, it should not have major direct effects on PcG marks. Indeed, we found that the pattern of both PcG marks at t = 0 hr was very similar along the X chromosome (Figure 7H). After 24 hr of DOX treatment, both marks accumulate globally along the X chromosome in a similar pattern to that observed in Hdac3+/+ cells (Figure 7H), albeit with slightly lower efficiency. This result differs from the patterns observed in Xist A-repeat mutant cells because, in Hdac3−/− cells, even gene-rich regions accumulate both PcG-associated marks. We then focused on what happens at genic regions. This revealed that genes that were not silenced in Hdac3−/− cells (i.e., sensitive) accumulate neither H2AK119Ub nor H3K27me3 (Figure 7I; Figure S7F). On the other hand, genes efficiently repressed in Hdac3−/− (resistant) robustly accumulated both marks. These relationships are exemplified by a AtrX (resistant gene) and Rnf12 (sensitive gene) (Figure S7G). Thus, Xist coating and intergenic PcG mark deposition are not sufficient for H2AK119Ub spread into active genes. Only the induction of transcriptional silencing enables H2AK119Ub spread to occur.

30In conclusion, we uncoupled the general deposition of the Polycomb marks over the X chromosome from transcriptional silencing. By doing so, we demonstrate that neither PRC1 nor PRC2 dependent marks are able to spread into actively transcribed regions. In the A-repeat mutant, this may be a result of defective Xist spreading or the presence of transcription. On the other hand, our experiments in Hdac3−/− cells revealed that reduced transcription and histone deacetylation are required to enable PcG spreading into actively transcribed regions.

31X chromosome inactivation is a powerful model for the study of transcriptional repression and the formation of facultative heterochromatin. The link between chromatin changes and transcriptional silencing during XCI has remained rather elusive. In this study, we provide the detailed choreography of early transcriptional and chromatin changes during the initiation of XCI at high molecular and temporal resolution (Figure S7H). We also explore the chromatin regulatory pathways that help to achieve efficient gene repression. All in all, our findings point to multiple parallel epigenetic mechanisms being at play during the initiation of XCI to ensure rapid and robust transcriptional silencing.

32Our results indicate that the dynamics and efficiency of chromatin alterations along the X chromosome are, in part, shaped by the spreading of Xist RNA itself. Indeed, rapid histone deacetylation is most efficient in regions proximal to the Xist locus, in line with previous allele-specific RNA-seq experiments.The PcG-dependent marks accumulate first at regions pre-marked by H3K27me3 and H2AK119Ub, which were previously identified as Xist entry sites (XESs), lying in 3D proximity to the Xist locus. This striking PcG mark enrichment and 3D proximity are reminiscent of the Polycomb-enriched regions specifically interacting together in Drosophila melanogaster. Thus, it is an exciting possibility that the chromatin landscape of the X chromosome prior to its inactivation instructs the folding of the chromosome, resulting in a specific pattern of Xist spreading and subsequent dynamics in gene silencing.

33Our high-resolution XCI analysis also sheds light on the kinetics of PcG mark deposition and spreading along the X chromosome. The very rapid accumulation of PRC1-dependent H2AK119Ub that occurs prior to the deposition of H3K27me3 is in contrast with some previous reports suggesting direct PRC2 recruitment by the Xist RNA but is in line with recent reports where a non-canonical PRC1 complex (including PCGF3 and PCGF5) is first recruited, and this then recruits PRC2. Using Xist transgenes on autosomes, this was proposed to occur through HNRNPK bridging the B-repeat region of Xist with PRC1. This finding appears to be contradicted by the defect in H2AK119Ub deposition in the Xist A-repeat mutant. However, we propose that transcriptional silencing may be required to allow for secondary canonical PRC1 recruitment. Indeed, initial H2AK119Ub enrichment is thought to recruit the PRC2 complex via JARID2 or AEBP2, allowing H3K27me3 accumulation, which, in turn, would allow canonical PRC1 binding. Further studies of how transcription affects this feedback and of the timing of PRC1 and PRC2 recruitment are necessary to further validate the revised model.

34Here we find that, during the first hours of XCI, PcG marks initially accumulate at intergenic regions and later spread into genes. In cells expressing the Xist A-repeat mutant, neither PRC1 nor PRC2-dependent marks are able to spread into genic regions because they are probably impeded by ongoing transcription. Alternatively, the Xist A-repeat deletion might lead to local defects in Xist spreading, as suggested by. Furthermore, our detailed analysis of Hdac3 mutant cells revealed that H2AK119Ub spreading to genes requires their—at least partial—deacetylation and transcriptional inactivation. This indicates that PRC1 mark spreading is not the trigger for initiating gene repression. On the other hand, because both processes seem to occur around the same time, there might be a potential role of PRC1 in facilitating efficient XCI but not initiating it. Indeed, in an autosomal context, a mutant of Xist, which is unable to recruit PRC1, has slightly reduced silencing capacity.

35Here we identify H3K27 deacetylation as one of the earliest chromatin alterations during XCI, tightly linked to transcriptional silencing. This is in line with previous IF/FISH studies in differentiating ESCs and embryos. Moreover, by using Hdac3 loss-of-function experiments, we demonstrate that histone deacetylation is not simply due to reduced transcription. Rather, it appears to be an active process that promotes gene silencing. We propose a model whereby histone deacetylation by the pre-bound HDAC3 promotes efficient silencing of most X-linked genes (Figure S7I). A previous study has shown that Hdac3 knockdown results in defective silencing of a single locus during XCI. However, our study reveals that other Xist A-repeat-dependent but HDAC3-independent mechanisms are also at play during XCI. This is presumably to safeguard and mediate timely gene repression, especially at genes that need to become rapidly silenced; e.g., Rnf12. Although we have excluded the involvement of other histone deacetylases during XCI initiation, future studies will reveal whether these HDAC3-independent mechanisms act via other chromatin-modifying or -remodeling activities or whether they occur through entirely different pathways.

36Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Edith Heard (edith.heard@curie.fr)

37TX1072 (mouse, female, [Mus musculus castaneus X C57BL/6] embryonic stem cells) cells have been previously derived in the lab. Hdac3−/− and Hdac3Flag/Flag clones have been derived from the TX1072 line by CRISPR/Cas9 targeting. Cells were cultured on gelatine-coated plates in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 15% fetal bovine serum, 2-mercaptoethanol (0.1mM), LIF(1000u/mL) and 2i (PD0325901 [0.4 mM], CHIR99021 [3 mM]). TXY:Xist and TXY:XistΔA (mouse, male embryonic stem cells) were obtained from the Wutz team. These lines were cultured on gelatine-coated plates in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 15% fetal bovine serum, 2-mercaptoethanol (50mM) and LIF (1000u/mL). All cells were incubated at 37°C with 8% CO2.

38TX1072 based ESC lines were plated at a density of 1.8mln/T75 flask. After 24hrs of culture in DMEM/15%FCS+LIF+2i cell the medium was supplemented with doxycycline (1ug/ml). Cells were then collected at regular intervals. For TXY based ESC lines we have plated them at a density of 1mln/T75. After 24hrs of culture in DMEM/15%FCS+LIF cells were washed twice with PBS and medium was changed to DMEM/10%FCS-LIF with or without doxycycline (1.5ug/ml). Cells were collected after 48hrs of differentiation. For HDAC inhibition cells were pre-treated for 1hr with: 10 μM BRD668; 10 μM RGFP966; 10 μM Tasquinimod; 5 μM Tubastatin A; 5 μM PCI-34051; 30nM LMK235.

39Cells were collected using Accutase (Thermo Fisher Scientific), washed twice in ice-cold PBS and counted. Typically, 3.5mln cells were used per immunoprecipitation (IP). A fraction of cells was always used for RNA/FISH verification of Xist induction. Cell pellet was resuspended in 90 μL (per 10 mln cells) of Lysis Buffer (50 mM Tris-HCl, pH7.5; 150mM NaCl; 0.1% sodium deoxycholate; 1% Triton X-100; 5mM CaCl2; Protease Inhibitor Cocktail; 5 mM sodium butyrate). After lysing cells on ice for 10 min we added 62 μL (per 10 mln cells) of Lysis Buffer with MNase (500 μL buffer + 0.5 μL MNase). Chromatin was digested for exactly 10 min at 37°C and reaction was stopped by the addition of 20 mM EGTA. To remove undigested debris the lysates were centrifuged at 13000 rpm for 5 min at 4°C. Supernatant was transferred to a fresh tube, an equal volume of STOP Buffer (50 mM Tris-HCl, pH7.5; 150mM NaCl; 0.1% sodium deoxycholate; 1% Triton X-100; 30 mM EGTA; 30 mM EDTA; Protease Inhibitor Cocktail; 5 mM sodium butyrate) was added, samples were stored on ice.

405 μL of lysate was digested in 45 μL of ProtK Digestion Buffer (20 mM HEPES; 1 mM EDTA; 0.5% SDS) for 30 min at 56°C. 50 μL of AMPure XP beads were added to the digested lysate together with 60 μL of 20% PEG8000 1.25M NaCl. After mixing the samples were incubated for 15 min at RT. Beads were separated on a magnet and washed twice with 80% Ethanol for 30sec. DNA was eluted in 12 μL of Low-EDTA TE and DNA concentration was measured using Qubit DNA High-Sensitivity kit. These measurements were used to normalize lysate concentration between samples. DNA isolated in this step was used as the input sample. The volume of each undigested lysate was adjusted for equal concentration and to obtain 1mL per IP using a 1:1 mix of Lysis Buffer and STOP Buffer.

41Anti-mouse Dynabeads (50ul/IP) and Protein-A Dynabeads (10ul/IP) were washed twice in Blocking Buffer (0.5% BSA; 0.5% Tween in PBS). Beads were then resuspended in Blocking buffer and coated with antibodies for 4hrs at 4°C (anti-mouse Dynabeads: H3K9ac[1ug/IP], H3K4me3[2.5ug/IP], H3K27ac[1ug/IP]; Protein-A dynabeads: H3K4me1[0.4ug/IP], H3K27me3[1ug/IP], H2AK119Ub[0.4ug/IP], H4ac[2ug/IP]). Once coated beads were magnet-separated and resuspended in 1mL of concentration-adjusted lysate. Samples were left rotating overnight at 4°C.

42Following day beads were magnet-separated and washed quickly with ice-cold washing buffers. anti-H3K4me3 IP was washed 8 times with Low Salt Buffer (0.1% SDS; 1% Triton X-100; 2 mM EDTA; 20 mM Tris-HCl, pH 8.1; 150 mM NaCl; 0.1% sodium deoxycholate). All remaining IPs were washed 4-times with Low Salt Buffer, 2-times with High Salt Buffer (0.1% SDS; 1% Triton X-100; 2 mM EDTA; 20 mM Tris-HCl, pH 8.1; 360 mM NaCl; 0.1% sodium deoxycholate) and 2-times with LiCl buffer (0.25 M LiCl; 1% NP40;1.1% sodium deoxycholate; 1 mM EDTA; 10 mM Tris-HCl pH 8.1). Prior to elution all samples were rinsed once in TE. ChIP-DNA was eluted in ProtK-Digestion buffer for 15 min at 56°C. Beads were separated and the supernatant was further digested for another 2 hr at 56°C. DNA was isolated using AMPure XP beads as described for the input sample.

43For each nChIP-seq, 0.5 μL of each sample was used for qPCR validation of enrichment at control regions. 0.5 μL of input samples were also used to verify the digestion efficiency using D1000 tapestation. Remaining DNA concentration was adjusted and used for library preparation using Ovation® Ultralow Library System V2 following suppliers protocol. Amplified libraries were size-selected for dinucleotide fraction (350-600bp fragments) using agarose gel-separation and MinElute Gel Extraction Kit (QIAGEN). Sample quality was inspected using D1000 tapestation. Samples were sequenced with HiSeq2500 using PE100 mode for nChIP-seq on TX1072 cell line and SE50 mode for nChIP-seq on TXY cell line.

442.5x105 of ESC were plated onto fibronectin-coated 6-well in N2B27 medium (50%DMEM/F12 and 50% Neurobasal Medium supplemented with: 2-Mercaptoethanol[0.1mM], L-Glutamine[20mM]; B27 supplement[1x], Ndiff Neuro-2 Medium Supplement[1x]) with bFGF (12 ng/ml) and ActivinA (20 ng/ml) and with or without doxycycline (1ug/mL). Medium was changed every day and cells were collected at appropriate intervals.

45ESC were dissociated using Accutase (Invitrogen) and adsorbed onto Poly-L-Lysine (Sigma) coated coverslips #1.5 (1mm) for 5 min. Cells were fixed with 3% paraformaldehyde in PBS for 10 min at room temperature and permeabilised for 5 min on ice in PBS containing 0.5%Triton X-100 and 2mM Vanadyl- ribonucleoside complex (New England Biolabs). Coverslips were preserved in 70% EtOH at −20°C. Prior to FISH, samples were dehydrated through an ethanol series (80%, 95%,100% twice) and air-dried quickly. For detecting Huwe1 and Atrx, a BAC spanning the respective genomic region (RP24-157H12, RP23-160I15) was labeled by nick translation (Roche) using spectrum green (Abbot). Per coverslip, 60ng probe was ethanol precipitated with Cot1 repeats, resuspended in formamide, denatured (10min 75°C) and competed for 45min at 37°C. For Xist detection or intron-spanning plasmid probe p510 was labeled by nick translation (Roche) using spectrum red (Abbot). Probe was ethanol precipitated and resuspended in formamide and denatured (10min 75oC). Both Xist and AtrX or Huwe1 probes were co- hybridized in FISH hybrization buffer (50% Formamide, 20% Dextran sulfate, 2x SSC, 1 μg/μl BSA, 10mM Vanadyl-ribonucleoside) over-night. Washes were carried out at 42°C three times 5min in 50% formamide in 2X SSC at pH = 7.2 and three times 5min in 2X SSC. 0.2mg/ml DAPI was used for counterstaining and mounting medium consisted of Vectashield (Vectorlabs). Images were acquired using an Inverted Confocal Spinning Disk Roper/Nikon.

46Cells were grown on fibronectin coated coverslips #1.5 (1mm) and treated with doxycycline (1ug/mL) for 24hrs. Cells were fixed with 3% paraformaldehyde in PBS for 10 min at room temperature and permeabilised for 5 min on ice in PBS containing 0.5%Triton X-100 and 2mM Vanadyl- ribonucleoside complex (New England Biolabs). Samples were blocked for 15 min at room temperature with 1%BSA/PBS. Coverslips were incubated with primary antibodies diluted in blocking solution (anti-H4ac: 1/500; anti-H3K27ac:1/100; anti-H2AK119Ub: 1/200) in the presence of a Ribonuclease Inhibitor (0.8u/mL; Euromedex) for 45 min at room temperature. Coverslips were washed thrice with PBS for 5min and subsequently incubated with secondary antibody solution (goat anti-mouse or rabbit conjugated with Alexa fluorophores; 1/500; supplemented with Ribonuclease Inhibitor [0.8u/mL; Euromedex]) for 45 min at room temperature. Coverslips were washed three times with PBS for 5 min at room temperature. Cells were fixed again with 2% paraformaldehyde in PBS for 10 min at room temperature. After rinsing twice in 2xSSC coverslips were hybridized for Xist detection. Here we used a custom designed strand-specific probe that covers all exons with ∼75 bp long oligo nucleotides end-labeled with the Alexa 488 fluorophore (Roche). Probes was hybridized in FISH hybridization buffer (50% Formamide, 20% Dextran sulfate, 2x SSC, 1 μg/μl BSA, 10mM Vanadyl-ribonucleoside) over-night. Washes were carried out at 42°C three times 5min in 50% formamide in 2X SSC at pH = 7.2 and three times 5min in 2X SSC. 0.2mg/ml DAPI was used for counterstaining and mounting medium consisted of Vectashield (Vectorlabs). Images were acquired using an Inverted Confocal Spinning Disk Roper/Nikon. For signal quantification, signal intensity was measured along an axis across the Xist domain (z-plane with Xist maximum intensity) and part of the nucleus. Signal was not measured across the nucleoli. Profiles from 70-100 nuclei were centered at the point of Xist cloud boundary (maximum inflection point of Xist signal profile). Signal was normalized at per cell basis. In Figure 5B and Figure S5B shown are average profiles from all nucleoli with shading representing 25 and 75 quantiles.

47RNA was extracted according to the manufacturer’s recommendations using RNeasy Mini Kit (QIAGEN) with on-column DNase digestion (QIAGEN). For cDNA synthesis 1.1 μg RNA was reverse transcribed using Superscript III Reverse Transcriptase (Thermo Fisher Scientific). For allelic-skewing analysis the cDNA was PCR-amplified with biotinylated primers and sequenced using the Pyromark Q24 system (QIAGEN). To quantify ChIP DNA enrichments at control regions 2x SybRGreen Master Mix (Applied Biosystems) and a ViiA7 system (Applied biosystems) were used.

48RNA was prepared as described above and quality of samples were verified by Tapestation. Only samples with RIN score above 9 were processed. 100 ng of RNA was used for library preparation using Ovation Mouse RNA-Seq System that uses mouse-specific InDA-C depletion of rRNA. Manufacturers recommendations were followed. Libraries were sequenced using HiSeq2500 at PE100 settings.

49The TT-seq protocol was performed as previously described with modifications. Cells were induced with DOX as described above in two T75 flasks per sample. After 0, 4, 8, 12 and 24 hr, RNA was labeled by the addition of 0.5 mM 4-thiouridine (4sU, Carbosynth) for 5 min. Medium was quickly removed and cells were lysed with 10 mL of TRI reagent (Sigma-Aldrich). Lysates were stored at −80°C.

50The spike-in mix, containing three 4sU-labeled and three unlabelled RNAs, was synthesized in vitro with MEGAscript T7 Transcription Kit (Thermo Fisher), where 1/10 UTP in the transcription reactions was replaced with 4sUTP (Jena BioScience) for the 4sU-labeled spike-ins. RNA lysates were thawed on ice and 24 ng of spike-in mix was added per 10 million cells. The RNA was then purified according to the manufacturer’s protocol. 4sU-labeled RNA was isolated from 300 μg total RNA, for this two aliquots of 150 μg were sheared in Covaris MicroTubes with the Covaris S220 System (Covaris) for 10 s with settings 100 W and 1% duty cycle. These were then pooled and 2 μg of sonicated total RNA was taken for total RNA fraction. The RNA was biotinylated with EZ-Link HPDP-Biotin (Thermo Fisher) in two biotinylation reactions per sample (150 μg RNA in 10 mM Tris pH 7.5; 1 mM EDTA, 40% DMSO; and 200 μg/ml HPDP-biotin) that were incubated 1.5 h at 24°C. The biotinylated RNA was extracted with chloroform and precipitated with isopropanol. 4sU-labeled RNA was isolated by incubating the biotinylated RNA with paramagnetic streptavidin μMACS MicroBeads (Miltenyi) for 15 min at room temperature and binding to MACS columns. The columns were washed three times with wash buffer (100 mM Tris pH 7.5; 10 mM EDTA; 100 mM NaCl; and 0.1% Tween 20) at 65°C and three times with wash buffer at room temperature. The RNA was eluted from the columns with 100 mM DTT. The eluted 4sU-labeled RNA and reserved total RNA were purified with the miRNeasy Micro kit (QIAGEN) according to the manufacturer’s protocol with on-column DNase I treatment. 150 ng and 500 ng of isolated newly synthesized RNA and total RNA fractions, respectively, were used to prepare NGS libraries with the TruSeq Stranded Total RNA Library Prep Kit (Illumina). The sample libraries were pooled and sequenced as a single run on the Nextseq550 (Illumina) using PE150 mode.

51Two gRNAs have been designed as previously described and cloned into pX462 vector allowing for their expression under U6 promoter. Additionally, this vector leads to expression of Cas9. TX1072 ESCs were transfected using AmaxaTM 4D-NucleofectorTM system (Lonza) as per manufacturer’s protocol. After a pulse of puromycin selection cells were plated at clonal density. Single clones were picked, expanded and screened for deletion by PCR. Final two knockout clones were further validated by sequencing both alleles. Hdac3−/−#1 harbors a deletion of exon 4-7 on the Cast allele and exon 9 on the B6 allele. Hdac3−/−#2 harbors a deletion of exon 4-7 on the B6 allele and exon 7 on the B6 allele. Both mutants showed negligible expression on the mRNA level as well as no detectable protein when using a N terminus specific antibody (Figure 4B).

52The Hdac3-Aid-3xFlag targeting construct was generated as follows: 500bp homology arms (flanking both sides of, but excluding the stop codon of Hdac3) were PCR amplified from mouse genomic DNA. 1-step Gibson cloning (NEB) was subsequently used to simultaneously surround the digested AID-3xFLAG insert (carrying a puromycin resistance gene under the control of the PGK promoter) in frame with the homology arms and clone the insert into the pBR322 vector. Synonymous mutations in the PAM/SEED target sequence (located on the 5′ homology arm) were then introduced using the QuickChange II XL site-directed mutagenesis kit (Agilent) to prevent Cas9 mediated cleavage of the targeting vector upon transfection and of the 3xFLAG tagged allele(s) upon integration.

53Single gRNAs has been designed as previously described and cloned into pX462 vector allowing for its expression under U6 promoter.

54Hdac3 targeting was performed using the 4-D nucleofector system from Lonza. 5 million cells were electroporated with 2.5 μg each of non-linearized targeting vector and gRNA/Cas9 encoding plasmids. The day after, puromycin was added at a concentration of 0.4ug/mL. Single clones were picked, expanded and screened for insertion by PCR. One clone showing homozygous insertion of the FLAG tagged cassette at the Hdac3 locus was validated by sanger sequencing and western blotting and kept for subsequent experiments (Figure S5).

5550 mln (per IP) ESCs (TX1072 –DOX; Hdac3Flag/Flag +DOX 24h) were trypsinised and washed in PBS. Pellets were snap frozen and stored at −80°C. Defrosted pellets were resuspended in 10 mL of buffer A (10 mM HEPES pH 7.9; 5 mM MgCl2; 10 mM KCl; 1 mM DTT; 0.1% NP-40, protease inhibitors) and incubated for 10 min at 4°C. Nuclei were pelleted at 2000 rpm at 4°C for 10 minutes. Nuclei were resuspended in 500μL of buffer C (20mM HEPES pH 7.9; 20% (v/v) glycerol; 150 mM KCl; 5 mM MgCl2; 2 mM EDTA; 2 mM EDTA; 1 mM DTT; protease inhibitors) and sonicated 3 times for 10 s (Bioruptor, medium setting). Lysates were cleared by centrifugation 12000 rpm at 4°C for 20 minutes. Lysate concentrations were equalised based on protein concentration. Anti-FLAG M2 Magnetic Beads (Sigma-Aldrich) were washed twice in buffer C and then added to each lysate (50μL per IP). Lysates were left rotating over-night at 4°C. Next day beads were washed 5 times in wash buffer (20 mM HEPES pH 7.9; 0.1% NP-40; 150mM KCl; 5mM MgCl2; 2mM EDTA; 2 mM EDTA; 1 mM DTT).

56Proteins on magnetic beads were washed twice with 100 μL of 25mM NH4HCO3 and on-beads digestion was performed with 0.2 μg of trypsin/LysC (Promega) for 1 hour in 100μL of 25mM NH4HCO3. Samples were then loaded onto a custom-made C18 StageTips for desalting. Peptides were eluted using 40/60 MeCN/H2O + 0.1% formic acid and vacuum concentrated to dryness.

57Online chromatography was performed with an RSLCnano system (Ultimate 3000, Thermo Scientific) coupled online to an Orbitrap Fusion Tribrid mass spectrometer (Thermo Scientific). Peptides were trapped on a C18 column (75 μm inner diameter × 2 cm; nanoViper Acclaim PepMap 100, Thermo Scientific) with buffer A (2/98 MeCN/H2O in 0.1% formic acid) at a flow rate of 4.0μL/min over 4 min. Separation was performed on a 50cm x 75 μm C18 column (nanoViper Acclaim PepMap RSLC, 2 μm, 100Å, Thermo Scientific) regulated to a temperature of 55°C with a linear gradient of 5% to 25% buffer B (100% MeCN in 0.1% formic acid) at a flow rate of 300 nL/min over 100 min. Full-scan MS was acquired in the Orbitrap analyzer with a resolution set to 120,000 and ions from each full scan were HCD fragmented and analyzed in the linear ion trap.

58Cells were grown in 145mm plate format. Cells were initially fixed in 1.5 mM ethylene glycolbis [succinimidyl succinate] (EGS) for 30 min, next 2% formaldehyde was added and further incubated for 10 min. After rinsing twice in PBS, cells were scraped and pellets directly used for ChIP.

59Fixed pellets were lysed in 6 mL Nuclear Lysis Buffer (0.5% Triton X-100, 0.1 M sucrose, 5 mM MgCl2, 1 mM EDTA, 10 mM Tris-HCl pH 8.0, 1x Protease Inhibitors) for 10min on ice and dounce-homogenized. Nuclei were pelleted and resuspended in 0.9 mL Lysis Buffer (1 mM EDTA, 0.5 mM EGTA, 10 mM Tris pH 8.0, 0.5% N-Lauroylsarcosine, 1x Protease Inhibitors). Chromatin was sheared using a BioruptorPlus (Diagenode) set to high for 30 cycles. Unsonicated chromatin was removed by centrifugation. Supernatant was diluted with 3.6 mL of Dilution Buffer (1.25% Triton, 0.125% sodium deoxycholate, 6 mM EDTA, 10 mM Tris-HCL pH8.0, 1x Protease Inhibitors) and antibody coated beads were added (per IP: 10 μL of anti-FLAG antibody). Samples were left rotating overnight at 4°C.

60Following day beads were magnet-separated and washed 5-times with LiCl buffer (0.50 M LiCl; 1% NP40; 1.1% sodium deoxycholate; 1 mM EDTA; 10 mM Tris-HCl pH 8.1) and 5 times with High Salt Buffer (0.1% SDS; 1% Triton X-100; 2 mM EDTA; 20 mM Tris-HCl, pH 8.1; 1M NaCl; 0.1% sodium deoxycholate). Each wash was performed for 10 min on a rotating wheel at 4°C. Prior to elution all samples were rinsed once in TE. ChIP-DNA was eluted in ProtK-Digestion Buffer (20mM HEPES; 1mM EDTA; 0.5% SDS; 0.8mg/mL Proteinase K) for 15min at 56°C. Beads were separated and the supernatant was further digested for another 2hrs at 56°C and then decrosslinked for 4hrs at 68°C. DNA was isolated using AMPure XP beads. Libraries were prepared using Ovation Ultralow Library System V2 following suppliers protocol. Samples were sequenced with HiSeq2500 using PE100 mode.

61Adapters and low quality bases (< Q20) have been removed with TrimGalore (v0.4.0; http://www.bioinformatics.babraham.ac.uk/projects/trim_galore) and Cutadapt (1.8.2). An ‘N-masked” genome has been generated with SNPSplit (0.3.2) which is a version of the mouse reference genome mm10 where all the polymorphic sites for the hybrid strain Mus musculus CAST/EiJ and Mus musculus C57BL/6 are masked by ambiguity nucleobase ‘N’. For all samples, reads were then mapped to the ‘N-masked” genome with Bowtie2 (2.2.5) with options [--end-to-end -N1 -q]. Duplicates were discarded with Picard MarkDuplicates (1.65) with options [REMOVE_DUPLICATES = true] (https://broadinstitute.github.io/picard/) and reads mapped on blacklisted regions from Encode Consortium were discarded. SNPSplit (0.3.2) was then used to generate allele-specific BAM files by separating the alignment into two distinct alleles (CAST and B6) based on SNPs information downloaded from Sanger. Bigwig files were created with bedtools genomeCoverageBed (2.25.0), using a scale factor calculated on the total library (10.000.000/total reads) for both allele specific bigwigs, and loaded on UCSC genome browser.

62Peak calling was done with MACS2 (2.0.10) with options [--broad -B -f BAMPE --broad-cutoff 0.01], on total (non allele-specific) ChIP-seq signal with input as control. Then, peaks with a fold change inferior to 3 were filtered out. For each histone mark, consensus peaks were defined as follow. For each replicate, all peaks coordinates were merged using bedtools merge (2.25.0). Then, common regions between merged peaks coordinates of each replicate were selected using bedtools intersectBed (2.25.0).

63Total and allelic reads overlapping consensus peaks were counted using featureCounts (1.5.1), with options [ -C -p -P]. Only peaks with more than 50 total allelic reads were selected. Next, ratios of allelic counts, called d-scores [readsB6/(readsCast+ readsB6)], were calculated.

64The dynamic evolution of d-score with time was analyzed for each peak with a d-score calculated on the 10 samples (5 times, 2 replicates). Moreover, only peaks that had a biallelic d-score at time 0 (comprised between 0.3 and 0.7) were selected. Then normalized d-scores were calculated by dividing each d-score by the corresponding initial d-score (time 0) and by 2, so that initial normalized d-scores start at 0.5. Initially we have fitted either exponential or sigmoidal curve to these values. We compared the quality of this fitting by taking into account the number of parameters used in the models (AIC and BIC criteria) and concluded that the sigmoidal model best reflects the observed dynamics (data not shown). Sigmoidal fitting of normalized d-scores in function of time has been done for each peak with a four-parameter log-logistic function from drc R package. Sigmoidal fittings with low residuals (< 0.2; corresponding to more than 98% of the peaks) were selected, and three parameters were obtained from these fittings: IC35, which is the time where the normalized d-score reaches 0.35 on the sigmoidal fitting; ED50, which corresponds to the maximum slope of the sigmoidal fitting; and the minimum second derivative used as a proxy for when the curve starts to decrease. The minimum of the second derivative has been calculated using optimize function from R in the interval between 0 and the calculated ED50 to avoid finding a local minimum not corresponding to the real minimum. IC/ED/minimum second derivative superior to 24h (outside the time range of the experiment) or not calculated were considered windows with a late loss of active histone mark, and replaced by 24h.

65Moreover, for all active histone marks, IC25, IC30 and IC40 (thresholds of normalized d-score:0.25; 0.3; 0.4) were also tested and were able to replicate all our major findings (data not shown).

66Early, intermediate and late silenced peaks located on TSS were defined with a 3 clusters k-means based on IC35 values. Early and late silenced peaks were compared for several features using Wilcoxon test: level of expression of associated genes (from TT-seq - see below), distance to Xist, distance to the closest TAD boundary, distance to the closest LAD, LINE density in a window of 100kb around the peak (from RepeatMasker database) and gene density in a window of 100kb around the peak. Moreover, IC35 was compared for peaks that are inside or close to Xist entry sites(< 100kb) and those that are distant.

67For comparison of dynamics (Figure 6E) of deacetylation and accumulation of repressive marks, early, mid-early, mid-late and late silenced genes were defined with a 4 clusters k-means based on IC35 values.

68Average plots of H3K4me1 signal were created using DeepTools (3.0.2). Matrix counts were created using DeepTools computeMatrix around active TSS (see above) on chrX and autosomes separately (with option [--binSize 50]), plots were then created using DeepTools PlotProfile.

69For repressive marks, global analysis was first done on fixed windows (10 kb) spanning the whole genome, then on different genomic subcategories: active gene bodies, active promoters, active enhancers and intergenic regions. Active genes were defined as genes with a transcript having its TSS (refFlat annotation) overlapping a consensus peak of H3K9ac and a consensus peak of H3K4me3. For genes having several active transcripts detected, the active gene was defined as starting at the minimum start of transcripts, and ending at the maximum end of transcripts. Then, the active gene bodies were defined as those active genes excluding the 2 first kb downstream of TSS. Active promoters were defined as ± 2kb windows around the TSS of active genes. Putative, active enhancers were defined as intersect between consensus H3K27ac peaks and consensus H3K4me1 peaks, located at a minimal distance of 1kb from a TSS. Intergenic regions were defined as 10 kb windows not overlapping a gene (active or inactive) and its promoter (2kb downstream) or an active enhancer. Enhancer coordinates were also used for the analysis of active histone mark (H3K4me1 and H3K27ac).

70It should be noted that a d-score analysis, as used for active mark loss, involves the use of the active allele as an internal control, but in the case of H2AK119Ub and H3K27me3, this was hindered by very low enrichment for repressive marks across the Xa. For all defined windows, total and allelic reads overlapping those features were then counted using featureCounts (1.5.1), with options [-C -p -P]. Then, analysis was done based on normalized reads from B6 allele (allele of the inactive X chromosome). For each sample, a normalization factor was calculated with the trimmed mean of M-values method (TMM) from edgeR package, based on B6 reads overlapping consensus peaks located on autosomes (identified and defined as consensus such as for active histone marks analysis). To correct for chromatin accessibility or mappability bias, 10kb windows with outliers counts in the input (counts superior or inferior to mean +-1.5 sd) were discarded from the analysis. Moreover, to represent B6 read accumulation compared to time 0, subtraction of normalized initial counts (time 0) was then applied to all other time points.

71Sigmoidal fitting of B6 read accumulation in function of time has also been done with the four-parameter log-logistic function from drc R package. Sigmoidal fittings with low residuals (< mean(residuals) + 1.5 sd(residuals)) were selected, representing more than 90% of the analyzed windows. The ED50 and the maximum second derivative of the sigmoidal fitting were calculated for each window. Windows with ED50/maximum 2nd derivative superior to 24h, or not calculated, were considered as windows with a late accumulation of repressive marks, those were then replaced by 24h. For pairwise comparison between marks, only windows with ED50/maximum 2nd derivative inferior to 24h for both marks were selected. For comparison of ED50 between different genomic subcategories (active promoters, gene bodies, enhancers, intergenic regions) for a same histone mark, no filter was done.

72To test the stability of our results, main analysis was also performed with different thresholds of residuals (0.2; 0.15; 0.1; 0.05). Moreover, repressive marks dynamics were also analyzed using allelic ratio (d-scores), such as for analysis of active histone marks, using ED50, IC60 and IC65 as criteria. All those analysis revealed stable and reproducible results (data not shown).

73Adapters (“TruSeq Universal Adapter”) and low quality bases (< Q20) have been removed with Cutadapt (v1.16). For all samples, reads were then mapped to the ‘N-masked’ genome (masked for SNPs of the hybrid strain Mus musculus CAST/EiJ and Mus musculus C57BL/6) with STAR 2.3.0 with maximum 2 percent mismatches. Only unique alignments were retained and soft clipping was disabled [mapping parameters: ‘--outFilterMismatchNoverReadLmax 0.02 --outFilterMultimapScoreRange 0 --alignEndsType EndToEnd’]. Samtools was used to quality filter SAM files, where alignments with MAPQ smaller than 7 (-q 7) were skipped and only proper pairs (-f 2) were selected. SNPSplit (v0.3.2) was used to generate allele-specific SAM files. Then, total and allelic reads per gene were counted with featureCounts (1.5.1) [options: -C -p -s 2 -T 8] on longest gene coordinates (refFlat annotation). The next steps of analysis (from d-scores to sigmoidal fitting) were done such as for nChIP-seq analysis, except that d-scores were calculated for genes with minimum 10 allelic reads instead of 50. Total normalized expression used for features comparison of early and late silenced peaks (see upper) was calculated based on signal in introns and exons, normalized by the size of the library and the size of the entire gene.

74Reads were first mapped on rRNA with Tophat (2.1.0), with options [-g 1 --no-coverage-search --library-type fr-secondstrand]. Paired unmapped reads were then used to reconstruct fastq files with bedTools bamToFastq (2.25.0). Those files were then mapped with Tophat (2.1.0), with options [-p 8 -g 1 -x 1 -N 3 --read-edit-dist 3 --no-coverage-search --library-type fr-secondstrand], with refFlat annotation. Reads covering exons of each gene were then counted with featureCounts (1.5.1) with options [-C -p -s 1 -T 8]. D-scores were then calculated for genes with minimum 10 allelic reads.

75For differential gene expression analysis prior to DOX treatment, analysis was done on genes for which 2 among the 4 samples have a TPM superior to 1. Normalization was done using TMM method from edgeR and differential analysis using Voom function from Limma R package.

76The processing from raw data (fastq) to clean mapped file was done such as for time course nChIP-seq. Then, for active histone marks, the reads counts of H3K27ac and H4ac nChIP-seq were done on consensus H3K27ac peaks and consensus H4ac peaks defined in time course analysis, using featureCounts (1.5.1), with options [-C -p -P]. D-score was calculated with a slight modification for one of the 2 mutant sample. Indeed, this sample showed a non biallelic signal on X chromosome in its input, due to the presence of some XO cells. To take this bias in account, a normalization factor was calculated as follow: normFactor = (readsB6/(readsB6 + readsCAST)) / (1-(readsB6/(readsB6 + readsCAST))).

77Then d-score for this mutant was adjusted from classic d-score as follow: d-score = d-score/(d-score + normFactor - normFactor ∗ d-score). After this adjustment, only the peaks that were biallelic at time 0 in WT and mutant samples were selected (d-scores between 0.3 and 0.7) for analysis. For repressive marks, analysis was done such as for time course nChIP-seq, based on signal accumulation of 10 kb windows all along the X chromosome.

78For differential peak signal analysis at t = 0hr, normalization was done using TMM method from edgeR and differential analysis using Voom function from Limma R package.

79Sensitive and resistant peaks/genes were defined as follow. For nChIP-seq on active histone marks and RNA-seq, only peaks/genes with d-scores considered as stable between replicates (diff replicate < median + 2sd of all differences between replicates) for both clones (WT, mutant) at each time point were selected. D-scores were normalized to time 0 (d-score / d-score t0), then only peaks/genes considered as not escaping were selected (normalized d-score at t24 < 0.8). Difference of normalized d-score at 24h between WT and mutant was then calculated, and divided by the difference of normalized WT d-score between 0 and 24h for the WT sample. K-means with 3 clusters was done on this normalized difference, identifying 3 groups: sensitive, intermediate and resistant peaks/genes. For nChIP-seq on repressive histone marks, differential enrichment between WT and mutant was defined as the difference between WT normalized BL6 enrichment relative to time 0 and Mutant normalized BL6 enrichment relative to time 0, divided by WT normalized BL6 enrichment relative to time 0.

80The processing from raw data (fastq) to d-score calculation was done such as for time course nChIP-seq, with slight modifications. The peak calling was done with MACS2 (2.0.10) with options [-B -f BAMPE], on total (non allele-specific) ChIP-seq signal against the FLAG ChIP-seq on TX1072 cell line as control. Consensus peaks were defined as peak regions present at least in 2 of the 4 samples, using bedtools multiIntersectBed (2.25.0). Only peaks with more than 20 total allelic reads were selected.

81Heatmaps and average plots of HDAC3 signal in comparison with H3K27ac and H4ac signal were created using DeepTools (3.0.2). Merged peaks of H3K27ac and H4ac at time 0 were split in 2 categories: active promoters for merged peaks overlapping a TSS and putative active enhancer for merged peaks at a distance of minimum 2 kb of a TSS. The merged peaks were ordered by normalized signal of HDAC3 peaks (signal/length of the peak) overlapping those peaks. For HDAC3, signal of the control (FLAG ChIP-seq on TX1072 cell line) was subtracted from the signal of FLAG ChIP-seq on Hdac3Flag/Flag ESC cell line using DeepTools bamCompare with options [--ratio subtract --binSize 100] before to proceed to the next step. Then, matrix counts were created using DeepTools computeMatrix around TSS of active promoters (with option [--binSize 100]) or centers of putative enhancers, plots were then created using DeepTools PlotHeatmap.

82For H3K27ac and HDAC3 peaks annotation comparison (Figure S5C), peaks were annotated as follow: peaks at a distance < 2000 bp of a TSS were considered in promoters, other peaks were considered as intragenic if overlapping a gene body, intergenic if not.

83Category of peaks accumulating signal or not on Xi (Figures S5E and S5F) were created based on clustering with hclust function with parameter [method = ’Ward.D’], using Pearson correlation as distance. 11 clusters were identified using cutree function. Those clusters were grouped in 2 super groups according to the presence or not of an accumulation of signal on Xi at 4hr in each cluster.

84Processing steps to clean bam files were similar to time course nChIP-seq analysis, except that mapping was not allele-specific but done on mm10 genome and that reads mapped with low quality (< q10) were removed with samtools (1.3). Such as for time course nChIP-seq analysis, ChIP-seq signal was analyzed differently depending on the histone mark: per peak for active mark (H3K27ac) and per window for repressive marks (H3K27me3, H2AK119Ub). Peak calling was done similarly to time course nChIP-seq analysis of active histone marks. Here, consensus peaks were defined as common regions between peaks identified in minimum 2 among the 4 samples using bedtools multiIntersectBed (2.25.0) and bedtools merge (2.25.0).

85The windows for repressive mark analysis were defined similarly as for nChIP-seq time course analysis for repressive marks, except that active TSS were defined here based on consensus peaks of H3K27ac overlapping TSS, and putative active enhancers as H3K27ac peaks located at minimum 1kb of TSS.

86Reads overlapping defined peaks (for active marks) or windows (for repressive marks) were then counted with featureCounts (1.5.1) with default options.

87For global analysis on repressive marks, counts normalization was done such as for time course nChIP-seq analysis, based on counts falling in autosomal consensus peaks. For windows and peaks analysis, windows that had less than one read per 50kb for more than 2 among the 8 samples were removed for the analysis. Normalization factors were calculated based on windows located on autosomes (with selection of the 10000 most higher signal for intergenic window analysis), with TMM method using edgeR.

88Because of the high variability in proportion of cells involved in X chromosome inactivation quantified by the presence of Xist Cloud by FISH experiments (TXY:XistWT#1 46.64%, TXY:XistWT#2 59.44%, TXY:XistΔA#1 50.61%, TXY:XistΔA#2 48.24%), linear regression including the percentage of induction calculated by FISH was fitted for each window according the following model: ∼0 + clones + clones: induction, using Voom function from Limma R package. The slope of this regression represents then the logFC between noDox and Dox conditions if the induction of the cell population was complete (corrected logFC).

89Metaplots were created using DeepTools (3.0.2). Bigwigs of log2(FC) between Dox and noDox samples were created with personalized scaling according to normalized factors calculated above using DeepTools bamCompare, bigwig of mean of log2(FC) between replicates was then created using DeepTools bigwigCompare (binSize: 100 bp for H3K27ac, 1000 bp for H3K27me3 and H2Ub), matrix counts was then created using DeepTools computeMatrix on active genes coordinates (see above) on chrX and autosomes separately and plots were then created using DeepTools PlotProfile.

90For identification, the data were searched against the Mus musculus (UP000000589) UniProt database using Sequest HF through proteome discoverer (version 2.2). Enzyme specificity was set to trypsin and a maximum of two-missed cleavage sites were allowed. Oxidized methionine, N-terminal acetylation, and carbamidomethyl cysteine were set as variable modifications. Maximum allowed mass deviation was set to 10 ppm for monoisotopic precursor ions and 0.6 Da for MS/MS peaks.

91The resulting files were further processed using myProMS v3.6. FDR calculation used Percolator and was set to 1% at the peptide level for the whole study. The label free quantification was performed by peptide Extracted Ion Chromatograms (XICs) computed with MassChroQ version 2.2.2. For protein quantification, XICs from proteotypic peptides shared between compared conditions (TopN matching) with no missed cleavages were used. Median and scale normalization was applied on the total signal to correct the XICs for each biological replicate. To estimate the significance of the change in protein abundance, a linear model (adjusted on peptides and biological replicates) was performed and p-values were adjusted with a Benjamini–Hochberg FDR procedure with a control threshold set to 0.05.

92The accession number for the sequencing datasets reported in this paper is GEO: GSE116480. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD011344