Xist RNA repeat E is essential for ASH2L recruitment to the inactive X and regulates histone modifications and escape gene expression

Long non-coding RNA Xist plays a crucial role in establishing and maintaining X-chromosome inactivation (XCI) which is a paradigm of long non-coding RNA-mediated gene regulation. Xist has Xist-specific repeat elements A-F which are conserved among eutherian mammals, underscoring their functional importance. Here we report that Xist RNA repeat E, a conserved Xist repeat element in the Xist exon 7, interacts with ASH2L and contributes to maintenance of escape gene expression level on the inactive X-chromosome (Xi) during XCI. The Xist repeat E-deletion mutant female ES cells show the depletion of ASH2L from the Xi upon differentiation. Furthermore, a subset of escape genes exhibits unexpectedly higher expression in the repeat E mutant cells than the cells expressing wildtype Xist during X-inactivation, whereas the silencing of X-linked non-escape genes is not affected. We discuss the implications of these results to understand the role of ASH2L and Xist repeat E for histone modifications and escape gene regulation during random X-chromosome inactivation.


Introduction
To equalize the imbalance of X-linked gene dosage between genders, one of the two X chromosomes in female mammals is converted to the transcriptionally inactive X-chromosome (Xi) at the early phase of development whereby the majority of X-linked genes on the Xi are silenced [1][2][3]. In mice, imprinted X-chromosome inactivation (XCI) starts from as early as 2-4-cell stage in which the paternal X-chromosome is always silenced, and the paternally imprinted Xi is sustained in the cells of the trophectoderm and primitive endoderm, which give rise to the extra-embryonic tissues [4,5]. In contrast, imprinted XCI is reactivate in the epiblast of the inner cell mass, and then induces random XCI during the peri-implantation stage. The paternal and maternal X chromosomes have an equal chance of inactivation in random XCI.
Once random XCI is established in epiblast lineage, the Xi is inherited throughout successive cell divisions to continue balancing X-linked gene dosage between males and females. While the majority of X-linked genes are subjected to silencing by XCI, a subset of genes called escape genes (approximately 3% in mice) is partially transcribed on the Xi [6,7]. Transcriptionally active compartments for escape genes are marked by characteristic epigenetic modifications which are distinct from heterochromatic regions across the Xi. Some escape genes are commonly expressed on the Xi in variable cell lines and tissues, whereas expression of other escape genes occurs in a cell line-or tissue-specific manner. Although several models have been proposed to explain how escape genes evade XCI under the chromosome-wide heterochromatic environment on the Xi [6,8], the underlying mechanism for escaping XCI remains unknown.
The X inactivation center (Xic) required for XCI harbors a number of non-protein-coding (non-coding) genes such as Xist and its antisense Tsix [9][10][11]. The expression of Xist is upregulated at the onset of random XCI by the coordinated action of several non-coding genes in the Xic. Xist RNA plays the pivotal role to initiate XCI and induce chromosome-wide X-linked gene silencing across the Xi except for escape genes [12]. Xist encodes a~17 kb-length noncoding RNA and is comprised of seven exons in mice [13]. In undifferentiated mouse embryonic stem (ES) cells, Xist is transcribed at a very low level from both paternal and maternal X chromosomes. When random XCI begins upon differentiation, the expression of Xist is increased in the future Xi; meanwhile, Xist expression is gradually erased from the other X [14]. The abundant Xist RNA covers the future Xi in cis and functions as a scaffold for various chromatin modifiers such as polycomb repressive complex 2 (PRC2) and PRC1 for histone 3 lysine 27 trimethylation (H3K27me3) and histone 2A lysine119 ubiquitination (uH2A), respectively, to decorate the Xi by a series of repressive epigenetic modifications [15]. Among various chromatin modifying enzymes recruited to the Xi in a Xist RNA-dependent manner, ASH2L is unique because it is known as a component of the MLL/SET histone 3 lysine 4 (H3K4) methyltransferase complexes for transcriptional activation [16,17]. To date, the function of ASH2L on the Xi remains mysterious.
Xist evolved from a protein-coding gene Lnx3 in early eutherians by pseudogenization, including the integration of mobile elements which yield simple tandem repeats within Xist (denoted as repeat A-F) [18,19]. These Xist repeat elements play various important roles to fulfill the unique function of Xist during XCI [9,10,20]. Xist repeat A is required for induction of X-linked gene silencing and Xist RNA spreading across the Xi [21,22]. A recent study identified the repeat A binding proteins as potential candidates responsible for induction of the Xist repeat A-dependent X-linked gene silencing [23]. The region across the Xist repeat F and B has an important role to recruit PRC2 to the Xi [24]. Interference using peptide nucleic acid (PNA) or locked nucleic acid (LNA) complementary to Xist repeat C revealed that the repeat C is important for Xist RNA localization [25,26]. In contrast to the intensely-studied Xist exon 1 which contains the repeat A-D and F, exon 7 of Xist has been overlooked even though it is the second largest exon with the repeat E element highly conserved among the eutherian mammals [27]. Although Xist exon 7 interacts with hnRNP U and is required for the stable Xist RNA localization to the Xi [28], the role of the Xist repeat E localized at the 5 0 -end of Xist exon 7 has not yet been addressed. Here, we explore the link between ASH2L and Xist RNA and demonstrate the role of the Xist repeat E for ASH2L recruitment to the Xi, regulation of histone modifications and escape gene expression.

ASH2L preferentially binds to the repeat E of Xist RNA
To explore the interaction between ASH2L and Xist RNA, we carried out formaldehyde-crosslinking RNA immunoprecipitations (RIPs) (Fig 1 and S1 Fig). Tsix-mutant female ES cells in which non-random XCI is induced in the Tsix-mutant X-chromosome were differentiated, and cell lysate for the RIP was prepared at day 8 upon differentiation [29]. While the control RIP assay using normal rabbit IgG did not show significant immunoprecipitation of the Xist RNA by any primer pairs for quantitative RT-PCR (RT-qPCR) across Xist, Xist RNA was immunoprecipitated by anti-ASH2L antibody. The ASH2L binding to Xist RNA was especially evident at the primer pair 11, which amplifies the internal region of the repeat E. RT-qPCR using primer pairs around the repeat E (positions 9 and 13) showed modest peaks, suggesting that ASH2L preferentially binds to the repeat E region of Xist RNA ( Fig 1B). Interestingly, UV-crosslinking RIP to address the direct protein-RNA interaction did not show significant precipitation of Xist RNA at any positions even with the anti-ASH2L antibody ( Fig 1C). These data suggest that ASH2L could bind to the repeat E region of Xist RNA indirectly via another Xist RNA-associated protein(s). Alternatively, Xist RNA-ASH2L interaction might be too weak or transient to be detected by UV-crosslinked RIP due to its inefficient crosslinking.
To test whether the repeat E of Xist RNA is essential for ASH2L recruitment to the Xi, we used a tetracycline (tet)-inducible Xist transgene system with the Xist mutant transgenes lacking a BssHII-Bsu36I 2.6-kb region including repeat E [21]. The Xist-inducible male ES cell lines carry Xist cDNA transgene under the control of a tet-inducible promoter in the Hprt locus. Robust full-length Xist (Xist-FL) expression induced by adding doxycycline (Dox) in the culture medium resulted in cis-accumulation of Xist RNA called Xist RNA clouds; these Xist RNA clouds were co-localized with strong focal ASH2L signal as well as repressive epigenetic modifications, H3K27me3 and uH2A (Fig 1D-1F). On the other hand, induction of Xist RNA lacking the repeat E (Xist-BB) failed to recruit ASH2L, although localization of Xist RNA, H3K27me3 and uH2A to the Xi was not altered (Fig 1D-1F). These results suggest that 2.6-kb region containing repeat E is crucial for ASH2L recruitment to the Xi and that Xist repeat E and ASH2L are not required for the cis-accumulation of Xist RNA, H3K27me3 and uH2A on the Xi.

Xist repeat E deletion induces slightly increased Xist expression
To further examine the role of Xist repeat E for ASH2L recruitment to the Xi and random XCI, we created Xist repeat E deletion mutant (XistΔE) female ES cells using standard gene targeting (Fig 2A and S2 Fig). It is likely that deletion of a specific region of Xist RNA which interacts with ASH2L could disrupt ASH2L function on the Xi but not affect global transcriptional activation by the MLL/SET complexes. To delete the 1.3 kb repeat E region at the 5 0 -end of Xist exon 7, we replaced the repeat E by a hygromycin resistant gene (Hyg) cassette flanked with two FRT derivative F5 sites (Fig 2A and S2A Fig) [30]. In this cell line, approximately 70-80% of the Xi is derived from the 129 allele due to the effect of an allele-specific X-controlling element (Xce) which controls probability of random XCI [31]. Since homology arms in the targeting construct were created from the phage P1 clone [32], which has the 129 genomic fragment, it was expected that targeted deletion of the repeat E element occurred on the 129 allele in the 16.7 female ES cells. To address whether the targeting of Xist repeat E occurs on the 129 or Cast allele, we carried out allele-specific RT-qPCR analysis using a primer pair to amplify an internal part of Xist repeat E. Since 129 allele-specific primer pairs could not amplify the fragment from both cDNA and genomic DNA of the mutant ES cells (Fig 2B), we concluded that the Xist repeat E deletion occurred on the 129 allele.
To address whether the XistΔE mutation affects Xist expression during random XCI, random XCI was induced upon differentiation by depletion of LIF and 2i inhibitors from the culture media. Using an allele-specific primer pair across Xist exon 1-3, we performed RT-qPCR analysis for Xist using the differentiating XistΔE cells (Fig 2C). The allele-specific RT-qPCR analysis for Xist showed that Xist expression was gradually increased from both 129 and Cast alleles upon induction of XCI in both wildtype and the XistΔE mutant cells. Interestingly, we found that the expression of Xist changed in three independent XistΔE mutant ES clones compared with that of wildtype cells. At the undifferentiated stage, all mutant cells showed a higher Xist expression level compared with wildtype. In the differentiating stages, all XistΔE mutant clones showed a higher Xist expression level than wildtype, although the extent of Xist upregulation varied between the mutant ES clones. These data indicate that repeat E might affect Xist expression.
To clarify the role of Xist repeat E for ASH2L recruitment and X-linked gene expression on the Xi during random XCI, we introduced an additional mutation in Tsix, an antagonist for Xist, to the XistΔE mutant 129 X-chromosome, which leads to non-random XCI of the XistΔE/ Tsix-mutant 129 X-chromosome ( Fig 2D and S3A Fig) [30,33]. As a control, we used the Tsix mutant female ES cells. To address whether the Tsix expression is disrupted by the targeted truncation (Tsix truncation mutation, TST), we performed allele-specific RT-qPCR analysis for Tsix and confirmed that the expression level of Tsix was significantly reduced in all TST and TST/XistΔE mutant cells compared with their parental cell lines ( Fig 2E). Next, to examine whether the XistΔE mutation affects XCI, we induced XCI upon ex vivo differentiation using the TST and TST/XistΔE mutant cell lines. The growth of the TST/XistΔE cells upon differentiation was normal and could not be distinguished from the control TST cells (S3B Fig).
To confirm non-random XCI induced by the Tsix mutation, we performed the RT-qPCR analysis for Xist and confirmed non-random expression of Xist from the 129 allele ( Fig 2F). As The positions of the primer pairs across Xist are shown as arrowheads. (B) Formaldehyde-crosslinked Xist RIP using anti-ASH2L antibody. A relative amount of Xist RNA immunoprecipitated by anti-ASH2L antibody to input was quantified by RT-qPCR. The means ± standard deviation (SD) from three independent experiments are shown. (C) UV-crosslinked Xist RIP using anti-ASH2L antibody. The mean ± SD from three independent experiments is shown. (D) Schematics of tet-inducible full-length and mutant Xist cDNAs used in this study. (E) ImmunoFISH and immunostaining of the tet-inducible Xist T20 ES cells at day 2 in the presence of Dox. Nuclei were counterstained by DAPI. Scale bar is 5 μm. (F) Summary of the tet-inducible mutant Xist cDNA experiment. Colocalization percentage was calculated by immunoFISH and immunostaining experiments in Fig 1E. More than 100 nuclei for each transgenic ES cell line in more than two independent experiments were counted. https://doi.org/10.1371/journal.pgen.1006890.g001 Role of Xist repeat E in X-inactivation To clarify whether the slightly higher Xist upregulation in the XistΔE mutant cell lines is a result of altered Xist RNA stability, we carried out the half-life assay for Xist transcript ( Fig  2G). After pulse-labeling of nascent transcripts by 5-ethynyl uridine (EU) for 24 hours, the EU-labeled cells cultured in normal medium were retrieved every 40 minutes. EU-labeled RNA clicked by biotin was captured by streptavidin beads and compared with the INPUT RNA by RT-qPCR analysis using a Xist RNA-specific primer pair. The half-life of Xist RNA was approximately 80 minutes and did not significantly differ between the TST and TST/ XistΔE mutant cells. This suggested that the upregulation of Xist expression observed in the XistΔE mutant cells might be a result of increased transcription, not alteration of its stability.
The XistΔE mutation abrogates ASH2L recruitment to the Xi but does not affect the deposition of repressive histone modifications During XCI, Xist RNAs cover the entire Xi and recruit various chromatin modifying enzymes to the Xi to induce chromosome-wide dynamic epigenetic alteration. To determine whether the XistΔE mutation affects Xist RNA localization and H3K27me3 deposition on the Xi, we performed the Xist RNA florescence in situ hybridization (FISH) combined with immunofluorescence with an anti-H3K27me3 antibody (Fig 3A and 3B). Neither Xist RNA clouds nor H3K27me3 strong foci could be seen in undifferentiated ES cells. As differentiation progressed, the co-localization of Xist RNA and H3K27me3 was gradually increased in both control TST and XistΔE TST cells. Interestingly, we observed dispersed Xist RNA localization in a subset of the XistΔE TST cells although H3K27me3 signal was focal and colocalized with the Xist RNA clouds.
We also examined the localization of other repressive epigenetic histone modifications, uH2A and H4K20me1, which are recruited to the Xi upon induction of random XCI [34][35][36]. Focal staining of these repressive epigenetic marks was colocalized with H3K27me3 and was not altered by the repeat E deletion (S4 Fig). Although the deposition of H3K27me3 and uH2A on the Xi was not affected by Xist repeat E deletion, we also examined the recruitment of EZH2 and RING1B, the catalytic core subunits of PRC2 and PRC1 complex, respectively, by immunoFISH (S5 Fig). At ES cell stage (day 0), neither Xist RNA cloud nor EZH2/RING1B focal signal was observed in both TST and XistΔE TST cells. As differentiation progressed, map. SA, Splice acceptor; IRES, internal ribosomal entry site; Hyg, hygromycin resistance gene; tpA, tandem polyadenylation signals. The positions of primer pairs used for allele-specific Xist expression analysis (B and C) are shown as asterisks. (B) Allele-specific RT-PCR and genomic PCR analysis for Xist using the primer pair (X7E) which amplifies the 3 0 -end of the Xist repeat E in the XistΔrepE mutant cell lines. (C) Allele-specific RT-qPCR of the Xist expression across exon 1-3 (X1-3) in the XistΔrepE mutant. Gapdh was used as an internal control for normalization. Each value was also normalized to that of the undifferentiated wildtype 16.7 cells which is set to 1. The mean ± SD from three independent experiments is shown. P-values were calculated by an unpaired t-test (*p<0.05, **p<0.01).  EZH2 and RING1B showed the highest colocalization with Xist RNA cloud at day 4. The percentage of their colocalization gradually decreased at day 8 and 12 in both TST and XistΔE TST cells as reported previously [35][36][37]. These results demonstrated that the Xist repeat E is not required for the recruitment of some repressive epigenetic histone modifications to the Xi. We also performed immunostaining analysis using anti-H3K27me3 and ASH2L antibodies and confirmed that the deletion of the repeat E resulted in the depletion of ASH2L from the Xi. Since the other common components of the MLL/SET complexes, WDR5 and RBBP5, were not enriched on the Xi upon the induction of random XCI (S6 Fig), function of ASH2L -which is recruited to the entire region of the Xi by Xist RNA-would be independent of the MLL/SET complexes' activity. Taken together, these data suggest that Xist repeat E is essential for the ASH2L localization but not for the deposition of repressive epigenetic modifications on the Xi.

Escape gene expression is compromised in the XistΔE mutant cells
Since the XistΔE mutation resulted in the slight upregulation of Xist expression and dispersed Xist RNA in a subset of differentiating cells, we next addressed whether the expression levels of X-linked genes are affected during XCI. First, we examined the expression pattern of four X-linked non-escape genes using the allele-specific RT-qPCR analysis (Fig 4A and 4B). Upon differentiation, these X-linked genes on the 129 Xi were gradually silenced and efficiently repressed less than 10% (at day 12) of that in undifferentiated cells (at day 0). Similar to the observation of in vivo X-linked gene silencing during imprinted XCI [38], each X-linked nonescape gene exhibited a diverse kinetics of gene silencing upon differentiation. Whereas Hprt and Usp9x were silenced by day 4 upon differentiation, Pgk1 silencing started after day 4 of differentiation. While Hprt silencing was slightly slower in two of three XistΔE/TST mutant clones at day 4 upon differentiation, all four non-escape genes were efficiently silenced as differentiation progressed. Combined with immunostaining data in Fig 3, these results indicate that the repeat E deletion does not affect the induction of X-linked genes silencing on the Xi despite of dispersed Xist RNA clouds in a subset of the XistΔE/TST cells.
The MLL/SET complexes generally induce open chromatin formation and transcriptional activation by the deposition of methylation at H3K4 [17]; thus, we next examined whether depletion of ASH2L, a component of the MLL/SET complexes, from the Xi by the Xist repeat E deletion affects the expression of escape genes. Due to fewer single-nucleotide polymorphisms (SNPs) in the exon regions than in the intron and intergenic regions, a limited number of escape genes have been identified by various methods, including allele-specific RNAseq in mice, to date [8]. To further examine the expression ratio of escape genes between the Xa and Xi, we picked 8 X-linked genes (1 non-escape and 6 escape genes included) and addressed the impact of the Xist repeat E deletion on the escape gene expression using pyrosequencing analysis ( Fig 4C). We confirmed dominant Xist expression and Pgk1 silencing on the 129 mutant X as a result of non-random XCI of the 129 X due to the Tsix mutation, which is consistent with the data in Figs 2F and 4B. In undifferentiated cells, escape gene expression in the XistΔE/TST mutation cells was comparable to that in the control TST cells. Upon differentiation, however, we observed the stochastic upregulation in a subset of escape genes on the Xi. To our surprise, three out of six escape genes' expression from Xi in the XistΔE mutation cells exhibited higher  Location of non-escape (left) and escape genes (right) on the X-chromosome we tested in Fig 4B-4D. Astarisk than control cells. Utx and Ftx expression was most affected by the Xist repeat E deletion, and the upregulation on the Xi increased to 63% compared to control cells expressing wildtype Xist. To further address X-linked gene expression, we performed allele-specific RNAseq analysis using 15 X-linked gene-specific primer sets (Fig 4D). Similar to pyrosequencing data, a subset of escape genes (3 out of 8 escape genes tested) was upregulated in some clones while expression of all 5 non-escape genes was stably silenced in the XistΔE/TST cells. Cdk16/Pctk1 and Usp9x, known escape genes in the brain, showed slightly higher expression than typical non-escape genes (Fig 4D) [39]; thus, they might be escape genes in female ES cells upon differentiation. Since escape gene expression ratio between the Xa and Xi was not altered between the TST and XistΔE/TST mutation cells in undifferentiated cells, deletion of Xist repeat E results in compromised escape gene expression in a differentiation-specific manner.
To explore a potential link between epigenetic alteration and higher escape gene expression induced by ASH2L depletion from the Xi, we performed allele-specific chromation immunoprecipatition (ChIP) using anti-H3K4me3 and -H3K27me3 antibodies ( Fig 5). As previously reported [40], H3K4me3 levels at X-linked gene promoter in the control TST cells (Fig 5A, white columns) were relatively comparable between the Cast Xa and 129 Xi even at day 12 upon differentiation, indicating that H3K4me3 might not be a major epigenetic regulator for X-linked gene silencing at the onset of random XCI. In contrast, H3K27me3 levels at the 129 Xi promoters in the control TST cells were higher than at the Cast Xa upon differentiation due to PRC2 recruitment to the Xi. Surprisingly, H3K4me3 levels at the promoter region of Xlinked genes were significantly reduced on both Xa and Xi in the XistΔE/TST cells especially upon differentiation. XistΔE/TST ES cells (day 0) exhibited largely similar level of H3K4me3 at X-linked gene promoters, suggesting Xist RNA-dependent ASH2L on the Xi might regulate a chromosome-wide H3K4me3 distribution. Since H3K4me3 levels at escape gene promoters in the XistΔE/TST cells at day 12 upon differentiation were comparable between the Xa and Xi (Fig 5A), H3K4me3 is unlikely to cause upregulation of escape gene on the Xi in the XistΔE/ TST cells upon differentiation. Interestingly, H3K4me3 at autosomal gene promoters was also affected by the Xist repeat E deletion (Fig 5B), suggesting that Xist repeat E might regulate global H3K4me3 distribution in female cells. Since H3K27me3 demethylase Utx expression was one of the escape genes which exhibited upregulated expression in the XistΔrepE cells, we next addressed whether reduction of H3K27me3 level at the promoter region is associated with abnormal upregulation of a subset of escape gene in the XistΔrepE/TST cells by allele-specific ChIP. Interestingly, the H3K27me3 modification at the promoter regions on the Xi significantly declined only in a subset of genes (Ddx3x, 5530601H04Rik and Kdm5c/Jarid1c) in the XistΔrepE cells, although EZH2 recruitment and H3K27me3 deposition to the Xi during EB differentiation were indistinguishable between control TST and XistΔE/TST cells (Figs 3  and S5), suggesting that the reduction of H3K27me3 levels at some escape gene promoters might cause stochastic upregulation of escape genes on the Xi. Since reduction of H3K27me3 occurred only in a subset of genes and not on a global level, Utx upregulation might not cause indicates a gene known as an escape gene in brain. (B) 129 mutant-allele-specific RT-qPCR analysis of X-linked Pgk1 and Mecp2. The data was normalized to Gapdh and undifferentiated TST cells (set to 1). The mean ± SD from three independent experiments is shown. P-values were calculated by an unpaired t-test (*p<0.05). (C) Allele-specific expression analysis of X-linked genes using pyrosequencing. The allelic ratio of expression for each sample was normalized to pyrosequencing data using genomic PCR fragment (X 129 :X Cast = 50%:50%). The mean ± SD from three independent experiments is shown. P-values were derived from an unpaired t-test (*p<0.05, **p<0.01). (D) Allele-specific RNAseq analysis using 15 X-linked gene-specific primer sets. The mean ± SD from three independent experiments is shown. Sequencing reads matching the sequences for each transcript/allele (129 or Cast) were counted by a custom python script. P-values were derived from an unpaired ttest (*p<0.05). https://doi.org/10.1371/journal.pgen.1006890.g004 Role of Xist repeat E in X-inactivation

Fig 5. H3K27me3 level at X-linked gene promoter is reduced on the Xi in XistΔE mutant. (A)
Allele-specific ChIP experiments of Xlinked genes with antibodies against H3K4me3, H3K27me3, and control mouse IgG. The mean ± SD from two independent experiments is escape gene upregulation in the XistΔE/TST cells. In sum, these results demonstrate that Xist repeat E is essential for the establishment of global H3K4me3 distribution and H3K27me3 modifications at the escape gene promoter regions on the Xi.

Discussion
In this study, we have identified the repeat E of Xist RNA as an essential region for ASH2L recruitment to the Xi by a formaldehyde crosslinked RIP experiment and targeted deletion. In our recent study, direct interaction between hnRNP U and Xist RNA was revealed using a UV-crosslinked RIP [28]. Interestingly, the repeat E region is the least preferential binding region for hnRNP U, which preferentially binds to exon 7 of Xist RNA where the repeat E resides at the 5 0 end. The repeat E region could have a distinct function from those required for Xist RNA localization on the Xi via hnRNP U, which has both DNA and RNA binding domains at N-and C-terminals, respectively [41,42]. Recent work attempted to uncover the conformation of full-length Xist RNA in vivo [43][44][45]. Whereas the repeat E in Xist RNA is predicted as an unstructured region using computational analysis, in vivo dimethyl sulfate (DMS) chemical probing analysis indicated that the repeat E of Xist RNA is protected from the DMS attack, implying that the repeat E region could be protected by protein-RNA interactions or by forming a paired secondary structure [43,45]. More recent analysis for global RNA duplex formation based on reversible psoralen crosslinking (PARIS) enables us to clarify basepairing interactions on an individual-molecular level [44]. PARIS analysis using human HEK293 cells indicates that the frequent local and long-range helix formation within XIST RNA exon 6 (a counterpart of Xist exon 7 in mice) occurs where the repeat E resides. Since this domain structure of RNA duplex formation across XIST exon 6 is conserved, it is possible that unique duplex folding including the repeat E might be recognized by a variety of chromatin regulators including ASH2L, such as XIST RNA repeat A-SPEN RNA-protein complex [23,44]. While ASH2L interacts with Xist RNA repeat E, ASH2L also has DNA-binding activity at its N-terminal forkhead-like helix-wing-helix domain [46]. Putative DNA-binding affinity of ASH2L and its association with Xist RNA indicate that ASH2L might play an important role to construct a unique conformational structure of the Xi during XCI and may be involved in epigenetic dynamics on the Xi and escape gene expression.
Although we have identified the repeat E of Xist RNA as a binding region of ASH2L, a recent formaldehyde-crosslinked RIP analysis using a human myelogenous leukemia cell line revealed that several chromatin-associated proteins, such as components of PRC2 (SUZ12 and EZH2) and a component of histone deacetylase NuRD complex (CHD4), also bind to repeat E [47][48][49]. Our data in Figs 3, 4 and S5, however, indicate that the repeat E is dispensable for Xlinked non-escape gene silencing and for the recruitment of EZH2 and H3K27me3; thus, Xist repeat E is unlikely to be a major EZH2 binding region observed in a human myelogenous leukemia cell line. Indeed, localization of PRC2 on the Xi is transient during ES cell differentiation (S5 Fig) [37]. H3K27me3 modification on the Xi is proposed to be maintained by the association with the perinucleolar region in differentiated cells [50]. Thus, the role of the repeat E of Xist RNA as a scaffold for chromatin modifying factor might differ in the context of cell type. However, we found that the H3K27me3 level was downregulated at the promoter in a subset of escape genes on the XistΔE Xi (Fig 5). Although the molecular mechanism underlying H3k27me3 modification at the escape gene promoters remains to be elucidated, Xist repeat shown. Statistical significance was calucuated by an unpaired t-test (*p<0.05, **p<0.01). (B) ChIP experiments of autosomal genes with antibodies against H3K4me3, H3K27me3, and control mouse IgG. The mean ± SD from two independent experiments is shown. P-values were derived from an unpaired t-test (*p<0.05, **p<0.01). https://doi.org/10.1371/journal.pgen.1006890.g005 Role of Xist repeat E in X-inactivation E and ASH2L could be involved in the H3K27me3 distribution at the escape gene promoters on the Xi.
Surprisingly, the depletion of ASH2L from the Xi by the repeat E deletion resulted in transcriptional upregulation of a subset of escape genes we tested without affecting stable silencing of non-escape genes during XCI. We initially hypothesized that escape gene expression on the Xi is repressed due to the depletion of ASH2L from the Xi, since ASH2L is a common component of the MLL/SET H3K4 methyltransferase complexes for transcriptional activation [17]. As shown in our results and previous report [16], other common components of the MLL/SET complexes such as WDR5 and RBBP5 are not enriched on the Xi during XCI, suggesting a unique role for ASH2L on the Xi. Because of the enrichment of H3K4me2 and 3 at escape gene loci [40,51], we cannot eliminate the possibility that ASH2L is required locally for escape gene expression on the Xi as a part of the MLL/SET complexes. It is also interesting that Xist repeat E deletion affects H3K4me3 levels at autosomal and X-linked gene promoters (Fig 5A and 5B). The alteration of global H3K4me3 distribution might disturb global gene expression and partly contribute to an increased ratio of escape gene expression from the Xi compared to the Xa in the XistΔE/TST mutant cells (Fig 4C and 4D). Since Xist repeat E deletion resulted in increased free ASH2L in the nucleus, the increased amount of ASH2L might disturb the MLL/SET complexes' activity. Xist RNA might act as a molecular sponge of ASH2L to regulate the MLL/SET complexes' activity. A similar sponge function of noncoding gene Jpx, a positive regulator for Xist, has been reported [52]. Jpx expression is essential for Xist upregulation at the onset of random XCI through eviction of Xist repressor CTCF by its binding affinity with CTCF.
The physiological roles of Xist repeat E, of which mutations affect escape gene expression, remain to be elucidated. Several X-linked genes encoding chromatin modifying enzymes and transcriptional regulators, such as Kdm5c/Jarid1c and Kdm6a/Utx (histone demethylases for H3K4 and H3K27, respectively) and Ddx3x (ATP-dependent RNA helicase), are known as escape genes [8]. Recent studies have identified frequent mutations in genes involving transcriptional regulation such as chromatin modifying enzymes in human cancer, suggesting that those genes could play a major role in cancer pathogenesis, metastasis and malignancy by altering global gene expression [53]. Aberrant X-linked gene expression, including transcriptional regulators, in cancer cells has been also reported [54]. Indeed, overexpression of the escape genes KDM5C, KDM6A and DDX3X is associated with cancer development, proliferation, invasion and metastasis in certain type of cancers such as breast cancer [55][56][57][58][59]. In addition, significantly higher expression of the escape long noncoding FTX gene has been reported in colorectal cancer, although the downregulation of FTX is also associated with primary breast tumors [60,61]. Ftx potentially activates Xist in random XCI in mice [62]. Since partial upregulation of a subset of X-linked genes by Xist knockout in the hematopoietic stem cells leads to hematological cancer in mice [63], the strict conservation of appropriate level of X-linked gene expression including escape genes would be critical for suppression of cancer.
In conclusion, we provide evidence for the unique role of Xist repeat E for the recruitment of ASH2L and epigenetic and transcriptional regulation of X-linked genes, including escape genes. Further studies on the interaction between ASH2L and Xist RNA though the repeat E will help us to elucidate an underlying mechanism whereby expression levels of escape genes are properly regulated by Xist RNA.

Methods
Mouse ES cell culture 16.7 female ES cells were used to generate Xist repeat E deletion mutant [30]. ES cells were grown under standard conditions as described except for the addition of 2i inhibitors (3 μM CHIR99021 and 1 μM PD0325901, LC laboratories) [28]. Xist expression from tet-inducible Xist cDNA transgene in T20 cell lines was induced by adding 1 μg/ml doxycycline (Dox) in the culture medium for differentiation for 2 days [21]. Cre-LoxP mediated insertion of the inducible Xist cDNA transgenes was done as described [21].

Quantitative allele-specific RT-PCR
Allele-specific RT-qPCR analysis using SNP-based allele-specific primers for Xist, Tsix, Mecp2, Pgk1 on X 129 or X Cast was performed as described previously [28]. In addition, primer sequences used in the analysis for Xist repeat E, Usp9x and Hprt are as following: Xist repeat E

Half-life assay
Half-life assay for Xist RNA was performed as described previously [28].

Immuno-FISH
Immuno-FISH was performed as described [28,67]. Antibodies used in this study are the following

Allele-specific expression analysis using pyrosequencing
To determine allelic gene expression ratio between 129 and Cast alleles, RT-PCR products amplified from cDNA were analyzed by pyrosequencing analysis by Pyromark Q96 MD (QIAGEN). The relative expression ratio between the two alleles was quantified by the Pyromark MD SNP/ Genotyping software according to the manufacturer's instructions and normalized by the pyrosequencing data obtained from PCR fragment amplified from the genomic DNA (X 129 :X Cast = 50%:50%). The following primers were used for pyrosequencing experiments is shown in S1 Table. Allele-specific RNAseq using gene-specific primer sets for X-linked genes To determine allelic gene expression ratio of X-linked genes between 129 and Cast alleles, RT-PCR products amplified from cDNA using gene-specific primer sets for 12 X-linked genes (12 non-escape and 12 escape genes) were sequenced by Illumina HiSeq2500. RT-PCR was performed as described previously [28]. Gene-specific primers for first strand synthesis and cDNA amplification were shown in S1 Table. Equal molar of RT-PCR products of 12 X-linked genes derived from the same sample were pooled, and the second-round PCR (v2_Ad1_noMX as a common forward primer and v2_Ad2.25-v2_Ad2.60 as reverse primers with an index) was performed to attach adaptors for index and Illumina sequencing as described previously [68]. A custom python script was used to count the reads exactly matching the sequences for each transcript/allele.