Contiguous erosion of the inactive X in human pluripotency concludes with global DNA hypomethylation

Abstract

Female human pluripotent stem cells (hPSCs) routinely undergo inactive X (Xi) erosion. This progressive loss of key repressive features follows the loss of XIST expression, the long non-coding RNA driving X inactivation, and causes reactivation of silenced genes across the eroding X (Xe). To date, the sporadic and progressive nature of erosion has obscured its scale, dynamics, and key transition events. To address this problem, we perform an integrated analysis of DNA methylation (DNAme), chromatin accessibility, and gene expression across hundreds of hPSC samples. Differential DNAme orders female hPSCs across a trajectory from initiation to terminal Xi erosion. Our results identify a cis-regulatory element crucial for XIST expression, trace contiguously growing reactivated domains to a few euchromatic origins, and indicate that the late-stage Xe impairs DNAme genome-wide. Surprisingly, from this altered regulatory landscape emerge select features of naive pluripotency, suggesting that its link to X dosage may be partially conserved in human embryonic development.

Keywords: DNA methylation; X chromosome; XIST; chromatin accessibility; escapee genes; expression; gene dosage; human endogenous retroviral elements; naive pluripotency; primed pluripotency.

Figure 1.. XIE reflected in high X-specific variance in female hPSCs

(A) β values distribution of probes on the X in female (red) and male (blue) samples. (B) Principal-component analysis (PCA) of female (red) and male (blue) samples using only X-linked probes with Brown-Forsythe p ≤ 0.01. (C and D) Density plots of probe β variance on X and autosomes, respectively, with red lines for female samples, and blue for male samples (Kolmogorov-Smirnov [KS] distance indicated, with KS p values of 5.6e–15 and 0.78 for C and D, respectively). (E) PCA of DMPs across female samples colored by their k-means assigned cluster (see Method details). Gray lines link different passages of the same cell line. Shaded areas represent the 95% confidence ellipse for each cluster. (F and G) Projection of C19 and C23 iPSC validation samples, using PCs from (E), labeled by passage numbers, and colored as in (E). (H) Fraction of DMPs demethylated to within 10% of expected DNAme loss captured in each passage of C19 and C23 iPSCs. (I) DNAme heatmap of high variance X probes in female hPSCs. Samples within each k-means cluster (A–F) are ordered by mean X DNAme. Probes (rows) are grouped by transition (lowest q ≤ 0.05), and β value change (↑ up, ↓ down, or ~ fluctuating irrespective of transition). Cluster transitions are numbered (1–5). Cell lines with data from multiple passages are described below the heatmap, with arrows drawn between samples and the passage numbers marked. (J) DNAme heatmap for C19 (dark gray) and C23 (light gray) iPSC samples, ordered by cluster and passage as annotated below heatmap.

Figure 2.. Differential DNAme analysis on Xe distinguishes early from late XIE

(A) Density plot of probes across X coordinates reveal transition 1 probes (1↓↑, blue) to concentrate irrespective of overall probe density (gray-shaded area), in contrast to fluctuating probes (~, black). KS distances quantify similarity to frequency distribution of the null probe density distribution (* KS test p value = 5.09e–11 for blue transition-1 probes). (B) Volcano plot of transition-1 differentially methylated probes (DMPs), colored by X coordinate (legend above the plot). Top increasing DMPs annotated by gene name. (C) DNAme of XIST and FIRRE probes across clusters A–F. Lines to (F) and (G) indicate the position of probes in relation to the genes and their annotated regions (* indicates DMPs from B). (D) XIST and FIRRE expression (VST + batch-corrected/log2-transformed), respectively, for all samples with available RNA-seq data (all but cluster E). (E) XIST CpGII/P2 ATAC peak accessibility score over DNAme (β) values, with symbol size indicating XIST expression (VST + batch-corrected/log2-transformed) across all A–F samples with all 3 datapoints. (F and G) ATAC-seq coverage for XIST and FIRRE, respectively. Horizontal lines below coverage indicate called peaks (gray, black), including differential, closing peaks (red). Bottom panels of (F) and (G) relate DMPs and ATAC peaks to cis-regulatory sites for XIST and FIRRE, respectively.

Figure 3.. Contiguous spread of XIE from euchromatic origins on the Xi

(A) Map of cumulative (gray) and transition-specific DMP (colors) changes in β values across the X plotted underneath Xi-specific H3K27me3 (dark teal) and H3K9me3 (brown) H9 chromatin immunoprecipitation sequencing (ChIP-seq) signal from Vallot et al. (2015). Gray vertical lines indicate escapee locations, determined from DNAme in cluster A, as in Cotton et al. (2015). (B) Differential ATAC-seq coverage (peak width sum per 100-kb intervals), across increasing and decreasing ATAC peaks, and correlations with DMP map (A) for each transition on bottom right. Transitions 4 and 5 are combined (purple), due to lack of ATAC-seq data for cluster E. Histograms of log2 fold change (log2FC) of differential peaks for each transition are shown on the top right (inset). (C) Pearson correlations of transition-specific DMP changes (colored in A) to H3K27me3 (dark teal) and H3K9me3 levels (brown). (D and E) DMP distance distributions relative to escapee locations for each transition on Xp and Xq, respectively, with black vertical lines denoting median distance. Null distance distributions using permuted escapees (broad gray) or permuted DMPs (narrow gray) plotted underneath actual distributions. Calculated p values (two-tailed rank test against the permutated distributions) indicated on the top left (permuted escapees) and top right (permuted DMPs) of each plot. (F) Pearson correlations of transition-specific ATAC-peaks (B) to H3K27me3 (dark teal) and H3K9me3 levels (brown). (G and H) ATAC-peak distance distributions relative to escapee locations for each transition on Xp and Xq, respectively, with black vertical lines denoting median distance.

Figure 4.. Global hypomethylation and emergent naive pluripotency markers in terminal XIE

(A) Autosomal methylation distributions across female hPSC clusters (A–F, red shades) and their KS distances relative to male hPSC distribution (teal). (B) Chromosome-resolved DMPs of the final XIE transitions (4 or 5). Total number of probes changing for autosomes and X listed with colors corresponding to increasing (red) or decreasing (blue) DNAme. The transparency is on a continuous scale based on the β-differential. (C) As in (B), but plotting differentially expressed genes (DESeq2) with concordant DNAme changes (B) and/or ATAC changes. The transparency is on a continuous scale reflecting the log2FC. (D) Transition- and autosome/X-resolved differentially expressed genes annotated in the COSMIC cancer gene census (Sondka et al., 2018) (also listed in Table S4). (E) Median X:autosome ratio for each female sample cluster (Wilcoxon p value relative to cluster A). (F) Differential expression scatterplot of transition-4/−5-specific genes (y axis) over log2FCs (x axis) of these genes between naive and primed hPSCs from Kilens et al. (2018). (G) Comparison of expression changes in transition-4/−5-specific differential genes (dark red) relative to corresponding changes (gray) in naive versus primed hPSCs from Kilens et al. (2018). (H) Cumulative changes (relative to cluster A) in all differentially expressed HERVs identified by Telescope (Bendall et al., 2019) (VST + batch-corrected/log2-transformed). Only significantly overrepresented HERVs (hypergeometric p ≤ 0.1) are labeled by colors (others in gray). Color legend lists total counts of differentially expressed HERV classes (with hypergeometric significance indicated as *p < 0.1, **p < 0.05, ***p < 0.01, ****p < 0.001). (I) Tally of differentially expressed HERV classes (H) overlapped by significantly differential ATAC peaks (Fisher right-side p value indicated as *p < 0.01, **p < 0.001, ***p < 0.0001). Negative overlaps indicate overlaps by closing ATAC peaks, and positive by opening peaks.