Differential epigenetic landscapes and transcription factors explain X-linked gene behaviours during X-chromosome reactivation in the mouse inner cell mass

X-chromosome inactivation (XCI) is established in two waves during mouse development. First, silencing of the paternal X chromosome (Xp) is triggered, with transcriptional repression of most genes and enrichment of epigenetic marks such as H3K27me3 being achieved in all cells by the early blastocyst stage. XCI is then reversed in the inner cell mass (ICM), followed by a second wave of maternal or paternal XCI, in the embryo-proper. Although the role of Xist RNA in triggering XCI is now clear, the mechanisms underlying Xp reactivation in the inner cell mass have remained enigmatic. Here we use in vivo single cell approaches (allele-specific RNAseq, nascent RNA FISH and immunofluorescence) and find that different genes show very different timing of reactivation. We observe that the genes reactivate at different stages and that initial enrichment in H3K27me3 anti-correlates with the speed of reactivation. To define whether this repressive histone mark is lost actively or passively, we investigate embryos mutant for the X-encoded H3K27me3 demethylase, UTX. Xp genes that normally reactivate slowly are retarded in their reactivation in Utx mutants, while those that reactive rapidly are unaffected. Therefore, efficient reprogramming of some X-linked genes in the inner cell mass is very rapid, indicating minimal epigenetic memory and potentially driven by transcription factors, whereas others may require active erasure of chromatin marks such as H3K27me3.


67
In mammals, dosage compensation between XX females and XY males is achieved by 68 inactivating one of two X chromosomes during early female embryogenesis 1 . In the mouse, 69 X-chromosome inactivation (XCI) occurs in two waves during early female development. 70 The first wave takes place during pre-implantation development and is subject to genomic 71 imprinting, resulting in preferential inactivation of the paternal X (Xp) chromosome 2 . In the 72 trophectoderm (TE) and the primitive endoderm (PrE), which contribute respectively to the 73 placenta and yolk sac, silencing of the Xp is maintained 3,4 . In contrast, in the epiblast 74 precursor cells within the inner cell mass (ICM) of the blastocyst, (which correspond to 75 mESCs), the Xp is reactivated and the second XCI wave and random inactivation of either 76 Xp or the maternal X chromosome (Xm), occurs shortly after 5,6 . The inactive state is then 77 stably maintained and transmitted through cell divisions in the soma. Lysine 27 (H3K27me3) 9 . The inactive X chromosome is also enriched for mono-methylation 88 of histone H4 lysine K20, di-methylation of histone H3 lysine K9 and the histone variant 89 demethylases [23][24][25] . JMJD3 appears to inhibit reprogramming 26 , whereas UTX plays a role in 114 differentiation of the ectoderm and mesoderm 27 and has been proposed to promote somatic 115 and germ cell epigenetic reprogramming 24 . Interestingly, the Utx gene escapes from X-116 chromosome inactivation (ie is transcribed from both the active and inactive X 117 chromosomes) 28 . This raises the intriguing possibility that Utx might have a female-specific 118 role in reprogramming the Xi in the inner cell mass of the mouse blastocyst. Utx knockout 119 mouse studies have suggested an important role of Utx during mouse embryogenesis and 120 germline development, but its exact molecular functions in X-linked gene transcriptional 121 dynamics have not been assessed 21,22,24,29,30 . 122 In this study we set out to obtain an in-depth view of the nature of the X-chromosome 123 reactivation process in the ICM in vivo. We have defined the chromosome wide timing of X-   We examined further genes for their timing of Xi reactivation in the ICM. We 170 performed RNA FISH in pre-implantation (E3.5, early) through to peri-implantation (E4.5, 171 late) blastocysts for 8 X-linked genes together with Xist (Atp6ap2,Fmr1,Kif4,Rnf12,Abcb7,172 Atrx, Atp7a and Pdha1) (Figure 2b). The genes were chosen based on their known range of 173 silencing kinetics during imprinted XCI in pre-implantation embryos, including genes 174 silenced early (prior to E3.0 such as Kif4, Rnf12, Atp7a, Atrx and Abcb7), late (after E3.0 e.g.

178
Increased frequencies of biallelic expression were observed for most genes in female 179 ICM cells from the E4.0 stage onward (Fmr1, Kif4, Atp7a and Pdha1 and Rnf12), indicating 180 that they have reactivated in a subset of ICM cells (presumably pre-epiblast cells) ( Figure   181 2c). However Atrx displayed biallelic expression as early as E3.5, similarly to Abcb7 gene 182 (also shown in Figure 2a). Thus, reactivation of Atrx and Abcb7 occurs in the early ICM cells 183 prior to any lineage segregation between epiblast (Epi) and primitive endoderm (PrE) cells 32-184 34 . Interestingly, just half a day later at E4.0, a decrease in biallelic expression of these two 185 genes was seen in 30% to 60% of ICM cells (Figure 2c). Previous studies have shown that X-186 chromosome reactivation occurs in epiblast cells 5,6 , whereas PrE-derived tissues maintain an 187 inactive Xp 4 . The decrease we observed in biallelic Atrx and Abcb7 expression at E4.0 and 188 E4.5 in ICM cells could indicate that these genes are silenced again, presumably in future 189 primitive endoderm cells. In the case of Atp6ap2, which is a gene that normally escapes from 190 XCI, as expected, it was found to be biallelically expressed in 60 to 80% of ICM cells at all 191 stages 11 (Figure 2c). Taken together, our data suggests that the reactivation of X-linked genes 192 occurs with very different timing in ICM cells during early to late blastocyst stages.

193
Furthermore, we find that a subset of genes may be reactivated early on, but then become 194 rapidly silenced again in a sub-population of cells, presumably destined to become PrE.

195
To examine whether biallelic expression of more slowly reactivated genes correlates 196 with pre-epiblast differentiation (and thus NANOG protein), we performed NANOG 197 immunofluorescence combined with RNA FISH for Xist and two such X-linked genes (Kif4 RNA coating are linked to biallelic expression of late reactivated genes, but that Nanog 208 expression alone is not sufficient. Taken together, our data point to reactivation in a lineage-209 specific manner beyond the mid ICM stage for genes that are late-reactivated. They also 210 reveal a lineage-independent reactivation of the early-reactivated genes at E3.5 ICM. The remarkable diversity in X-linked gene reactivation observed above ( Figure 2) prompted 214 us to explore the Xp reactivation process on a chromosome-wide scale. Furthermore, given 215 the mixture of cells in the ICM, some of which are destined to become PrE, while others will 216 become Epiblast, we were interested to know whether reactivation or silencing maintenance 217 of Xp-linked genes correlated with PrE factor (eg. Gata4 or Gata6) and/or pluripotency factor 218 expression (eg. Nanog, Oct4, Sox2) at the single-cell level 35 Figure 3a). We found that E3.5 ICM cells still showed substantial heterogeneity compared to 229 E4.0 ICM single cells, which clustered into two distinct groups. Nevertheless, some signs that 230 2 sub-populations are emerging could be seen at E3.5 for some ICM cells. This revealed that 231 developmental stage (E3.5 versus E4.0) does not seem to be the primary source of variability 232 but that lineage specification between primitive endoderm and epiblast precursor cells could  expressed genes when compared to TE. We found that 51% of X-linked genes were 271 expressed from both X-chromosomes in E3.5 ICM (55 biallelic genes out of 107 in total, e.g. give rise to the PrE, where XpCI is known to be maintained, ultimately 4 . Our RNA-FISH data 280 confirms that Atrx is transiently expressed from both X chromosomes even in the cells that  Interestingly, some genes classified as "very late reactivated" still appear to be repressed on  Table 2). This could be explained by 294 differences between nascent (RNA-FISH) and mature RNA (scRNAseq) for this gene, if the 295 levels of paternal mRNA are not yet high enough for scRNAseq detection even though the 296 gene has begun to be transcribed. 297 We describe here that X-chromosome reactivation can initiate for some genes   Figure 2d). We thus hypothesize that late and very late reactivated genes may have acquired 329 an epigenetic signature that prevents their rapid reactivation in early ICM cells, compared to 330 early-reactivated genes. Early-reactivated genes on the other hand, may become expressed 331 more rapidly due to specific TFs overriding their silent state. 332 We first examined recent allele-specific ChIPseq data for H3K27me3 and H3K4me3 333 in ICM of pre-implantation embryos (pooled between E3.5-E4.0) 38 . We overlapped the genes 334 for which there is allelic information between this study and our different reactivation-timing   (Supplementary Figures 3c, 3d). H3K27me3 during XCI and may thus be more prone to rapid reactivation, remain unknown.

529
Interestingly, expression of several epigenetic modifiers appeared to correlate with X-530 chromosome reactivation (Supplementary Table 3        Immunofluorescence followed by RNA-FISH were carried out as described previously 5 .

727
Images were acquired using Inverted laser scanning confocal microscope with spectral 728 detection (LSM700 -Zeiss) equipped with a 260nm laser (RappOpto), with a 60X objective 729 and 0.2 µm Z-sections or a confocal wide-field Deltavision core microscope (Applied were analysed using ImageJ software (Fiji, NIH).

734
ICMs obtained from Utx FDC/wt females were PCR-genotyped after image acquisition (details 735 available upon request).

736
All the antibodies and probes used in this study are listed in Supplementary Table 3 along   737 with the information on dilution ratios.   Table 1).

767
Quality control and filtering of raw data 768 Quality control was applied on raw data as already described in Borensztein et al., 2017 7 . 769 Sequencing reads characterized by at least one of the following criteria were discarded from 770 the analysis: 771 1. More than 50% of low quality bases (Phred score <5).

857
For genes whose length was below 5kb, gene size was taken as window. (Distribution of gene 858 size for each group was not significantly different, data not shown).  Kruskal-Wallis and Post-hoc test were used to analyse non-parametric and unrelated samples.

881
The statistical significance has been evaluated through two-sided Dunn's Multiple The authors declare no competing financial interests.