In this study, we analysed one male cell line (GHP7/7) and two independently derived female XEN cell lines (GHP7/9 and GHP7/3) 26 . All three cell lines harboured morphological features typical of XEN cells and expressed PrE-specific lineage markers (Additional file 1). The female cell lines carry an Xp of 129Sv.Pgk1a origin and an Xm of 129Sv origin, allowing the identification of allele origin through the extensive polymorphisms located in the large Pgk1a-derived region surrounding the Xist gene.

In a previous report, the authors indicated that I-XCI in XEN cells, unlike that in TS cells, is not accompanied by PRC2 recruitment and accumulation of associated H3K27me3 on the inactive Xp and suggested that XEN cells might express a distinct set of chromatin remodellers 26 . To address this question we compared, in XEN and in TS cells, the mRNA steady-state levels of 40 known epigenetic regulators using single-cell RT-qPCR analysis (Figure 1A and Additional file 2A and B). On the basis of their expression profiles, ten of the forty epigenetic regulators were able to discriminate between the two cell populations. These XEN/TS epigenetic discriminators included members of the PRC1 and PRC2 complexes. Unexpectedly, given the previously reported findings 26 , mRNAs encoding the EZH2 H3K27 methyltransferase appeared more abundant, globally, in XEN cells than in TS cells.

This observation prompted us to reassess the distribution of H3K27me3 in XEN female cells using three-dimensional immuno-RNA-FISH for Xist and H3K27me3, followed by a deconvolution step (Figure 1B). Xist coating of the Xi was associated with a coenrichment in H3K27me3 in the majority of XEN cells, similarly to what we observed in TS cells (Figure 1C and Additional file 3). Although the volumes of H3K27me3 enrichment and of Xist domains on the Xi were not significantly different (Figure 1D), H3K27me3 and Xist nuclear foci did not show strict superimposability in the majority of XEN nuclei (Figure 1B), indicating that certain regions of the Xi, coated by Xist, may lack H3K27me3.

These results suggest that XEN and TS cells have distinct epigenetic identities. Notably, XEN cells highly express the Ezh2 gene. Consistent with this, the Xi in XEN cells is globally enriched in H3K27me3.

In order to confirm these results and identify the genomic regions of the inactive Xp showing lower levels of H3K27me3, we performed ChIP-chip analysis for this histone mark on male and female XEN cells using a high-resolution microarray covering the X chromosome and a subpart of chromosome 17, which served as a ChIP-chip efficiency control (see the Methods section for data treatment and quality controls). Comparing male and female profiles allowed us to detect differences marking specifically the Xi.

X-linked genes, but not intergenic intervals, appeared significantly more enriched in H3K27me3 in female compared to male XEN cells, indicating that genes constitute the preferential targets of H3K27me3 accumulation on the Xi (Figure 2A). In addition, the majority of expressed genes, as opposed to silent genes, exhibited higher levels of H3K27me3 in female compared to male XEN cells. This is suggestive of an accumulation of this modification on Xi alleles, although we noticed that some expressed genes lacked H3K27me3 in both male and female extraembryonic cells, which is indicative of biallelic H3K27me3 depletion (Figure 2B). We were able to identify these H3K27me3-depleted Xi genes (H3K27me3-low) using unbiased k-means clustering based on the statistical comparison of H3K27me3 percentages along gene bodies in male and female XEN cells (see Figure 2C, the Methods section and Additional files 2C and 4A). Approximately 15% of expressed genes exhibited such a profile in a female XEN cell line (GHP7/9), as opposed to 19.5% that showed similar behaviour in TS cells and only about 7% in liver cells (Figure 2B and Additional file 4B and C). Amongst these H3K27me3-low genes, 19 were common to both XEN and TS female cells, and of those, only 2 were associated with reduced levels of H3K27me3 in the female liver (Additional file 4C).

We then assessed whether this lack of H3K27me3 on specific Xi genes in female XEN cells was associated with a gain of active histone mark H3K4me2 using a similar ChIP-chip approach. Lowered H3K4me2 levels in female XEN cells than in male XEN cells were observed at the majority of expressed genes, consistent with an absence of active histone marks on Xi alleles (Additional file 5A). In contrast, H3K27me3-low Xi genes appeared significantly enriched in H3K4me2 compared to H3K27me3-high Xi genes (P = 0.037 by Kolmogorov–Smirnov test) (Additional file 5B and C). We noted, however, that this effect was not observed at all H3K27me3-low genes and that some H3K27me3-high genes may also show high levels of H3K4me2 in female cells, indicating that the anticorrelation between H3K4me2 and H3K27me3 is not strict.

On the basis of these ChIP-chip analyses, we conclude that both the active and repressive histone marks H3K4me2 and H3K27me3 are distributed in a lineage-specific manner on Xi genes. Many of the genes lacking H3K27me3 on the Xi of XEN cells specifically showed a gain of H3K4me2 compared to Xi genes enriched in H3K27me3, indicating that some of these H3K27me3-low genes adopted the characteristics of active chromatin.

In order to address whether these active chromatin features are associated with gene expression from the Xi in XEN cells, we performed two types of analyses. First, in single XEN cells, we measured by allele-specific RT-qPCR, the maternal and paternal expression of five genes with high H3K27me3 levels (Atp7a, Atrx, Rlim, Apool and Sh3bgrl) and five genes with low H3K27me3 levels (Ftx, Jpx, Pbdc1 (also known as 2610029G23Rik), Brwd3 and Rps6ka6), as well as Xist transcripts (see Additional file 6A for RT-qPCR controls). As expected in cells subjected to I-XCI, Xist expression appeared to be restricted to the Xp and genes showing high H3K27me3 levels were expressed predominantly from the active Xm (Figure 3). All five genes lacking global H3K27me3, on the other hand, showed a significant level of paternal expression in XEN individual cells, indicating that these genes escaped from I-XCI (Figure 3). We noted that the relative levels of expression from both the Xp and the Xm were extremely variable, both amongst cells and between escaper loci, leading to a highly heterogeneous XEN cell population. Consistent results were obtained with mixed cell populations (Additional file 6B).

Second, we performed RNA-FISH on 13 different X-linked regions: 9 regions showing high levels of H3K27me3 (Fgd1, Atp7a, Pgk1, Elf4-Aifm, Taf1-Ogt, Flna, Cenpi, Tcfe3 and Rlim) and 4 genes showing low levels of H3K27me3 (Kdm5c, Ftx, Hdac8 and Kdm6a) (Additional file 2D). Importantly, some of these regions (namely, Taf1-Ogt, Flna, Cenpi, Tcfe3, Hdac8 and Kdm6a) displayed different H3K27me3 and H3K4me2 levels in XEN than in TS cells (Additional file 2C). The comparison of RNA-FISH expression profiles between these two extraembryonic cell types allowed us to assess the lineage specificity of Xp gene silencing in extraembryonic lineages (Figure 4).

In agreement with our chromatin profile analysis, most H3K27me3-high Xi genes in each extraembryonic cell type were transcribed predominantly from the Xa (Xist-lacking X) and not from the Xi (Xist-coated X), indicating that these genes are subject to I-XCI. H3K27me3-low genes, on the other hand, appeared to be transcribed from the Xi in a significant percentage of nuclei in both cell types, indicating that they escaped from I-XCI (Figure 4). Two exceptions were observed. Hdac8, which we expected to escape from I-XCI in XEN cells based on its H3K27me3 profile, showed no significant expression from the Xi in these cells (Figure 4B). This may be related to its globally highly reduced expression level (low steady-state level of mature transcripts 29 ), which may introduce some stochasticity into transcription initiation. Conversely, the Kdm6a gene is transcribed from the Xi in a significant percentage of TS nuclei, despite showing significant levels (about 47%) of H3K27me3 on the Xi in this cell line (Figure 4D). However, the proportion of nuclei showing this behaviour was significantly smaller than that seen in female XEN cells in which Kdm6a was poorly enriched (about 10%) in H3K27me3. Kdm6a is known to robustly escape from XCI in somatic tissues 30 .

As previously described for TS cells 13 and other adult cell types 31 , we observed that maternal allelic expression prevailed over that of the paternal allele in XEN cells for all the escapers examined, suggesting that, for most XCI escapers, the escape mechanism was not fully efficient in relieving the inactivation. Indeed, the degree of escape, as estimated both by the number of nuclei exhibiting transcriptional activity from the Xi detectable by RNA-FISH and by the number of cells showing significant paternal expression detectable by single-cell RT-qPCR, appeared to be extremely variable, not only between cell types (for example, Ftx showed different levels of escape in TS cells compared to XEN cells) but also from gene to gene. This supports the idea that expression from the Xi at loci escaping from I-XCI is a locally regulated phenomenon.

In order to address whether the molecular mechanism controlling X-linked gene silencing and/or escape in XEN cells involves exclusively the specific distribution of H3K27me3 along the Xi, XEN cells were treated with GSK126, a potent and specific inhibitor of EZH2, the PRC2 catalytic subunit responsible for H3K27me3 apposition 32 . Importantly, the inhibitor had no visible impact on XEN cell morphology or growth rate and did not significantly change the expression of XEN-specific markers, indicating that GSK126 did not have important pleiotropic effects on XEN cell homeostasis (Additional file 7A and C). Western blot analysis of H3K27me3 on GSK126-treated compared to untreated XEN cells showed that the EZH2 inhibitor induced a progressive reduction in the global levels of H3K27me3, leading to barely detectable H3K27me3 levels after 5 days of treatment (Additional file 7B). Immunofluorescence analysis of H3K27me3, coupled with RNA-FISH for Xist, confirmed that most XEN cells (65%) treated with GSK126 lacked H3K27me3 accumulation on the Xi (Xist-coated X) after 5 days treatment (Figure 5A).

To assess the effect of GSK126-induced depletion of H3K27me3 on the expression of genes on the Xi, we analysed the expression of Xist, three genes subjected to I-XCI (Atp7a, Chic1 and Hmgn5) and three genes escaping from I-XCI (Ftx, Pbdc1 and Jpx) using allelic RT-qPCR on pools of XEN cells treated for 5 days with the drug. The expression levels of some of the X-linked genes, including Xist, were significantly increased after GSK126 treatment. Importantly, this increase affected exclusively alleles that were already active prior to any treatment—that is, the paternal Xist allele and maternal alleles of some other X-linked genes—suggesting that the inhibitor treatment leads to further derepression of active promoters (Figure 5B). In contrast, no significant change in expression levels of Xp alleles (except from Xist, as mentioned previously) was observed in GSK126-treated compared to untreated XEN cells, indicating that the marked reduction of H3K27me3 on the Xi of XEN cells did not significantly modify the efficiency of Xi gene silencing (Figure 5B). Similar results were obtained with two other H3K27me3-low genes (Kdm6a and Nkap) at the single-cell level using RNA-FISH following H3K27me3 immunostaining (Additional file 7D and E). Note that neither of the two analyses completely excludes the possibility that highly localised H3K27me3 may have remained associated with gene promoters, after GSK126 treatment, in some cells.

This analysis shows clearly that, although H3K27me3 enrichment allows the selective identification/segregation of genes subjected to I-XCI compared to genes escaping from I-XCI, this histone mark is probably not, by itself, determinant for X-linked gene expression control. This implies that other epigenetic marks or transcriptional mechanisms might be implicated in the regulation of Xi expression in XEN cells.

Because observations of expression of genes from the Xp have previously been reported in the VE of EED mutants 24 33 , we hypothesized that this extraembryonic tissue might be prone to labile Xp expression. Thus, we treated our female XEN cells with bone morphogenetic protein 4 (BMP4), a factor known to trigger XEN differentiation towards the VE fate 34 35 , and then assessed the status of transcription at the Xi. After 5 days of induction, XEN cells formed a homogeneous, epithelium-like layer of cells characteristic of VE conversion (Figure 6A). Our expression analysis of lineage determinants at the single-cell level shows that the vast majority of cells had repressed PrE- and PE-specific markers and activated VE-specific genes, thus confirming the VE differentiation (Figures 6B and C and Additional file 2E). Transcription from the Xi, which we assessed by RNA-FISH, was not significantly induced upon VE conversion at the H3K27me3-high genes Taf1, Rlim and Pgk1 or at the H3K27me3-low gene Kdm5c (Figures 6D and E). This observation indicates that, unlike what we saw during TS cell differentiation (Additional files 2E and 8), no change in Xi expression—at least for the genes we tested—is associated with XEN differentiation into VE.

Male GHP7/7 and female GHP7/9 and GHP7/3 XEN cell lines used in this study and have been described previously 26 . GHP7/9 and GHP7/3 cell lines carry an Xp from a 129.Pgk1a origin and an Xm from a 129/Sv origin, providing polymorphisms allowing for allelic distinction. XEN cells were grown on gelatin-coated culture dishes and routinely cultured at 37°C in 8% CO₂ in Dulbecco’s modified Eagle’s medium (DMEM) (Life Technologies, Carlsbad, CA, USA) supplemented with 15% foetal bovine serum (Life Technologies) and 100 mM β-mercaptoethanol 26 . XEN cell conversion into visceral endoderm was induced upon addition of 25 ng/ml BMP4 (R&D Systems, Minneapolis, MN, USA) in serum containing DMEM 34 . XEN cell treatment with GSK126 was carried out as follows: three concentrations of inhibitor were tested during a 5-day treatment period (0.5 μM, 2 μM and 3 μM GSK126). Male F2 and female F3 TS cell lines and culture conditions have been described previously 12 18 .

ChIP assays were performed as previously described 54 using an H3K27me3 antibody (07–449) and an H3K4me2 antibody (07–030; both from EMD Millipore, Billerica, MA, USA) and analysed as described previously 18 (see also Additional file 9 for the detailed protocol). ChIP-chip data have been deposited under Gene Expression Omnibus accession number [GSE:50587].

Raw data were analysed using the Tobias Straub protocol (http://www.epigenesys.eu/images/stories/protocols/pdf/20111025114444_p43.pdf) with the Bioconductor R interface. This protocol includes quality assessment, data normalization and transformation. Raw H3K27me3 data in male and female livers were taken from the Gene Expression Omnibus database [GSE:20617] 31 and subjected to the same statistical treatment as the TS and XEN data sets. The gene expression data of TS and XEN cells were extracted from [GSE:15519] 29 , and gene expression data of liver cells were extracted from [GSE:13583] 55 .

RNA-FISH and immuno-RNA-FISH procedures were carried out as described at http://www.springerprotocols.com/Abstract/doi/10.1007/978-1-59745-406-3_18?verPrint=print 56 (H3K27me3 antibody, 07–449; EMD Millipore). For details of our experimental procedures, see Additional file 9. For probe localisation, see Additional file 2D. Z-stacks were captured (step = 0.2 μm) on a Zeiss Axioplan 2 microscope (Carl Zeiss Microscopy, Thornwood, NY, USA) equipped with a Hamamatsu ORCA ER charge-coupled device camera (Hamamatsu City, Japan) and controlled using Volocity Acquisition software (PerkinElmer, Waltham, MA, USA). Z-stacks deconvolution consisted in 50 iterations per each channel using Huygens Professional software (Scientific Volume Imaging, Hilversum, the Netherlands). Volume measurements were performed using Volocity Professional software, and intensity plot profiles were obtained on an image analysis platform in ImageJ software (National Institutes of Health, Bethesda, MD, USA).

Single-cell gene expression analysis was performed as described previously 18 57 58 and as recommended by Fluidigm (http://www.fluidigm.com/single-cell-expression.html; South San Francisco, CA, USA). Briefly, TS and XEN cells were sorted by fluorescence-activated cell sorting using the MoFlo system (Beckman Coulter, Brea, CA, USA), and individual cells were distributed into wells or 96-well plates containing 5 μl of CellsDirect resuspension buffer (Invitrogen, Carlsbad, CA, USA). The preamplification step consisted of 20 cycles using a mix of universal primer pairs to preamplify each gene simultaneously. Preamplification was followed by exonuclease I treatment (New England Biolabs, Ipswich, MA, USA), and allelic qPCR was performed on a BioMark thermal cycler (Fluidigm). Raw efficiencies of each PCR assay and allelic specificity were measured on control DNA within each experiment. Transcript levels were extrapolated using the raw PCR efficiencies, thus allowing the direct comparison of different genes. Controls for allelic specificities of PCR assays are available upon request. See Additional file 2A for primer sequences.