Erosion of X Chromosome Inactivation in Human Pluripotent Cells Initiates with XACT Coating and Depends on a Specific Heterochromatin Landscape

Article Erosion of X Chromosome Inactivation in Human Pluripotent Cells Initiates with XACT Coating and Depends on a Specific Heterochromatin Landscape Graphical Abstract Authors Céline Vallot, Jean-François Ouimette,..., Marc Lalande, Claire Rougeulle Correspondence claire.rougeulle@univ-paris-diderot.fr In Brief Rougeulle and colleagues show that XCI erosion in human pluripotent cells is related to a pluripotency-specific epigenomic landscape on the inactive X chromosome, and that coating by the XACT lncrna and gene reactivation precedes loss of XIST expression during this process. Highlights d The inactive X chromosome in hpscs has a specific partitioned chromatin structure Accession Numbers GSE62562 d d d Erosion of X inactivation occurs in a subset of heterochromatin domains X inactivation instability is restricted to pluripotent cells Re-expression of XACT is an early step in the erosion of X inactivation Vallot et al., 215, Cell Stem Cell 16, 533 546 May 7, 215 ª215 Elsevier Inc. http://dx.doi.org/1.116/j.stem.215.3.16

Cell Stem Cell Article Erosion of X Chromosome Inactivation in Human Pluripotent Cells Initiates with XACT Coating and Depends on a Specific Heterochromatin Landscape Céline Vallot, 1,2,6 Jean-François Ouimette, 1,2,6 Mélanie Makhlouf, 1,2 Olivier Féraud, 3 Julien Pontis, 1,2 Julien Côme, 4 Cécile Martinat, 4 Annelise Bennaceur-Griscelli, 3 Marc Lalande, 5 and Claire Rougeulle 1,2, * 1 Epigenetics and Cell Fate, Université Paris Diderot, Sorbonne Paris Cité, 7513 Paris, France 2 CNRS, UMR7216 Epigenetics and Cell Fate, 7513 Paris, France 3 ESTeam Paris Sud, INSERM U935, Université Paris Sud 11, AP-HP, Villejuif 9482, France 4 INSERM/UEVE UMR 861, ISTEM, AFM, 913 Evry Cedex, France 5 Stem Cell and Systems Genomics Institutes, University of Connecticut, Farmington, CT 63, USA 6 Co-first author *Correspondence: claire.rougeulle@univ-paris-diderot.fr http://dx.doi.org/1.116/j.stem.215.3.16 SUMMARY Human pluripotent stem cells (hpscs) display extensive epigenetic instability, particularly on the X chromosome. In this study, we show that, in hpscs, the inactive X chromosome has a specific heterochromatin landscape that predisposes it to erosion of X chromosome inactivation (XCI), a process that occurs spontaneously in hpscs. Heterochromatin remodeling and gene reactivation occur in a non-random fashion and are confined to specific H3K27me3-enriched domains, leaving H3K9me3-marked regions unaffected. Using single-cell monitoring of XCI erosion, we show that this instability only occurs in pluripotent cells. We also provide evidence that loss of XIST expression is not the primary cause of XCI instability and that gene reactivation from the inactive X (Xi) precedes loss of XIST coating. Notably, expression and coating by the long non-coding RNA XACT are early events in XCI erosion and, therefore, may play a role in mediating this process. INTRODUCTION Human pluripotent stem cells (hpscs) display an unlimited growth capacity as well as the ability to differentiate into a wide variety of highly specialized cell types and are fundamental tools of applied biology and regenerative medicine (Tabar and Studer, 214; Takahashi and Yamanaka, 213). hpscs and, in particular, human embryonic stem cells (hescs) also represent an emerging model for studying human development (Zhu and Huangfu, 213). hescs are now widely used to decipher the epigenetic remodeling that characterizes embryonic stages and accompanies early cell fate decisions (Bártová et al., 28; Brunner et al., 29; Guenther et al., 21; Hawkins et al., 21; Shipony et al., 214). This research into hesc pluripotency is also critical for therapeutic approaches because hpscs have to maintain epigenetic features that mirror, as much as possible, normal development. Indeed, although epigenetic marks may not be key regulators of pluripotent cell identity, they control genes involved in lineage commitment as well as specific regions of the genome, such as centromeres, imprinted loci, and one X chromosome in female cells (Vallot and Rougeulle, 212). Perturbations of the epigenetic landscape of embryonic stem cells might not directly affect undifferentiated cells but may have drastic consequences on their differentiated progeny. Monitoring epigenetic modifications and their instability in hpscs is, therefore, an important step on the road to therapeutic applications. A major level of epigenetic variation in hpscs arises at the level of the X chromosome. In female mammals, one X chromosome is transcriptionally inactivated during early development. X chromosome inactivation (XCI) is controlled by XIST, a master long non-coding RNA (lncrna) that coats the X chromosome in cis and triggers transcriptional silencing and accumulation of repressive histone modifications. The inactive X (Xi) is characterized by hypermethylation of histone H3 lysine 9 (H3K9) and histone H3 lysine 27 (H3K27) (Wutz, 211). XIST is essential for the initiation of the X inactivation process (Pontier and Gribnau, 211) but appears to be mostly dispensable for maintenance of the inactive state (Brown and Willard, 1994; Csankovszki et al., 1999), which, when established, is stably propagated throughout cell divisions for the entire life of the individual. hpscs represent, however, a notable exception to the stable transmission of an inactive X chromosome. XCI has already occurred in the majority of female hpsc lines but is highly unstable in culture, and key features of the inactive state are gradually lost in a manner that appears to be irreversible (Hall et al., 28; Mekhoubad et al., 212; Shen et al., 28; Silva et al., 28). This erosion of XCI (Mekhoubad et al., 212) in hpscs is characterized by loss of XIST expression and XIST coating, global loss of H3K27 tri-methylation (me3) as monitored by immunofluorescence (IF), reduced promoter DNA methylation, and reexpression of several X-linked genes. We recently discovered XACT, a lncrna that displays the unique property of coating the active X (Xa) chromosome specifically in hpscs, and we showed that coating by XACT is also a feature of the eroded X (Xe) (Vallot Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. 533

A B C D E Figure 1. The Pluripotent-Specific Epigenomic Landscape of the Human X Chromosome (A) Schematic of XCI status and its analysis in hescs. Scale bars, 5 mm. (B) ChIP-seq profiles (log2 enrichment over input) and chromosomal density maps of peaks detected for H3K27me3 (green tracks) and H3K9me3 (blue tracks) in human male H1 and female H9 hescs and in IMR9, a fibroblast female cell line derived from normal fetal lung. The position of the XIST locus is indicated at the top. (C) Percentage of peak occupancy per chromosome for H1, H9, and IMR9. Red dotted lines represent median genomic occupancy. 534 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. (legend continued on next page)

et al., 213). The XCI-eroded phenotype has begun to be characterized (Mekhoubad et al., 212; Nazor et al., 212), and modification of expression patterns, altered growth, and differentiation properties have been associated with XCI erosion (Anguera et al., 212). However, the consequences of this instability on stem cell identity and fate remain to be precisely assessed. In addition, the exact extent of XCI erosion and the causes of this instability must be identified to develop strategies aiming at stabilizing the XCI status. In this context, it is particularly intriguing that erosion of XCI has never been reported to occur in differentiated cells. Here we explore the heterochromatin organization of the X chromosome in hescs and its link to XCI erosion. Chromatin immunoprecipitation sequencing (ChIP-seq) analysis reveals that the Xi in hescs is partitioned in distinct, non-overlapping epigenomic domains characterized by either K3K9 trimethylation (H3K9me3) or H3K27 trimethylation (H3K27me3) and that this bi-partite organization is not conserved in primary differentiated cells. By combining epigenomic studies with allelic RNA sequencing (RNA-seq), we show that erosion of XCI is not a chromosome-wide process. Rather, it appears to be an organized phenomenon, restricted to H3K27me3-rich domains, in which chromatin and transcriptional alterations occur. We developed an innovative RNA fluorescence in situ hybridization (FISH) approach to probe this instability at the single cell level, and, using this strategy, we linked erosion of XCI to pluripotency. Finally, we provide evidence that loss of XIST expression is not the cause of X inactivation instability and show that gene reactivation from Xi precedes loss of XIST coating. In particular, expression and coating of the X chromosome by XACT appears to be a precocious event in XCI erosion, which could suggest a role for XACT in the epigenetic instability of hescs. RESULTS A Pluripotent-Specific Epigenomic Partitioning of the Inactive X Chromosome X inactivation is highly unstable in hescs, and cells harboring various XCI states can co-exist within the same culture. We therefore established an experimental procedure to isolate and maintain homogeneous hesc populations with respect to their XCI status (Figure 1A; Supplemental Experimental Procedures; Vallot et al., 213), which is an absolute pre-requisite for robust ChIP-seq and RNA-seq analyses. We obtained pure populations of post-inactivation H9 and HUES1 hescs carrying either an Xi coated by XIST and enriched in H3K27me3 or an Xe lacking both XIST coating and accumulation of H3K27me3. We termed these populations XaXi and, respectively. Importantly, the XCI status was monitored on a regular basis and prior to any molecular analysis, and samples deriving from the original homogeneous XCI configuration were discarded. We first investigated, by ChIP-seq, the heterochromatin organization of the X chromosome in XaXi H9 cells, analyzing specifically H3K9me3 and H3K27me3 because both marks characterize the Xi facultative heterochromatin in humans (Boggs et al., 22; Chadwick and Willard, 24). Visualization of H3K9me3 and H3K27me3 profiles indicates that these marks are already highly abundant on the X chromosome in female hescs (Figure 1B). Discretization of the ChIP-seq profiles using jahmm (Filion and Cuscó, 214) shows that 25% of the X chromosome is covered by H3K27me3 peaks and 32% by H3K9me3 peaks and that these regions do not seem to be randomly distributed. H3K27me3 peaks are organized in several large and well delimitated domains, the major one spanning around 5 Mb in the middle of the long arm. The XIST locus lies within an H3K27me3 domain. However, XIST itself is devoid of this mark, as expected for an expressed gene (Figure S1A). H3K9me3 peaks cover most of the short arm as well as the most proximal and distal parts of the long arm. A second observation is that X is the chromosome that displays the highest accumulation of H3K9 and K27 tri-methylation in female hescs (Figure 1C). Very little enrichment is found on the autosomes (median coverage of 1% and 5% for H3K27me3 and H3K9me3, respectively), with the exception of chromosome 19, on which several clusters of ZNF genes are enriched for H3K9me3 but not for H3K27me3, as described previously (Vogel et al., 26). The Xa is also mostly devoid of those two repressive marks (.5% and 1% coverage for H3K27me3 and H3K9me3, respectively), as seen in male hescs (H1, Table S1; Figure 1B), indicating that the enrichment detected in female hescs corresponds to the Xi. Strikingly, visualization of the H3K9me3 and H3K27me3 peak density map along the X chromosome reveals that these two marks, overall, occupy distinct chromosome domains in female hescs. H3K9 and H3K27 tri-methylation indeed shows an anti-correlated profile (Pearson s correlation score between log2 enrichments, r =.61, p < 1 6 ; Figure 1D), and they co-occupy only 3% of the X chromosome. Such an anticorrelated distribution was confirmed in two additional XaXi hescs datasets (Figure S2A). In female differentiated cells, H3K9me3 and H3K27me3 profiles are markedly divergent from that of hescs regarding both X versus autosome distribution as well as the partitioning of the two marks on the Xi chromosome (Figures 1B, 1C, and 1E). First, as reported previously (Hawkins et al., 21), there is a marked increase relative to hescs in the fraction of autosomes covered by these heterochromatic marks, ranging from 1% 49% of their sequence in differentiated cells, depending on the modification and the chromosome studied (Figure 1C; Figures S1B S1C). The overall coverage of H3K9me3 on the X chromosome is within the same range as that of autosomes (24%; median, 25%). H3K27me3 coverage on the X chromosome remains the highest (32%; median, 16%) but is within the range of some autosomes (chr2 and chr17, for example). Strikingly, this observation does not apply to the Xa, which is overall depleted of H3K27me3 and displays variable enrichment of H3K9me3 (Figures S1B and S1C). The relative hypomethylated status of the (D) Top: Euler diagram displaying megabases of the X chromosome covered by H3K27me3 and H3K9me3 peaks in XaXi H9 cells. Bottom: superposition of H3K9me3 and H3K27me3 ChIP-seq profiles in H9 hescs. Log2 ratios were compared using Pearson s correlation test, and the associated p value was calculated using random permutations of the datasets. (E) Same analysis as in (D) for IMR9 cells. See also Figures S1 and S2 and Table S1. Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. 535

A B C D E F G H Figure 2. XCI Erosion Is Associated with Restricted Epigenomic Remodeling (A) ChIP-seq profiles and density maps of peaks detected for H3K27me3 (green tracks) and H3K9me3 (blue tracks) in XaXi and H9 hescs. Traces show log2 enrichment over input. The dotted rectangle delineates the zoom shown in (F). (B) Percentage of peak occupancy per chromosome in XaXi (light color) and (dark color) H9 hescs for H3K27me3 (green) and H3K9me3 (blue). (C) Scatterplot comparing H3K27me3 and H3K9me3 log2 enrichment ratios between XaXi and H9 cells for chromosomes X and 19. Log2 ratios were compared using Pearson s correlation tests, and the p value was computed as in Figure 1. (D) Scatterplot comparing H3K27me3 log2 enrichment ratios on the X chromosome in male versus female XaXi or H9 cells. (legend continued on next page) 536 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc.

Xa might be linked to its known upregulation, which compensates for the X/autosome gene dosage imbalance in mammalian cells (Deng et al., 211). Second, there is a partial redistribution of H3K27me3 and H3K9me3 along the Xi between pluripotent and differentiated cells (Figure 1B; Figures S1D S1G); only half of the H3K27me3 and H3K9me3 peaks in pluripotent cells are conserved in differentiated cells. Moreover, the overlap between H3K9me3 and H3K27me3 peaks on the Xi is increased in differentiated cells compared with pluripotent cells (Figures 1D and 1E, top; Fisher s exact test, p < 1 16 ), and H3K27me3 and H3K9me3 profiles are no longer anti-correlated in IMR9 cells (Figure 1E, bottom, r =.7). This observation was confirmed in two additional primary cell types (Figure S2B). In contrast, immortalized cells display an anti-correlated distribution of H3K9me3 and H3K27me3, as described previously (Chadwick and Willard, 24; Nozawa et al., 213), and are, in this respect, clearly distinct from primary cells (Figure S2C). Our analysis therefore reveals a partitioning of the facultative heterochromatin on the Xi in hescs, characterized by either H3K9 or H3K27 tri-methylation. This organization is not maintained in primary differentiated cells, where more regions are co-enriched in both marks. This pronounced change in the epigenomic landscape of the Xi between pluripotent and differentiated cells suggests that XCI, although already initiated, is not in its final state in hescs. This might contribute to the different sensitivity of the two cell types to XCI instability. Epigenomic Instability of the Inactive X Chromosome Is Restricted to H3K27me3 Domains Erosion of XCI in hpscs has been shown to be associated with aberrant DNA methylation profiles at gene promoters (Mekhoubad et al., 212; Nazor et al., 212; Shen et al., 28), but the global heterochromatin landscape of the Xe has not been investigated. To do so, we conducted a ChIP-seq analysis in pure hesc populations (Figure 2). This analysis reveals that the Xe is characterized by a massive loss of H3K27me3 (Figure 2A), in agreement with this mark being dependent on continuous XIST expression and coating. The percentage of the Xe occupied by H3K27me3 peaks is reduced to 1%, within the same range as that of the Xa and autosomes (Figures 1, 2A, and 2B). This loss is specific to the X chromosome; the few loci enriched in H3K27me3 on chromosome 19 do not undergo such remodeling (Figure 2C). The H3K27me3 profile on the Xe strongly resembles that of the Xa in male hesc (Figure 2D, r =.73, p < 1 6 ). Although H3K27me3 peaks on the Xi cluster in large domains (largest peak cluster of 31 kb) and are equally distributed between intergenic and genic regions, the peaks on the Xe and Xa aggregate into smaller clusters (maxima of 29 kb and 26.5 kb, respectively) that are restricted to 5 and, to a lesser extent, 3 regions of X-linked genes (Figure 2E). Such a distribution suggests a role in local, gene-specific silencing, as seen on autosomes. In marked contrast to H3K27me3, the overall profile of H3K9me3 enrichment is maintained on the Xe (r =.87, p < 1 6 ; Figures 2A 2C). The distribution of the peaks is maintained (Figure 2A), but we observed, in cells, a lower performance of peak detection compared with XaXi cells, which is more pronounced for the X chromosome compared with chromosome 19 (41% and 11% variation in H3K9me3 peak coverage, respectively; Figure 2B). H3K9me3 peaks cluster in larger domains on the Xi and Xe compared with the Xa (largest clusters of 278 kb, 223 kb, and 51 kb, respectively). The boundaries of the H3K9me3 domains on the X chromosome are also largely conserved between XaXi and hescs. In some cases, however, limited spreading of H3K9me3 to nearby H3K27me3- depleted regions was observed (see zoom, Figure 2F). Maintenance of H3K9me3 in hescs was confirmed in an additional cell line, WIBR3 (Figure S2D). These results indicate that maintenance of H3K9 tri-methylation on the Xi is not dependent on XIST. They also imply that H3K9me3 distribution is not massively constrained by H3K27me3 because this would have led to global accumulation of H3K9me3 on the Xe. Together, our findings reveal that erosion of XCI is not accompanied by a complete reorganization of the heterochromatin landscape of the X chromosome but that chromatin changes appear to be limited to specific domains, in particular those enriched in H3K27me3. To summarize our findings, we compared the heterochromatin landscape of the X chromosome across all previously mentioned cellular states, pluripotent (XaXi and ) and differentiated (primary and immortalized cells), through a correlative matrix of H3K9me3 and H3K27me3 enrichment profiles along the X chromosome (Figures 2G and 2H). Hierarchical clustering confirmed that both H3K27me3 and H3K9me3 profiles discriminate hescs from differentiated cells. Within these cellular contexts, XaXi hescs are distinguished from hescs by their H3K27me3 profile (Figure 2G), whereas primary differentiated cells differ from immortalized cells by their H3K9me3 profile (Figure 2H). Transcriptional Reactivation of the Xe Is Partial and Corresponds to H3K27me3 Domains A substantial variation in X-linked gene expression associated with XCI erosion has been reported, but this was mostly based on microarray analyses (Mekhoubad et al., 212), which assess expression levels rather than allelism. To probe X-linked gene reactivation and to integrate this information with the epigenomic variations we identified, we generated RNA-seq data from pure (E) Genomic association between H3K27me3 peaks and gene annotation. Log2 enrichment and associated q values were computed using the Genomic Association Tester (GAT). (F) Zoom of H3K27me3 (green) and H3K9me3 (blue) ChIP-seq profiles for the X chromosome region most enriched in H3K27me3 (85 145 Mb) in XaXi (light) and (dark) H9 hesc. The graphs show log2 enrichment over input. Black horizontal lines correspond to regions with higher H3K9me3 enrichment in versus XaXi H9. (G) Hierarchical clustering of correlation scores between H3K27me3 log2 enrichments along the X chromosome in all cellular contexts studied. hescs are indicated in black and differentiated cell lines in dark yellow; XCI status is indicated in brackets. The intensity and size of circles is relative to the correlation score between two H3K27me3 ChIP-seq datasets. Black squares indicate clustering groups. (H) The same analysis for H3K9me3. See also Figure S2 and Table S1. Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. 537

A B C D E Figure 3. Transcriptional Reactivation upon XCI Erosion Is Restricted to H3K27me3 Domains (A) Chromosomal maps of informative SNPs detected in genomic DNA (black) and in the transcriptome of XaXi and H9 cells, aligned with the epigenomic map of the XaXi H9 (H3K27me3, green; H3K9me3, blue). Blue and red lines indicate biallelic expression (escapees [biallelic in XaXi and ] and reactivated genes [biallelic in only], respectively), whereas gray lines correspond to monoallelic expression in XaXi and cells. Dotted rectangles delineate the zoom shown in (C E). (B) Enrichment of H3K27me3 and H3K9me3 along the 78 genes with respect to their expression status in XaXi and H9 cells. The lines display the average log2 ratio of immunoprecipitation (IP) over input, and shaded color represents SEM. The color code is the same as in (A). (C E) Zoom of X chromosome regions with corresponding H3K27me3 (green) and H3K9me3 (blue) profiles in XaXi (light) and (dark) cells. The color of the vertical bars between the two ChIP-seq profiles corresponds to gene expression status as in (A) (red for reactivated genes and gray for monoallelically expressed genes). Black horizontal lines correspond to regions with higher H3K9me3 enrichment in versus XaXi H9 cells. See also Figure S3 and Table S2. populations of XaXi H9 and H9 cells. We took advantage of the clonal XCI pattern of H9 cells (Mitjavila-Garcia et al., 21; Shen et al., 28) to analyze the data in an allelic manner using a list of H9-informative SNPs we identified using a dataset of whole genome sequencing for H9 (Supplemental Experimental Procedures). This approach allowed us to investigate the allelic expression of 78 genes evenly distributed along the X chromosome for which one or more informative SNPs are available and that are covered by ten or more reads in both RNA-seq datasets (Figure 3A; Table S2). Three categories of genes were identified based on their allelic expression profiles (Figure 3A). In the first category are genes that are biallelically expressed in XaXi and H9 cells. These are transcripts that escape XCI; i.e., escapees. They mostly localize to the short arm of the 538 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc.

chromosome, and some of them have been identified previously as escapees. The percentage of escapees identified in hescs (13%) falls within the range that has been described in differentiated cells (1% 15%) (Carrel and Willard, 25). The second class corresponds to genes biallelically expressed specifically in cells (Fisher s exact test, p <.5), confirming that the Xe is characterized by some degree of gene reactivation. These reactivated genes are not distributed evenly along the chromosome but tend to cluster, notably in a 38-Mb region on the long arm (96 134 Mb). More importantly, only 28% of the assessed genes are reactivated in cells. In contrast, the majority of X-linked genes (59%), constituting the third category, termed mono-allelic, remains mono-allelically expressed in cells. Notably, a large 46-Mb chromosomal domain (from 39 85 Mb) appears to be resistant to transcriptional reactivation. This demonstrates that X chromosome reactivation, although substantial, is incomplete. The RNA-seq results were validated using pyrosequencing and FISH for several genes belonging to each of the three classes and for novel lncrnas we identified (Figure S3). Because re-expression from the Xe is not a global event but is limited to only a subset of genes, we integrated epigenomic features to our expression data to identify potential epigenomic signatures of X chromosome reactivation. At the chromosomal level, reactivated genes map within domains enriched in H3K27me3 on the Xi. In contrast, genes that are maintained silent on the Xe tend to be localized in H3K9me3 regions (Figure 3A). To further probe the link between expression status and chromatin modifications, we compiled the H3K9 and H3K27me3 profiles along the 78 SNP-carrying genes (including upstream and downstream regions) for each category (Figure 3B). This analysis reveals a strong correlation between H3K27me3 and susceptibility to reactivation. Indeed, reactivated genes are characterized by significantly higher levels of H3K27me3 on the Xi relative to escapees and inactive genes (1 SEM difference in average log2 enrichments). On the Xe, all genes display low H3K27me3, in agreement with this mark being lost (Figure 2; Hall et al., 28; Shen et al., 28; Silva et al., 28). In contrast, H3K9me3 preferentially marks mono-allelic genes, although reactivated ones and escapees also display some level of enrichment, mainly around the transcription start site (TSS). On the Xe, inactive genes maintain higher levels of H3K9me3 compared with escapees and reactivated ones, but promoter regions are no longer enriched. A higher-resolution display of several regions of the X chromosome further illustrates the epigenomic features associated with reactivation (Figures 3C 3E). Genes located in H3K27me3 domains, whether on the central part of the largest domain on the long arm of the X chromosome (from MORF4L2 to RP1-24P17.1, Figure 3C) or in a more distal domain (lnc121, Figure 3D) show reactivation concomitant with H3K27me3 loss. Genes located in H3K9me3-rich regions (CSTF2 and ARMCX3, Figure 3C; CDKL5, Figure 3E) are maintained silent on the Xe. Interestingly, genes located at the edge of H3K27me3 domains, such as WDR44, DOCK11, and BTB33, also fail to reactivate (Figure 3C). This could be linked to local spreading of H3K9me3 because these boundary regions display elevated levels of H3K9me3 on the Xe compared with the Xi. Together, our analysis reveals that, although erosion of XCI in hescs is associated with reactivation of genes from the Xe, this reactivation is only partial and linked to the epigenomic signature, with only genes within H3K27me3-rich domains being prone to reactivation. Erosion of XCI Is Linked to Pluripotency Large-scale characterization of transcription by RNA-seq is a very powerful approach but does not routinely probe for cellto-cell variability because it is usually performed on cell populations. It is, in addition, costly and time-consuming, especially regarding the analysis of the large datasets produced. To circumvent these problems, we developed a method for largescale visualization of transcriptional activity at the single-cell level and applied it to monitor X chromosome activity in the context of pluripotent and differentiated cells. This method, called RNA-paint, relies on an adaptation of the RNA-FISH procedure to be used with commercially available chromosome paint probes (Experimental Procedures; Figure S4A). An X chromosome paint probe can indeed detect, by RNA-FISH, a unique broad transcription signal in female cells carrying an Xi in both hescs (Figure 4A, left) and fibroblasts (Figure S4B). The RNA origin of the hybridization signal is confirmed by RNase treatment of the cells prior to RNA-FISH (Figure 4A, right). RNA/ DNA-FISH shows that the signal indeed originates from one of the two X chromosomes, most likely the Xa (Figure 4B). Indeed, this broad X-paint RNA cloud overlaps that of XACT RNA (Figure 4C), which coats the Xa in XaXi hescs (Vallot et al., 213). The X-paint RNA signal also encompasses that of ATRX (Figure 4D), a gene resistant to X chromosome reactivation and expressed solely from the Xa (Figures S4C and S4D), but is clearly distinct from the XIST RNA cloud, which coats Xi (Figure 4E). These experiments indicate that the X-paint RNA cloud corresponds to the Xa. The RNA-paint technique can also be applied to other chromosomes, as shown for chromosome 4, where two broad domains are observed in 1% of the cells (Figure 4F), an observation that is consistent with both chromosomes 4 being active. Our RNA-paint approach therefore allows the visualization of chromosome-specific transcription territories within the nuclear space. We then tested whether RNA-paint could be used to probe XCI erosion in hescs. In striking contrast to XaXi hescs, X-RNApaint detects a second transcription territory in the majority of H9 cells (62%, n = 1, Fisher s exact test, p < 1 3 ; Figure 4G; Figure S4E). Similar results were obtained in another hesc cell line, HUES1, with 81% of HUES1 cells displaying two X chromosome transcription territories (Figure S4F). This secondary signal corresponds to the Xe chromosome, and more specifically to the outer edge of the chromosome territory, as shown by successive RNA/DNA-FISH (Figure S4G) and by co- RNA-FISH with ATRX (expressed only from the Xa) and POLA1 (expressed from Xa and Xe; Table S2; Figure 4H). The Xe transcription territory is located in a nuclear region enriched for H3K4me2 (Figure 4I), a chromatin mark that characterizes potentially active regions and is globally excluded from the Xi (Boggs et al., 22; Heard et al., 21). Together, these results confirm the substantial reactivation of the Xe in hescs and indicate that RNA-paint can not only efficiently detect XCI erosion but also assess the nuclear localization of altered X chromosome activity. Measurement of the size of the transcription territories provides further information on the extent of transcription. The Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. 539

A C E H F B I G D Figure 4. Single-Cell Monitoring of X Chromosome Activity in hescs (A) X-paint RNA-FISH (yellow) in XaXi H9 cells with or without initial RNase treatment. (B) RNA/DNA FISH in XaXi H9 cells using X-paint RNA-FISH (yellow) prior to X chromosome DNA- FISH (green). Only one of the two Xs displays an X- paint RNA signal and is, therefore, the Xa, the other being the Xi (arrows). (C) Double RNA-FISH staining with the X-paint (yellow) and the long non-coding RNA XACT (red) probes in XaXi H9 cells. (D and E) Double RNA-FISH staining with the X- paint (yellow) and ATRX or XIST probes (green) in XaXi H9 cells. (F) RNA-FISH with a chromosome 4 paint (yellow) in XaXi H9 cells. (G) X-paint RNA-FISH in H9 cells (left). Shown is the quantification of the ratio between the volume of Xe versus Xa chromosome transcription territories for nuclei with two such territories (right, n = 5). The black line indicates the average value (24.1%). (H) Double RNA-FISH staining with X-paint (yellow) and ATRX (left, green) or POLA1 (right, green) probes in H9 cells. The Green asterisk indicates the presence of pinpoints for ATRX or POLA1. (I) Immunofluorescence staining for histone H3 lysine 4 dimethylation (H3K4me2, green) prior to X- paint RNA-FISH (red) in H9 cells. The quantification of IF and RNA-FISH signals along the white line starting from the top left (position mm, x axis) to the bottom right (position 35 mm, x axis) is shown. The numbers in the right bottom corners of the images indicate the percentage of nuclei displaying the staining shown. Blue shows DAPI staining. Scale bars, 5 mm. See also Figure S4. transcription territory corresponding to the Xe occupies a nuclear volume four times smaller than that of the Xa (24% on average, n = 5; Figure 4G, right). This parallels the proportion of reactivated genes identified by allelic RNA-seq (28%). Erosion of XCI occurs spontaneously in hpscs but has never been reported in differentiated cells. Indeed, targeted deletion of XIST from a stably inactivated X chromosome does not result in large-scale X chromosome reactivation in the either mouse or human (Brown and Willard, 1994; Csankovszki et al., 1999). To assess the dependence of XCI erosion on cellular context, we induced H9 and HUES1 hescs cells to differentiate and monitored the transcriptional activity of the Xe chromosome by RNA-paint. cells differentiated efficiently into neural stem cells (NSC), as judged by morphology (Figure 5A), decreased expression of the pluripotency marker OCT3/4, and expression of the NPC markers NESTIN and SOX2 (Figures 5B and 5C; Figure S5A). We observed, upon neural differentiation, a progressive disappearance of the Xe RNA-paint signal (Figure 5D; Figure S5B). In contrast, transcription from the Xa continued to be detected both in XaXi and cells. We confirmed, by pyrosequencing and sequencing, re-silencing of some individual genes and lncrnas (Figures S5C and S5D). Others, however, maintained their biallelic expression status in a differentiated context (Figure S5E), as described previously (Mekhoubad et al., 212). Xe silencing was not mediated by XIST, which remained repressed throughout differentiation (Figure 5B). Global re-silencing of the Xe was also observed when an alternative differentiation protocol was used (Figures S5F and S5G). These results suggest that, in a XIST-negative environment, erosion of XCI is a feature of pluripotent but not differentiated cells. To further assess the link between XCI erosion and pluripotency, we took advantage of a model system in which H9 cells are submitted to successive rounds of differentiation and reprogramming (Giuliani et al., 211; Vallot et al., 213). In this system, H9 cells are differentiated into first generation mesenchymal stem cells (MSC-1) and subsequently reprogrammed into induced pluripotent cells (ipscs). These ipscs were then differentiated into second-generation mesenchymal stem cells (MSC-2). Although none of these cells expressed XIST, significant transcription from the Xe can be detected in pluripotent cells (ESCs and ipscs) but not in differentiated MSCs (Figure 5E; n = 1, Fisher s exact test, p < 1 3 ). Our results indicate that RNA-paint is a powerful approach to visualize chromosome transcription territories and to monitor X chromosome reactivation in a single-cell manner. This approach 54 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc.

also reveals that erosion of dosage compensation occurs specifically in a cellular context linked to pluripotency. Gene Reactivation and XACT Coating of the Xe Precedes XIST Silencing Erosion of XCI is characterized by several features, including heterochromatin remodeling, altered coating of X chromosomes by XIST and XACT, and X-linked gene reactivation. To determine the order in which these events take place, we decided to monitor the kinetics of the transition between the inactive and the eroded state. In H9 cells, the transition between the XaXi and states appears to be a quick event because the vast majority of cells are found in either configuration. We therefore searched for other hesc lines in which we could efficiently capture the transition state and selected WIBR2, which was initially described as being in a pre-inactivation state, carrying two Xa under low-oxygen conditions (Lengner et al., 21). In our hands, WIBR2 cells had already initiated XCI and retained a Xi coated by XIST, even in 5% O 2, as described elsewhere (Gafni et al., 213). However, this XCI status is much more stable at lower than at higher O 2, with a much slower progression of XCI erosion, as monitored by XIST expression. 1% of WIBR2 cells displayed XIST coating of the X chromosome from passages 29 38 (Figure 6A), suggesting a XaXi state for these cells. However, when we used RNA-paint to assess the transcriptional status of the X chromosomes in WIBR2 cells, we could detect nuclei with a second X chromosome transcription cloud, indicating that erosion of XCI is already occurring in these cells (Figure 6B). Strikingly, X chromosome reactivation occurs while XIST is still expressed and coating the chromosome, indicating that X chromosome reactivation takes place prior to XIST silencing. We named the chromosome marked by XIST and X-paint RNA Xt, for X in a transition state. Similar observations were made when XACT expression was assessed. In WIBR2 cells at passage 29, XACT co-accumulates with XIST in 35% of nuclei (Figure 6C). This fraction increases with passage number, indicative of a progression within the population toward the eroded state (Figure S6A). H3K27me3 enrichment is also maintained on the Xt upon XACT accumulation (Figure S6B). In contrast, we never observed cells in which the Xt was neither coated by XIST (or H3K27me3) nor by XACT, which would have been indicative of XIST repression and chromatin remodeling taking place prior to XACT re-expression. We also observed XACT re-expression from the Xi prior to XIST silencing in H9 cells. Indeed, when H9 XaXi cells were maintained in culture, up to 1% of cells could be observed in the transition phase and displayed co-accumulation of XACT and XIST on the Xt (Figure S6C). Together, these data indicate that X chromosome reactivation initiates before loss of XIST accumulation and epigenomic remodeling. Next we compared the kinetics of expression of XACT and other X-linked genes in relation to XIST by performing three-color RNA-FISH in WIBR2 cells. We analyzed two transcripts identified previously in H9 cells as being re-expressed from the Xe, POLA1, and lnc121 (Figures 6D and 6E). RNA-paint could not be used in such a three-color configuration because of reduced sensitivity. First, we did observe reactivation of these two genes in WIBR2 cells as in H9 cells, suggesting that the erosion pattern is not cell line-specific. Expression from the Xt was found to initiate only at late passages (Figure 6D) without loss of XIST coating, confirming our previous observation. At passage 38, WIBR2 cells expressed POLA1 and lnc121 from the XIST-coated chromosome in 28% and 22% of nuclei, respectively (Figures 6D and 6E). However, no or very few cells (1%) showed expression of either gene on the XIST-coated chromosome in the absence of XACT. Conversely, co-accumulation of XACT and XIST occurred in the majority of nuclei with no apparent expression of POLA1 or lnc121 (52% and 5%, respectively). These results suggest that XACT expression from the Xt does not merely reflect X chromosome reactivation. Instead, X chromosome reactivation appears secondary to XACT expression and coating of the Xt. We further investigated the concomitant accumulation of XACT and XIST on the Xt and observed that XIST and XACT systematically occupy distinct nuclear domains. Very little overlap could be detected between the two RNA-FISH signals (median co-occupied volume of.3 mm 3 ; Figure 6F). In addition, the respective volumes occupied by XIST and XACT when co-accumulating on the Xt were significantly smaller than the volume occupied by either RNA alone. XIST RNA occupied 3.6 mm 3 on the Xi versus 2.6 mm 3 on the Xt (median values, p =.2), and XACT RNA occupied 4 mm 3 on the Xa and.8 mm 3 on the Xt (median values, p < 1 6 ). This suggests that these lncrnas could affect each other s accumulation and that XACT re-expression and accumulation on the Xi could impact XIST accumulation. In conclusion, our kinetic analysis revealed that XIST repression is not the first event in the XCI erosion process and appears to be a consequence rather than a cause of erosion (Figure 6G). Partial reactivation of the Xi precedes the loss of XIST coating. In particular, reactivation and accumulation of XACT on the inactive X take place prior to robust gene reactivation and loss of XIST expression. XACT is therefore an early marker of XCI erosion in hpscs. DISCUSSION Here we unravel a pluripotent specific epigenomic landscape of the inactive X chromosome in humans and demonstrate that the erosion of XCI, which occurs spontaneously in hpscs, is not a chromosome-wide process but, rather, is restricted, both at the epigenomic and transcriptional levels, to subdomains of the X chromosome. We show that erosion of XCI is tightly associated with pluripotency, and we demonstrate that, surprisingly, loss of XIST expression is a consequence rather than a cause of this event. We reveal that gene reactivation and, in particular, XACT expression and coating precede XIST repression and are therefore more precocious markers of the erosion process. Characterizing the details of XCI in hpscs is essential but challenging, given the major instability of the process. Because drift in the XCI status occurs in only few passages and because multiple XCI states can co-exist within a given culture, conclusions drawn from experiments performed on cell populations must be interpreted with care. Purification of homogenous cell populations and constant monitoring of their XCI status has allowed us to precisely assess epigenomic and transcriptomic alterations associated with XCI erosion. Allelic investigation of RNA-seq data was also instrumental in unambiguously identifying relaxed genes and genes resistant to XCI erosion and correlating their genomic localization with epigenomic features. Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. 541

A B C D E Figure 5. Erosion from XCI Is Linked to Pluripotency (A) Bright-field images of differentiation of H9 cells into NSCs from day (d) to day 9 (d9). Scale bars, 15 mm. (B) RT-qPCR analysis of OCT3/4 and XIST expression during H9 differentiation. (C) Characterization of NSCs by immunofluorescence using antibodies against NESTIN (green) and SOX2 (red). Quantification of positive cells was performed by ArrayScan, and the percentage of positive cells is indicated. Scale bars, 1 mm. 542 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. (legend continued on next page)

Such an approach is more precise in terms of allelic determination than analysis of microarray datasets because elevated expression levels in cells compared with XaXi cells can result from mechanisms other than biallelic expression. Our findings that not all genes are equally susceptible to X chromosome reactivation add to the complexity of the system and suggest that discordant conclusions regarding the XCI status can be drawn when only single genes are analyzed by RNA- FISH. Such gene-specific differences in allelic expression emphasize the importance of monitoring the entire chromosome for transcriptional activity. RNA-paint proved to be a powerful approach, allowing the visualization of chromosome-specific transcription territories within the nucleus. RNA-paint is not, however, suitable to detect individual gene expression, such as nascent transcripts from individual escapees or XIST RNA from the Xi. This could be due to the probe s coverage of individual loci (for example, the 32-kb XIST locus represents only.21% of the X chromosome DNA) or to the genomic content of the chromosome paint probe. In this context, maintenance of the reactivated status during differentiation for specific genes, as described here and by Mekhoubad et al. (212), would not be detected by RNA-paint. The ability to assess global chromosome activity directly with single-cell resolution has several important implications for both fundamental and applied biology. This method combines largescale transcriptome analysis with single-cell characterization, enabling the simultaneous assessment of 3D nuclear organization and chromosome-wide transcription. In the case of the X chromosome, RNA-paint can be used to gain insights into XCI at the single-cell level in heterogeneous samples and in lowcell-number and valuable material, such as pre-implantation embryos and human blastocysts. RNA-paint could also be used to probe for aberrant profiles of X chromosome activity in heterogeneous tumor samples. This is of relevance because several findings suggest that enhanced transcription of the X chromosome could participate in tumorigenesis and that the number of active Xs rather than the total number of X chromosomes, might be a key feature of an aggressive cancer cell (Richardson et al., 26; Spatz et al., 24; Vincent-Salomon et al., 27). Finally, RNA-paint can be useful to assess transcriptional activity in the case of chromosome translocations. So far, abnormal X chromosome activity profiles have only been reported for hpscs and aggressive cancer cell lines, potentially revealing a phenotypic advantage in rapidly dividing cells for increased X chromosome activity. The latter could be linked to the upregulation of X-linked oncogenes (Anguera et al., 212). On the other hand, XCI erosion has never been described in differentiated cells. XIST expression is stably maintained in such cells, both in culture and in vivo. In addition, induced deletion of XIST from the Xi does not lead to massive gene reactivation (Brown and Willard, 1994; Csankovszki et al., 1999). Here we demonstrated that, in an isogenic, XIST-negative context, global erosion of XCI as detected by RNA-paint is strictly linked to pluripotency. It is important to note, however, that some genes are maintained expressed in differentiated cells. This suggests that a subset of genes might not be properly dosagecompensated in differentiated cells derived from female hescs. This result has safety implications for the clinical use of hesc-differentiated derivatives; the long-term impact of bi-allelic expression for several X-linked genes should be assessed. The opposite susceptibility of pluripotent versus differentiated cells toward XCI erosion might be explained by the existence of protective mechanisms in the latter, the presence of promoting factors in the former, or a combination of both. Our finding that the epigenomic landscape of the Xi is different in hpscs and differentiated cells may provide insights. The redistribution of H3K27me3 and the resulting overlap between H3K27me3 and H3K9me3 in differentiated cells may ensure stabilization of the inactive state. Such a reorganization also demonstrates that the Xi in hpscs and the Xi in differentiated cells are distinct, differing at the epigenomic levels and probably regarding other features. It suggests that, although initiated, XCI is not in its final state in hpscs. XCI in these cells would rather correspond to an intermediate, transitory state, artificially but inefficiently fixed in culture. This is reminiscent of the short time window described in early differentiating mouse ES cells, during which XCI shifts from a reversible, Xist-dependent process to an irreversible, Xist-independent mechanism (Wutz and Jaenisch, 2). What triggers XCI instability in hpscs is unknown, but we show here that XIST is unlikely to play a causal role in this process. Indeed, we demonstrate that XACT expression and coating are so far the first detectable markers distinguishing the Xi from the Xe, preceding loss of XIST expression, loss of H3K27me3 accumulation, and global gene reactivation. Whether XACT directly promotes XCI erosion remains to be determined, but, in this context, it would be important to decipher the regulatory mechanisms controlling XACT to elucidate the mechanisms of XACT reactivation from the Xi and to further develop strategies aimed at preventing instability of XACT expression in hpscs. In addition to the precise characterization of XCI erosion in hpscs and its possible causes, our findings provide insights into the regulation of XCI in humans and the role of XACT in this process. Our observation that XIST and XACT can transiently co-accumulate on the X chromosome, but with limited overlap, suggests that XIST and XACT transcripts do not interact physically and that the two RNAs might compete for coating the chromosome. We propose that re-expression of XACT from the Xi in hpscs repels XIST RNA, initially from the XACT locus and progressively upon coating by XACT, from all XIST contact points. This would lead to epigenomic reorganization and to reactivation of genes within these domains. Transcriptional silencing of XIST, likely by mechanisms involving DNA methylation (Lengner et al., (D) X-paint RNA-FISH (yellow) during NSC differentiation of hesc cells. Staining for hescs (left) and NSCs (right) is shown with the percentage of cells displaying the corresponding staining. Scale bars, 5 mm. The graph indicates the percentage of nuclei (n > 1) with one (black) or two (white) X-paint RNA clouds for all time points. Nucleus distribution was compared every day to XaXi H9 differentiation with Fisher s exact test. ***p < 1 3 ; **p < 1 2 ; n.s., not significant. (E) H9 hescs were differentiated into MSC-1 cells, which were reprogrammed into ipscs, and these ipscs were differentiated to generate MSC-2 cells. Top: bright-field images. Scale bars, 1 mm. Bottom: X-paint RNA-FISH (yellow) with DAPI (blue). The numbers correspond to the percentage of nuclei displaying the corresponding staining. Scale bars, 5 mm. See also Figure S5. Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. 543

A B C D E F G Figure 6. XACT Transcription and Coating from the Xe Precedes XIST Silencing and X-Linked Gene Reactivation (A) XIST RNA-FISH (green) in WIBR2 cells at passages 29 (left) and 38 (right). (B) X-paint RNA-FISH in WIBR2 cells at passage 29. (C) Double RNA-FISH staining with XIST (green) and XACT (red) probes in WIBR2 cells at passage 29. Quantification is shown in Figure S6A. (D) Triple RNA-FISH with XACT (red), POLA1 (blue), and XIST (green) probes in WIBR2 cells at passages 29 and 38. The images show the most abundant expression pattern. The barplots describe the quantification of the observed expression patterns. Numbers of nuclei (n = 1) with each expression pattern were compared between the two passages using Fisher s exact test. ***p < 1 3. (E) Triple RNA-FISH with XACT (red), lnc121 (blue), and XIST (green) probes in WIBR2 passage 38 cells as in (D). (F) Computation of the volumes occupied by either XIST (green) or XACT (red) RNA on the Xi, Xa, or Xt chromosome in WIBR2 cells at passage 29 using 3D image segmentation. Also shown is the volume co-occupied by XIST and XACT RNA on the Xt (XACTXXIST). Volumes occupied by XIST or XACT RNA on Xt and Xi or Xa were compared using a Mann-Whitney-Wilcoxon test. Left and center: two opposite examples of XIST and XACT staining and volumes coating on the Xt with measured volume values (cubic micrometers) for XIST on the Xt and XACT on the Xa and Xt and for the overlap (yellow). Scale bars, 5 mm. 1 nuclei were analyzed. (G) Model for the transition of XaXi to hescs. The first events observed are the activation of XACT transcription and subsequent coating of the Xt, accompanied by a reduction of the volume occupied by XIST RNA on the Xi (i). Progression of erosion leads to transcription of other X-linked genes (ii). Erosion is eventually complete following full destabilization of XIST and expansion of the volume occupied by XACT RNA (iii). In (A F), the numbers in the right bottom corner correspond to the percentage of nuclei displaying the staining. Blue shows DAPI staining. See also Figure S6. 544 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc.

21; Shen et al., 28), would ultimately occur. In this model, XACT would not be directly involved in the activity of the chromosome but would rather act as a XIST antagonist, competing with XIST for binding to the chromosome and/or limiting chromosome accessibility. Interestingly, in human pre-implantation embryos, XIST is expressed by default and accumulates around every X chromosome, in males and in females, in the absence of K3K27me3 enrichment and gene silencing (Okamoto et al., 211). It would be interesting to test whether XACT is expressed at this stage and whether it plays a role in preventing XIST from silencing X chromosomes in early human development. EXPERIMENTAL PROCEDURES Cell Culture The hesc lines H9, HUES1, and WIBR2 and ipscs were cultured in DMEM F-12 medium supplemented as described previously (Vallot et al., 213), except for WIBR2 cells, which were kept in 5% O 2. Fetal female fibroblast IMR9 cells and adult female fibroblasts (Coriell Institute, catalog no. AG963) were cultured in DMEM supplemented with 1% serum,.1 mm non-essential amino acids (NEAAs), 2 mm L-glutamine, and.5 mm b-mercaptoethanol. MSCs were cultured in minimum essential medium Eagle, a modification (a-mem) supplemented with 1% FBS,.1 mm NEAA, 2 mm L-glutamine, and 1 ng/ml b-fibroblast growth factor (FGF). The use of hesc lines was licensed by the French BioMedicine Agency in December 28 and in January 212. Immunofluorescence All cells were cultured on 12-mm coverslips and were first fixed in 3% paraformaldehyde/pbs for 1 min at room temperature (RT) and then permeabilized for 5 min at RT with cytoskeleton (CSK) buffer (NaCl, MgCl2, sucrose, and piperazine-n,n -bis(2-ethanesulfonic acid [ph 6.8]) supplemented with.5% Triton and 2 mm EGTA. Cells were incubated for 45 min with.2% gelatin/ PBS and for 45 min with rabbit polyclonal primary antibody (anti-h3k4me2, catalog no. 7-3, Millipore, or anti-h3k27me3, catalog no. 7-449, Millipore) diluted in.2% gelatin/pbs, washed three times in PBS, incubated for 3 min with an Alexa Fluor anti-rabbit secondary antibody (Life Technologies), and washed three times with PBS. Coverslips were mounted on slides with Vectashield (Vector Laboratories) supplemented with 4,6 -diamidino-2-phenylindole (DAPI). ChIP-Seq ChIP experiments were carried out as described previously (Navarro et al., 21) using antibodies against H3K9me3 (catalog no. pab-56-5, Diagenode) and H3K27me3 (catalog no. 7-449, Millipore). 1 ng of immunoprecipitated and input DNA was processed for sequencing (5-bp reads) on an Illumina Hiseq 2 using the Truseq DNA sample prep kit V2 (Illumina) according to the manufacturer s recommendations. As a validation of our ChIP-seq analysis workflow, our datasets for IMR9 were compared with published ChIP-seq profiles (GSE16256; Figure S2E). A summary of all publicly available datasets used in this study is provided in Table S1. RNA-Seq and Allelic Expression Analysis of X-Linked Transcripts Total RNA was extracted using Trizol and treated with Turbo DNase (Life Technologies). RNA-seq was performed and aligned as described previously (Vallot et al., 213). We took advantage of the clonal XCI pattern of H9 cells (Mitjavila- Garcia et al., 21; Shen et al., 28) to analyze the data in an allelic manner. We identified 198 SNPs on the X chromosome, corresponding to 78 genes that could be used for allelic analysis (Table S2). We considered a transcript as biallelic when at least 3% of reads originated from the second allele. For each selected SNP, we measured the significance of the difference in allelic balance between the XaXi and libraries using Fisher s exact test. RNA-FISH with Chromosome Paint Chromosome paint probes were obtained from Metasystems (chromosome 4 paint and X chromosome paint #1) and Cambio (X chromosome paint #2). X chromosome paint #1 was used in most experiments, except when indicated otherwise. For probe preparation, 5 ml of concentrated chromosome paint was diluted with 5 mg of human Cot-1 DNA (Life Technologies) in 3 ml water and precipitated using 1/1 3M NaAc and 2.5 volumes of ethanol overnight. The pellet was resuspended in 5 ml of 5% formamide/5% hybridization buffer (43 saline sodium citrate (SSC), 2% dextran sulfate, 2 mg/ml BSA, and 2 mm vanadyl ribonucleoside complex) and dissolved for 1 min at 37 C. Chromosome paint was denatured for 7 min at 75 C and incubated for 3 min at 37 C prior to dispensation on coverslips overnight at 37 C in a humid chamber. Coverslips were washed as described previously (Vallot et al., 213) and mounted in Vectashield (Vector Laboratories). For dual RNA-FISH involving chromosome paint, RNA-FISH was performed in two steps. First, RNA-FISH with chromosome paint was performed overnight as described above. Coverslips were washed three times in 5% formamide/23 SSC at 37 C and re-incubated overnight with the second probe as described in the Supplemental Experimental Procedures. Both signals were observed simultaneously. Successive RNA/DNA-FISH with Chromosome Paint RNA-FISH was performed on slides as described above, pictures were taken, and cell coordinates were noted using a stage-motorized Leica microscope (see Microscopy and Image Analysis). Slides were then washed, and cells were permeabilized for 1 min in.1 N HCl/.7% Triton on ice and treated with 1 mg/ml RNase in 23 SSC for 1 h at 37 C. Slides were denatured for 1 min at 8 C in 7% formamide/23 SSC. The X chromosome paint was denatured for 2 min at 75 C and incubated with slides overnight. Slides were washed three times with 23 SSC at 45 C and.13 SSC at 6 C. ACCESSION NUMBERS All datasets analyzed in this manuscript are referenced in Table S1. The GEO accession number for the ChIP-seq data produced in this manuscript is GSE62562. SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures, six figures, and two tables and can be found with this article online at http:// dx.doi.org/1.116/j.stem.215.3.16. AUTHOR CONTRIBUTIONS C.V. and J.F.O. performed the experiments. C.V. performed the bioinformatics analyses. M.M. and O.F. contributed to hesc manipulation. A.B. provided the H9-MSC-iPS system. J.C. and C.M. performed the NSC differentiation. J.P. helped with ChIP-seq. M.L. supervised the RNA-seq experiment. C.V., J.F.O., and C.R. conceived and designed the experiments and wrote the manuscript. ACKNOWLEDGMENTS We thank members of our laboratories for stimulating discussions and critical reading of the manuscript. We thank Claire Francastel for fruitful discussions about FISH experiments, Guillaume Filion for help with the use of jahmm, and the Genopole at Pasteur Institute for ChIP-seq sequencing. The research leading to these results has received funding from the European Research Council under the European Community s Seventh Framework Programme (FP7/27-213)/ERC Grant Agreement 26875 and EpiGeneSys FP7 25782 Network of Excellence, the Inserm (Avenir Program R721HS), and the Agence Nationale de la Recherche (ANC) Investissements Avenir, INGES- TEM Infrastructure. J.F.O. was supported by a fellowship from DIM-Stem Pole and the Canadian Institutes of Health Research. Received: October 22, 214 Revised: January 7, 215 Accepted: March 22, 215 Published: April 23, 215 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc. 545

REFERENCES Anguera, M.C., Sadreyev, R., Zhang, Z., Szanto, A., Payer, B., Sheridan, S.D., Kwok, S., Haggarty, S.J., Sur, M., Alvarez, J., et al. (212). Molecular signatures of human induced pluripotent stem cells highlight sex differences and cancer genes. Cell Stem Cell 11, 75 9. Bártová, E., Galiová, G., Krejcí, J., Harnicarová, A., Strasák, L., and Kozubek, S. (28). Epigenome and chromatin structure in human embryonic stem cells undergoing differentiation. Dev. Dyn. 237, 369 372. Boggs, B.A., Cheung, P., Heard, E., Spector, D.L., Chinault, A.C., and Allis, C.D. (22). Differentially methylated forms of histone H3 show unique association patterns with inactive human X chromosomes. Nat. Genet. 3, 73 76. Brown, C.J., and Willard, H.F. (1994). The human X-inactivation centre is not required for maintenance of X-chromosome inactivation. Nature 368, 154 156. Brunner, A.L., Johnson, D.S., Kim, S.W., Valouev, A., Reddy, T.E., Neff, N.F., Anton, E., Medina, C., Nguyen, L., Chiao, E., et al. (29). Distinct DNA methylation patterns characterize differentiated human embryonic stem cells and developing human fetal liver. Genome Res. 19, 144 156. Carrel, L., and Willard, H.F. (25). X-inactivation profile reveals extensive variability in X-linked gene expression in females. Nature 434, 4 44. Chadwick, B.P., and Willard, H.F. (24). Multiple spatially distinct types of facultative heterochromatin on the human inactive X chromosome. Proc. Natl. Acad. Sci. USA 11, 1745 17455. Csankovszki, G., Panning, B., Bates, B., Pehrson, J.R., and Jaenisch, R. (1999). Conditional deletion of Xist disrupts histone macroh2a localization but not maintenance of X inactivation. Nat. Genet. 22, 323 324. Deng, X., Hiatt, J.B., Nguyen, D.K., Ercan, S., Sturgill, D., Hillier, L.W., Schlesinger, F., Davis, C.A., Reinke, V.J., Gingeras, T.R., et al. (211). Evidence for compensatory upregulation of expressed X-linked genes in mammals, Caenorhabditis elegans and Drosophila melanogaster. Nat. Genet. 43, 1179 1185. Filion, G., and Cuscó, P. (214). jahmm: a tool for discretizing multiple ChIP-seq profiles. Published online May 21, 214. http://arxiv.org/pdf/145.5467.pdf. Gafni, O., Weinberger, L., Mansour, A.A., Manor, Y.S., Chomsky, E., Ben- Yosef, D., Kalma, Y., Viukov, S., Maza, I., Zviran, A., et al. (213). Derivation of novel human ground state naive pluripotent stem cells. Nature 54, 282 286. Giuliani, M., Oudrhiri, N., Noman, Z.M., Vernochet, A., Chouaib, S., Azzarone, B., Durrbach, A., and Bennaceur-Griscelli, A. (211). Human mesenchymal stem cells derived from induced pluripotent stem cells down-regulate NKcell cytolytic machinery. Blood 118, 3254 3262. Guenther, M.G., Frampton, G.M., Soldner, F., Hockemeyer, D., Mitalipova, M., Jaenisch, R., and Young, R.A. (21). Chromatin structure and gene expression programs of human embryonic and induced pluripotent stem cells. Cell Stem Cell 7, 249 257. Hall, L.L., Byron, M., Butler, J., Becker, K.A., Nelson, A., Amit, M., Itskovitz- Eldor, J., Stein, J., Stein, G., Ware, C., and Lawrence, J.B. (28). X-inactivation reveals epigenetic anomalies in most hesc but identifies sublines that initiate as expected. J. Cell. Physiol. 216, 445 452. Hawkins, R.D., Hon, G.C., Lee, L.K., Ngo, Q., Lister, R., Pelizzola, M., Edsall, L.E., Kuan, S., Luu, Y., Klugman, S., et al. (21). Distinct epigenomic landscapes of pluripotent and lineage-committed human cells. Cell Stem Cell 6, 479 491. Heard, E., Rougeulle, C., Arnaud, D., Avner, P., Allis, C.D., and Spector, D.L. (21). Methylation of histone H3 at Lys-9 is an early mark on the X chromosome during X inactivation. Cell 17, 727 738. Lengner, C.J., Gimelbrant, A.A., Erwin, J.A., Cheng, A.W., Guenther, M.G., Welstead, G.G., Alagappan, R., Frampton, G.M., Xu, P., Muffat, J., et al. (21). Derivation of pre-x inactivation human embryonic stem cells under physiological oxygen concentrations. Cell 141, 872 883. Mekhoubad, S., Bock, C., de Boer, A.S., Kiskinis, E., Meissner, A., and Eggan, K. (212). Erosion of dosage compensation impacts human ipsc disease modeling. Cell Stem Cell 1, 595 69. Mitjavila-Garcia, M.T., Bonnet, M.L., Yates, F., Haddad, R., Oudrhiri, N., Féraud, O., Magniez, A., Makhlouf, M., Vallot, C., Rougeulle, C., et al. (21). Partial reversal of the methylation pattern of the X-linked gene HUMARA during hematopoietic differentiation of human embryonic stem cells. J. Mol. Cell Biol. 2, 291 298. Navarro, P., Oldfield, A., Legoupi, J., Festuccia, N., Dubois, A., Attia, M., Schoorlemmer, J., Rougeulle, C., Chambers, I., and Avner, P. (21). Molecular coupling of Tsix regulation and pluripotency. Nature 468, 457 46. Nazor, K.L., Altun, G., Lynch, C., Tran, H., Harness, J.V., Slavin, I., Garitaonandia, I., Müller, F.J., Wang, Y.C., Boscolo, F.S., et al. (212). Recurrent variations in DNA methylation in human pluripotent stem cells and their differentiated derivatives. Cell Stem Cell 1, 62 634. Nozawa, R.S., Nagao, K., Igami, K.T., Shibata, S., Shirai, N., Nozaki, N., Sado, T., Kimura, H., and Obuse, C. (213). Human inactive X chromosome is compacted through a PRC2-independent SMCHD1-HBiX1 pathway. Nat. Struct. Mol. Biol. 2, 566 573. Okamoto, I., Patrat, C., Thépot, D., Peynot, N., Fauque, P., Daniel, N., Diabangouaya, P., Wolf, J.P., Renard, J.P., Duranthon, V., and Heard, E. (211). Eutherian mammals use diverse strategies to initiate X-chromosome inactivation during development. Nature 472, 37 374. Pontier, D.B., and Gribnau, J. (211). Xist regulation and function explored. Hum. Genet. 13, 223 236. Richardson, A.L., Wang, Z.C., De Nicolo, A., Lu, X., Brown, M., Miron, A., Liao, X., Iglehart, J.D., Livingston, D.M., and Ganesan, S. (26). X chromosomal abnormalities in basal-like human breast cancer. Cancer Cell 9, 121 132. Shen, Y., Matsuno, Y., Fouse, S.D., Rao, N., Root, S., Xu, R., Pellegrini, M., Riggs, A.D., and Fan, G. (28). X-inactivation in female human embryonic stem cells is in a nonrandom pattern and prone to epigenetic alterations. Proc. Natl. Acad. Sci. USA 15, 479 4714. Shipony, Z., Mukamel, Z., Cohen, N.M., Landan, G., Chomsky, E., Zeliger, S.R., Fried, Y.C., Ainbinder, E., Friedman, N., and Tanay, A. (214). Dynamic and static maintenance of epigenetic memory in pluripotent and somatic cells. Nature 513, 115 119. Silva, S.S., Rowntree, R.K., Mekhoubad, S., and Lee, J.T. (28). X-chromosome inactivation and epigenetic fluidity in human embryonic stem cells. Proc. Natl. Acad. Sci. USA 15, 482 4825. Spatz, A., Borg, C., and Feunteun, J. (24). X-chromosome genetics and human cancer. Nat. Rev. Cancer 4, 617 629. Tabar, V., and Studer, L. (214). Pluripotent stem cells in regenerative medicine: challenges and recent progress. Nat. Rev. Genet. 15, 82 92. Takahashi, K., and Yamanaka, S. (213). Induced pluripotent stem cells in medicine and biology. Development 14, 2457 2461. Vallot, C., and Rougeulle, C. (212). Epigenetic Stability of Human Pluripotent Cells. In Epigenomics: from Chromatin Biology to Therapeutics, K. Apasani, ed. (C.U. Press), pp. 118 133. Vallot, C., Huret, C., Lesecque, Y., Resch, A., Oudrhiri, N., Bennaceur-Griscelli, A., Duret, L., and Rougeulle, C. (213). XACT, a long noncoding transcript coating the active X chromosome in human pluripotent cells. Nat. Genet. 45, 239 241. Vincent-Salomon, A., Ganem-Elbaz, C., Manié, E., Raynal, V., Sastre-Garau, X., Stoppa-Lyonnet, D., Stern, M.H., and Heard, E. (27). X inactive-specific transcript RNA coating and genetic instability of the X chromosome in BRCA1 breast tumors. Cancer Res. 67, 5134 514. Vogel, M.J., Guelen, L., de Wit, E., Peric-Hupkes, D., Lodén, M., Talhout, W., Feenstra, M., Abbas, B., Classen, A.K., and van Steensel, B. (26). Human heterochromatin proteins form large domains containing KRAB-ZNF genes. Genome Res. 16, 1493 154. Wutz, A. (211). Gene silencing in X-chromosome inactivation: advances in understanding facultative heterochromatin formation. Nat. Rev. Genet. 12, 542 553. Wutz, A., and Jaenisch, R. (2). A shift from reversible to irreversible X inactivation is triggered during ES cell differentiation. Mol. Cell 5, 695 75. Zhu, Z., and Huangfu, D. (213). Human pluripotent stem cells: an emerging model in developmental biology. Development 14, 75 717. 546 Cell Stem Cell 16, 533 546, May 7, 215 ª215 Elsevier Inc.

Cell Stem Cell Supplemental Information Erosion of X Chromosome Inactivation in Human Pluripotent Cells Initiates with XACT Coating and Depends on a Specific Heterochromatin Landscape Céline Vallot, Jean-François Ouimette, Mélanie Makhlouf, Olivier Féraud, Julien Pontis, Julien Côme, Cécile Martinat, Annelise Bennaceur-Griscelli, Marc Lalande, and Claire Rougeulle

Supplemental*Information* Erosion' of' X' chromosome' inactivation' in' human' pluripotent' cells' initiates' with'xact'coating'and'depends'on'a'specific'heterochromatin'landscape*! Céline!Vallot 1,2,!6,!Jean1François!Ouimette 1,2,6,!Mélanie!Makhlouf 1,!2,!Olivier!Féraud 3,!!Julien!Pontis 1,2,! Julien!Côme 4,!Cécile!Martinat 4,!Annelise!Bennaceur1Griscelli 3,!Marc!Lalande 5!and!Claire!Rougeulle* 1,2! * Contents:* I.*Supplemental*Figures*and*Tables* II.*Supplemental*Experimental*Procedures* III.*Supplemental*References* *

! I.*Supplemental*Figures*and*Tables*! 2

A 72,75 NAP2K4P1 73 XIST 73,25 (Mb) D 4 H3K27me3 r=.35 p<1-6 XaXi H9 4 4 CDX4 H3K27me3 H3K9me3 CHIC1 TSIX JPX log2 IP/INPUT XaXi H9 2-2 -2 2 4 B chrx chr22 chr21 chr2 chr19 chr18 chr17 chr16 chr15 chr14 chr13 chr12 chr11 chr1 chr9 chr8 chr7 chr6 chr5 chr4 chr3 chr2 chr1 C H3K27me3 XaXi Colonic Mucosa 1 2 3 4 % occupied H3K9me3 XaY Lung 1 2 3 % occupied XaY Duodenum Mucosa 4 1 2 3 4 % occupied XaY Duodenum Muscle 1 2 3 % occupied 4 XaY Pancreas 1 2 3 % occupied 4 E log2 IP/INPUT XaXi H9 F 4 2-2 -2 log2 IP/INPUT IMR9 H3K9me3 2 log2 IP/INPUT IMR9 H3K27me3 r=.44 p<1-6 4 XaXi Colonic Mucosa XaY Lung XaY Duodenum Mucosa XaY Duodenum Muscle XaY Pancreas chrx chr22 chr21 chr2 chr19 chr18 chr17 chr16 chr15 chr14 chr13 chr12 chr11 chr1 chr9 chr8 chr7 chr6 chr5 chr4 chr3 chr2 chr1 G 2.7 H3K9me3 23.66 1 2 3 4 % occupied 1 2 3 % occupied 4 1 2 3 % occupied 4 1 2 3 % occupied Figure S1: Epigenomic analyses (Related to Figure 1). (A) Zoom on the XIST neighboring region, with genomic coordinates indicated (in Mb, hg19) and genes represented as black boxes. H3K27me3 (green) and H3K9me3 (blue) ChIP-seq data in XaXi H9 cells is represented as log2 enrichment ratio of IP over input. (B, C) H3K27me3 and H3K9me3 occupancy of the active X chromosome and autosomes in one additional female (Colonic Mucosa cells) and four male differentiated cell types (AG445 male lung cells, Duodenum Mucosa cells, Duodenum Muscle cells and Pancreas cells). ChIPseq were downsampled to the smallest sample for each category. (D, E) Scatter plots comparing H3K27me3 (D) and H3K9me3 (E) log2 enrichment over input for 1kb bins on the X chromosome between XaXi pluripotent (H9) and differentiated cells (IMR9). (F, G) Euler diagrams displaying Mb of X chromosome covered by H3K27me3 peaks (F) or H3K9me3 peaks (G) in XaXi H9 cells versus IMR9 cells. 3" 4 1 2 3 % occupied 4

A log2 IP/input 4 2 XaXi HUES6 H3K27me3 H3K9me3 r= -.56 p<1-6 log2 IP/input 4 2 XaXi HUES48 H3K27me3 H3K9me3 r= -.64 p<1-6 B log2 IP/input -2 4 2 XaXi Adult adipocyte H3K27me3 H3K9me3 5 1 15 chrx (Mb) r=.2 p=.12 log2 IP/input -2 4 2 XaXi Adult colonic mucosa H3K27me3 H3K9me3 5 1 15 chrx (Mb) r= -.1 p<1-6 C -2 4 XaXi vhmec H3K27me3 H3K9me3 5 1 15 chrx (Mb) r= -.47 p<1-6 -2 4 XaXi RPE-hTERT H3K27me3 H3K9me3 5 1 15 chrx (Mb) r= -.74 p<1-6 log2 IP/input 2 log2 IP/input 2 D log2 IP/input -2 4 2-2 WIBR3 H3K27me3 H3K9me3 5 chrx 1 15 (Mb) r=.6 p=.1 E IMR9 (Ren Lab) Upstate 7-449 -2 2-2 H3K27me3-2 5 1 15 chrx (Mb) r=.87 p<1-6 2 IMR9 (Ren Lab) Abcam ab8898 2-2 H3K9me3-2 r=.69 p<1-6 2 5 1 15 chrx (Mb) IMR9 (Vallot, Ouimette et al.) Upstate 7-449 IMR9 (Vallot, Ouimette et al.) Diagenode pab-56-5 Figure S2: H3K27me3 and H3K9me3 profiles in hescs and differentiated cells (Related to Figure 1 and 2). (A-D) Superposition of H3K9me3 and H3K27me3 ChIP-seq for XaXi hesc (A), differentiated normal cells (B), immortalized cells (C) and WIBR3 hesc. Log2 ratios were compared using Pearson correlation test and the associated p-value was calculated using random permutations of the data sets. (E) Scatter plots of log2 enrichment over input for 1kb bins along the X chromosome for H3K27me3 and H3K9me3 comparing our IMR9 ChIPseq with datasets from the UCSD Human Reference Epigenome Mapping Project (GSE16256). For H3K27me3 ChIPseq (left panel), the same antibody was used in both experiments, and the two datasets display high correlation (r=.87, p<1-6). For the H3K9me3 ChIPseq (right panel), we used a different antibody from the one previously used, which could explain the variations observed bewteen the two datasets. 4"

A 1% TSPAN6 monoallelic 1% POLA1 reactivated 1% CTPS2 escapee 8% 8% 8% 6% 4% 2% 6% 4% 2% 6% 4% 2% Allele B Allele A % gdna XaXi H9 H9 % gdna XaXi H9 H9 % gdna XaXi H9 H9 B lnc13 12,98 13 13,2 (Mb) C lnc121 121,34 121,36 121,38 121,4 (Mb) 1 3 3,25 SNP1 SNP3 ** SNP2 * H9 RNAseq (+ strand) H1 RNAseq (+ strand) K4me3 Raw Signal K36me3 Raw Signal 5 1 15 1 SNP1 SNP3 * * SNP2 * * SNP4 1% p<1-8 p=.2 p<1-6 p=.9 p<1-4 1% p<1-3 p=.2 % reads 75% 5% 25% Allele A Allele B % reads 75% 5% 25% Allele A Allele B % XaXi XaXi XaXi SNP1 SNP2 SNP3 % XaXi XaXi XaXi XaXi SNP1 SNP2 SNP3 SNP4 XaXi H9 lnc13 H9 lnc13 XaXi H9 lnc121 H9 lnc121 9% 71% 86% 76% Figure S3: Validation of allelic RNA-seq (Related to Figure 3). (A) Assessment of allelic expression by pyrosequencing for TSPAN6, POLA1 and CTPS2. (B, C) Identification of two lncrnas relaxed from XCI in cells, lnc13 and lnc121. Upper panel: we compared our RNA-seq data ( H9) to published strand-specific RNA-seq from H1 hesc (UCSC accession number: wgencodeeh132). Both show transcription originating from the plus strand over two large intergenic regions. ChIP-seq data for H1 hesc from the ENCODE project shows large H3K36me3 peaks, usually found within the body of transcribed genes, over the identified transcription unit. Furthermore, one large H3K4me3 peak is located at the 5' end of each region. We used the H3K36me3 peak calling to delineate the two lncrna genes: lnc13 (12,995,14-13,55,964) and lnc121 (121,352,438-121,385,472). Middle panel: Allelic counts assessed by RNA-seq in XaXi and H9 cells were compared using a Fisher s exact test (corresponding p-values are indicated above each SNP). Lower panel: reactivation observed by RNA-seq was confirmed by RNA-FISH using fosmids spanning each lncrnas in XaXi and H9 cells. Predominant expression pattern is displayed for XaXi and H9 cells, numbers of nuclei with mono or biallelic pattern were compared using a Fisher s exact test. White scale bars represent 5 µm. 5"

1µg/µl Cot-1 X-paint RNA B 2µg/µl Cot-1 X-paint RNA X-paint RNA fibroblast µg Cot-1 X-paint RNA XaXi H9 A 1% log2 IP/input C chrx 4 75 76 77 ATRX MAGT1 LOC11928469 4 D 1% X-paint RNA #2 mono-allelic bi-allelic 4% 2% % X-paint RNA XaXi X-paint RNA Xe Xa H9 Xe 6% G HUES1 Xa 8% 1% F E H9 ATRX H3K27me3 H9 XaXi H9 H3K9me3 H9 XaXi H9 1% H9 % of nuclei XaXi H9 ATRX Mb X-paint DNA Xe Xe Xa 81% Xa Figure S4: X chromosome paint RNA-FISH to monitor XCI erosion (Related to Figure 4). (A) 5µl of X-paint (yellow) were co-precipitated with various amounts of human Cot-1 DNA prior to overnight hybridization (see Experimental Procedures). White scale bars represent 5µm. (B) X-paint (yellow) in adult female fibroblasts (AG963). White scale bars represent 5µm. (C) ChIP-seq data of the ATRX locus for H3K27me3 and H3K9me3 in XaXi and H9 cells. The charts shows the log2 enrichment signal of IP over input over 1kb binning windows. (D) Allelic expression analyzed by RNA-FISH using a BAC probe covering ATRX (green) in XaXi and H9 cells. White scale bars represent 5µm. The barchart displays the percentage of nuclei with mono and bi-allelic pattern of expression (n=1). (E) Validation of RNA FISH chromosome paint approach using an independent X-paint (yellow; Cambio). (F) X-paint (yellow) in HUES1. (G) X-paint RNA-FISH (yellow) followed by X-paint DNA-FISH (green) in H9 cells. White scale bars represent 5µm. 6"

A C % peak height HUES1 NSC 1% 8% 6% 4% 2% NESTIN GPM6B 9% 1% 8% 6% 4% 2% SOX2 94% lnc13 B Allele A Allele B D HUES1 d NSC X-paint RNA X-paint RNA 81% 89% lnc121 gdna cdna H9 H9 H9 NSC E % gdna H9 POLA1 H9 NSC % gdna H9 COL4A6 H9 NSC G/A G/A A 1% 1% % peak height 8% 6% 4% 2% 8% 6% 4% 2% Allele A Allele B F Expression relative to GAPDH % gdna,6,4,2 H9 OCT3/4 H9 NSC % Expression relative to GAPDH gdna,4 4,3 3,2 2,1 1 H9 NANOG d d5 d d5 H9 NSC G H9 d X-paint RNA 65% d5 X-paint RNA 9% Figure S5: Additional monitoring of XCI erosion during differentiation (Related to Figure 5). (A) Characterization of NSCs derived from HUES1 by immunofluorescence using NESTIN (green) and SOX2 (red) antibodies. The numbers in white indicate the percentage of positive cells as quantified using ArrayScan. White scale bars correspond to 1µm. (B) X-paint RNA-FISH as in Figure 5. (C) Analysis of allelic expression by pyrosequencing of genes relaxed from XCI in H9: GPM6B, lnc13, For each SNP, H9 genomic DNA (gdna) and H9 and NSC cdna were studied. The bar chart indicates the percentage of the peak height corresponding to each allele. (D) Allelic expression analysis of lnc121 during differentiation using sequencing. (E) Same as in (C) on POLA1 and COL4A6 genes. (F, G) Characterization of H9 hesc during undirected differentiation using RT-qPCR (F) and X-paint RNA-FISH (G). White scale bars correspond to 5µm. 7

p=.4 A % nuclei 1% 8% 6% 4% 2% % wibr2 p29 wibr2 p38 WIBR2 5%O2 B H3K27me3 p33 XACT merge 1% p33 42% p33 42% XaXi H9 p.36 C XIST XACT merge 1% 1% 1% Figure S6: XACT coating precedes the loss of XIST coating and H3K27me3 accumulation on the Xi (Related to Figure 6). (A) Quantification corresponding to experiment in Figure 6C, where XACT is in red and XIST in green. The barplot displays the percentage of nuclei (n>1) with corresponding expression pattern for passage 29 and 38. Number of nuclei were compared using a Fisher s exact test. (B) Immunofluorescence coupled to RNA-FISH to simultaneously detect H3K27me3 (green) and XACT (red) in WIBR2 nuclei. (C) RNA-FISH for XIST (green) and XACT (red) in H9 cells shows co-accumulation of XIST and XACT on one X chromosome in 1% of the cells. 8"