哺乳动物胚胎发育过程中依赖H3K9me3的染色质重编程
Reprogramming of H3K9me3-dependent heterochromatin during mammalian embryo development
Result
1 小鼠配子和早期胚胎中H3K9me3的全基因组特点
Fig1b The UCSC genome browser view of H3K9me3 signals around Zscan4d (right panel) and Gnas (left panel) in mouse gametes, early embryos, mESCs (from either this study or from ENCODE) and mouse TSCs. Signals represent the log2-transformed H3K9me3/input ratio. Chr, chromosome.
In b–f, ChIP–seq for H3K9me3 was performed two times for each indicated stage, except for TSCs (three times) and E7.5 epiblast tissue (four times), and data shown represent the average values for each stage.
Fig1c, Bar plots show the enrichment of H3K9me3, H3K4me3 and H3K27me3 in promoter, SINE, LINE and LTR regions.
promoter, SINE, LINE 和 LTR 区域的 H3K9me3, H3K4me3, H3K27me3 富集
In c, H3K4me3 and H3K27me3 ChIP–seq data for MII oocyte, sperm, 2-cell-, 4-cell-, 8-cell- and morula-stage embryos, E3.5 ICM, E3.5 trophectoderm and TSCs were from our previous publication (GSE73952)20 and others were derived from this study: H3K4me3 ChIP–seq was performed two times except for E7.5 epiblast and E7.5 extraembryonic tissues (three times), and H3K27me3 ChIP–seq was performed three times except for E7.5 epiblast and ESCs (two times).
Fig1d, Graph showing the percentage of genomic regions covered by H3K9me3 peaks.
H3K9me3峰覆盖的基因组区域百分比
Fig1e, Graph showing the fraction of established and disappeared H3K9me3 domains during embryonic development.
Fig1f, Graph showing the number of H3K9me3-marked LTR regions (left panel) and promoter regions (right panel) during embryonic development.
2 小鼠早期胚胎依赖H3K9me3的异染色质动态变化
a, Left panel, heatmaps showing the dynamics of H3K9me3 domains during mouse pre-implantation embryo development. The colours represent the log2-transformed H3K9me3/input ratio scaled by row.
Middle and right panels, heatmaps showing the H3K27me3 level and DNA methylation level on H3K9me3 clusters during mouse pre-implantation embryo development. Heatmaps were generated using the same order of H3K9me3 clusters. The colours represent the log2-transformed H3K27me3/input ratio scaled by row (middle panel) and the absolute DNA methylation levels (right panel).
For each cluster, the averaged distance to the TSS, LTR regions and mESC HP1-binding sites are also plotted. The colours represent distance (log10). Data shown represent the averages for two independent H3K9 ChIP–seq experiments. WGBS was performed once for each sample. H3K27me3 ChIP–seq data for zygotes were from experiements performed three times and data for pre-implantation embryos were from our previous publication (GSE73952).
b, The UCSC genome browser view of representative regions of oocyte-specific (top), cleavage-specific (middle) and blastocyst-specific (bottom) H3K9me3 domains. Signals represent the average log2-transformed H3K9me3/input ratio from two independent ChIP–seq analyses.
c, Left, gene ontology analysis of oocyte-specific, cleavage-specific and blastocyst specific H3K9me3-covered genes. P values were determined based on a modified Fisher’s exact test.
三组domains的GO分析
Right, box plot for the expression level of oocyte specific, cleavage-specific and blastocyst-specific H3K9me3-covered genes during pre-implantation embryo development.
3 受精后，父母源基因组都经历了大量的H3K9me3重编程
a, Hierarchical clustering of H3K9me3, H3K4me3, H3K27me3 and DNA methylation levels in gametes and two alleles of embryos. H3K4me3 and H3K27me3 ChIP–seq data for MII oocyte, sperm, early 2-cell, late 2-cell and E3.5 ICM stages were from public data (GSE71434 and GSE76687)27,42. M, maternal; P, paternal.
b, Graphs showing the number of lost, inherited and gained regions during fertilization for H3K9me3, H3K4me3, H3K27me3 and DNA methylation.
c, Heatmaps showing the H3K9me3 enrichment in lost, inherited and gained regions of gametes and two alleles of zygote samples. Left panel shows the maternal regions and the right panel shows the paternal regions. The colours represent the normalized SNP-trackable H3K9me3 read counts and the values are scaled by row.
d, The UCSC genome browser view of the representative lost, inherited and gained regions for maternal (left) and paternal (right) H3K9me3 signal during fertilization. Signals represent ChIP–seq RPM.
4 小鼠植入前胚胎H3K9me3的allelic-specific图谱
接下来，作者想知道这种亲本H3K9me3的不对称性是否会在早期发育阶段保持下去。
a, Scatter plots showing the number of H3K9me3 reads assigned to each allele. The red, green and blue colours denote maternal-specific, bi-allelic and paternal-specific H3K9me3 regions. The numbers indicate the percentage of paternal-specific H3K9me3 regions (top left) and maternal-specific H3K9me3 regions (bottom right) for each developmental stage.
b, Line chart showing the ratio of paternal-specific regions versus maternal-specific regions for epigenetic modifications during mouse pre-implantation development (log2). The dashed lines represent non-consecutive stages. The black dashed line across the 0 mark indicates equal number of paternal and maternal regions.
c, Heatmap showing the dynamics of the H3K9me3 AS score during pre-implantation development. The colours represent the H3K9me3 AS score.
d, The UCSC genome browser view of allelic H3K9me3 signals near the Myoc gene (cluster 2 in c) during development. Signals represent ChIP–seq RPM.
e, The UCSC genome browser view of allelic H3K9me3 signals near the Snrpn gene (cluster 4 in c) during development. Signals represent ChIP–seq RPM.
5 发育过程中逆转座子的转录活化，从DNA甲基化调控到H3K9me3调控
a, Scatter plots for the association between the log2-transformed H3K9me3/input ratio and DNA methylation levels during pre-implantation in promoter and LTR regions. The x axis represents the Pearson’s correlation coefficients (PCCs) and the y axis represents the P values of the two-sided association test (n = 7 biologically independent samples). The blue horizontal line corresponds to P = 0.05. The orange dashed line separates the positive and negative associations. The total number of significant positive (+) and negative (–) correlations (P < 0.05) is shown at the top of each plot.
b, Box plots showing the log2-transformed H3K9me3/input ratio, the log2-transformed H3K27me3/input ratio, the DNA methylation, the ATAC signal and the expression levels of the 50 H3K9me3-increased LTRs during pre-implantation.
c, The UCSC genome browser view of the log2-transformed H3K9me3/input ratio, the absolute DNA methylation level and the normalized RNA-seq read counts on representative LTRs (the MERVL internal sequences, MERVL-int (left) and the IAPEz internal sequences, IAPEz-int (right)).
d, Line charts showing the H3K9me3, H3K27me3, DNA methylation, ATAC signal and expression levels of MERVL and IAPEY3 during pre-implantation.
e, Scatter plots showing the association between the normalized H3K9me3 read counts and the DNA methylation level during pre-implantation in promoter and LTR regions (two-sided association test, n = 7 biologically independent samples). The different colours represent the four major LTR families. The total number of significant positive and negative correlations (P < 0.05) is shown at the top of each plot.
f, Venn diagram shows the overlap between the 50 H3K9me3-increased LTRs and the 28 H3K27me3-increased LTRs.
g, Box plot showing the enrichment of chromatin factor-binding sites on the 50 H3K9me3-increased LTRs and the 8 H3K9me3-decreased LTRs. The y axis represents the Jaccard index between ChIP–seq peaks and LTRs.
6 早期胚胎中，CHAF1A作为核心因子在H3K9me3调控的LTR沉默中起作用
a, Principal component (PC) analysis of H3K9me3 ChIP–seq data (left panel) and RNA-seq data (right panel) for LTRs. Data were obtained at the morula stage in control mouse embryos or embryos with siRNA-mediated KD of the indicated six genes; n = 7 biologically independent samples.
b, Graphs showing the number of LTRs with increased and decreased H3K9me3 signals (left panel) and upregulated and downregulated expression levels (right panel).
c, Heatmaps showing the H3K9me3 and expression levels of 50 H3K9me3-increased LTRs in control and siRNA-KD samples. The colours represent the log2-transformed H3K9me3/input ratio and the normalized RNA-seq read counts scaled by row.
d, Scatter plot showing the changes in H3K9me3 level and expression level of all LTRs in Chaf1a-KD embryos. LTRs with both significant H3K9me3 level and expression level changes are labelled on the graph. The grey points indicate other LTRs.
e, The UCSC genome browser view of H3K9me3 and expression levels on representative LTR (MMERGLN). Signals represent the log2-transformed H3K9me3/input ratio and the normalized RNA-seq read counts.
f, Depletion of H3K9me3-related chromatin factors impaired the development of pre-implantation embryos. The siRNA mix was injected into MII oocytes. The injected oocytes then underwent ICSI to initiate further development. Data are presented as the mean ± s.d. siUbe2i (n = 35, 40, 37), siSetdb1 (n = 34, 30, 24), siZfp809 (n = 33, 34, 20), siSumo2 (n = 30, 31, 40), siChaf1a (n = 30, 30, 28), siTrim28 (n = 25, 26, 36), control (n = 35, 35, 21). n refers to the number of 2-cell stage embryos.
g, Representative images of blastocyst-stage embryos produced from siRNA injection at 4 days after fertilization. This experiment was repeated three times independently with similar results. Scale bar, 100 μm.
7 启动子上的lineage-specific异染色质在植入后胚胎中建立
a, Principal component analysis of promoter H3K9me3, H3K27me3 and H3K4me3 signals at the blastocyst stage and post-implantation embryos (n = 8 biologically independent samples). The red dashed line represents the differentiation lineage after the ICM stage, and the blue dashed line represents the differentiation lineage after the trophectoderm stage.
b, Scatter plots showing the comparison of promoter H3K9me3 signals between two different lineages at 3.5, 6.5 and 7.5 days. The diagonal dashed lines indicate promoters with significant H3K9me3 signal preference (log2-transformed absolute fold change > 0.5). The numbers of genes with lineage-specific H3K9me3 signal are labelled.
c, The UCSC genome browser view of H3K9me3 and H3K27me3 levels on representative genes with epiblast-specific H3K9me3 signals (top panel) and extraembryonic-specific H3K9me3 signals (bottom panel). Signals represent the log2-transformed ChIP/input ratio.
d,e, Right panel, venn diagram shows the overlap between E6.5 and E7.5 epiblast-specific (d) or extraembryonic-specific (e) H3K9me3-marked genes. Left panel, gene ontology analysis of overlapped epiblast-specific (d) or extraembryonic-specific (e) genes. P values were generated using a modified Fisher’s exact test.
Method
ChIP–seq, RNA-seq and bisulfite (BS)-seq data processing and normalization
ChIP–seq reads were aligned to the mouse genome build mm9 using the bwa (v0.7.12) mem command46. Signal tracks for each sample were generated using the MACS2 (v2.0.10.20131216) pile-up function and were normalized to 1 million reads (RPM). To examine the reproducibility of the ChIP–seq experiments, we calculated the correlation of the normalized signal intensity between biological replicates on merged H3K9me3 peaks across all stages. As the replicates were highly correlated to each other (Pearson’s correlation > 0.8), we then pooled the biological replicates together for each stage. H3K4me3 and H3K27me3 ChIP–seq data in mouse pre-implantation embryos from our previous publications were used in the analysis20. To minimize the effect of chromatin structure and sequencing bias, we re-normalized the ChIP–seq signal by input samples. We first divided the genome into 1-kb consecutive bins and calculated the normalized input signal (RPM) for each bin. Regions with extremely low input signal were assigned with a genomic average to prevent over amplification, then all the histone modification signal tracks were divided by the input signal in its corresponding bins and the histone modification/input ratio were log2 transformed to generate the input normalized histone signal tracks. The RNA-seq reads were mapped to the mm9 reference genome using TopHat (v1.3.3)47. Expression levels for all RefSeq genes were quantified to fragments per kilobase million (FPKM) using Cufflinks (v1.2.0), and FPKM values of replicates were averaged48. All the BS-seq reads were first processed using TrimGalore (v0.3.3) to trim adaptor and low-quality reads. Adaptor-trimmed reads were then mapped to a combined genome with mm9 and 48052 lambda sequence using bsmap (v2.89)49. The methylation level of each CpG site was estimated using mcall (v1.3.0)50.
Identification of ChIP–seq peaks and comparison between stages
All the ChIP–seq peaks were called by MACS14 (v1.4.2)21 with the parameters –nomodel –nolambda –shiftsize = 73 over input files. As the H3K9me3 peak number detected at each stage could be affected by the sequencing depths, we use the same number of reads (80 million for the H3K9me3 samples and 20 million for the input samples) when available that were randomly selected from samples of each stage. To define the established and disappeared H3K9me3 domains, we split the H3K9me3 peaks into 1-kb domains and compared them between consecutive stages: both ICM domains and trophectoderm domains were compared with the morula stage and domains of the E6.5 epiblast and extraembryonic ectoderm were compared with the ICM and the trophectoderm separately. Domains with no surrounding domains within 300 bp in the previous stage were defined as established domains, otherwise they were defined as maintained domains. Domains with no surrounding domains within 300 bp in the next stage were defined as disappeared domains. Enrichment of histone modification peaks in promoter, short interspersed nuclear element (SINE), long interspersed nuclear element (LINE) and LTR regions was calculated using observed versus expected probability. The observed probability was calculated using the length of the histone modification peaks that covers the related genomic regions versus the length of the total histone modification peaks, and the expected probability was calculated using the length of the total related genomic regions versus the length of the mouse genome.
Clustering analysis of stage-specific H3K9me3 domains
To classify the stage-specific H3K9me3 domains, we first split the H3K9me3 peaks into 1-kb domains. We merged the domains from MII oocytes to the ICM stage and filtered the domains that cover less than two stages. The input normalized H3K9me3 signal was calculated for each domain. We then performed k-means clustering on the filtered domains, setting k = 7. Cluster 1 was defined as ‘oocyte-specific domains’; clusters 2–4 were merged as ‘cleavage-specific domains’, as they were formed temporarily during the developmental process; and clusters 5–7 were merged as ‘blastocyst-specific domains’. Heatmaps of input normalized H3K27me3 signal and DNA methylation level of each domain were generated according the order of the H3K9me3 domains. Promoters of the genes (defined as ±2 kb around the transcription start site (TSS)) that only overlapped with ‘oocyte-specific domains’ were defined as ‘oocyte-specific genes’, only overlapped with ‘cleavage-specific domains’ were defined as ‘cleavage-specific genes’ and only overlapped with ‘blastocyst-specific domains’ were defined as ‘blastocyst-specific genes’.
Gene ontology analysis
Functional annotation was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) Bioinformatics Resource51. Gene ontology terms for each functional cluster were summarized to a representative term, and P values were plotted to show the significance.
Allele assignment of sequencing reads
To assign each read to its parental origins, we first obtained the SNPs between C57BL strains and DBA strains, as well as C57BL strains and PWK strains from the Sanger Institute (http://www.sanger.ac.uk/science/data/mouse-genomes-project). We then examined all the SNPs in each read that showed high-quality base calling (Phred score ≥ 30). For paired-end reads, SNP information from both reads was summed, and when multiple SNPs were present in the same read, the parental origin was determined by votes from all SNPs. The read was assigned to the allele that at least two-thirds of the total votes supported. The public SNP-trackable H3K4me3, H3K27me3 and ATAC data sets in mouse pre-implantation development were included in the analysis22,27,34.
Definition of lost, inherited and gained regions during fertilization
To analyse the inheritance and establishment of allelic-specific epigenetic modification from gamete to zygote, we first divided the mouse genome into 1-kb consecutive bins and only those bins that were covered by at least 10 SNP-trackable reads were considered. For histone modifications, we calculated the normalized read counts of the gametes, maternal and paternal zygotes on these regions. For maternal signals, regions with a >0.2 MII oocyte signal and a <0.2 maternal zygote signal were defined as ‘lost regions’; regions with both a >0.2 MII oocyte signal and a >0.2 maternal zygote signal were defined as ‘inherited regions’; regions with a <0.2 MII oocyte signal and a >0.2 maternal zygote signal were defined as ‘gained regions’. Similar calculations were made for paternal signals. For DNA methylation, the cut-off for gametes was regarded as 0.7, whereas the cut-off for maternal and paternal zygotes was regarded as 0.5; similar calculations were performed for DNA methylation levels on these regions.
Identification of allelic-specific regions and genes
To identify the allelic-specific regions after fertilization, we first divided the mouse genome into 1-kb consecutive bins, and only those bins covered by at least 10 SNP-trackable reads were considered. For histone modifications, the significance of allele bias was assessed by the binomial test. For DNA methylation, the significance was assessed using the Fisher’s exact test. The AS score was defined as –log10 (P value), which was defined as positive for maternal-specific regions and negative for paternal-specific regions. Allelic-specific regions were identified using a cut-off of 3 (the absolute value). Similar AS scores were calculated for RefSeq genes, with the scores calculated on the promoter regions (defined as ±2 kb around the TSS) and the minimum SNP-trackable reads as 20. Allelic-specific genes were identified using a cut-off of 2 (the absolute value). We obtained 150 imprinted genes from the mouse book (http://www.mousebook.org/mousebook-catalogs/imprinting-resource) and compared them with the allelic-specific genes defined based on H3K9me3, H3K27me3 and DNA methylation data. Paternally imprinted and maternally imprinted genes were analysed separately.
Expression, DNA methylation and histone modification level quantification of repeats elements
To assess the expression level of repeats elements, all the RNA-seq files were re-mapped to the mm9 genome using the STAR aligner software, allowing up to three mismatches and filtering out reads that mapped to >500 positions in the genome52. Mapped files were then processed using the makeTagDirectory script of HOMER with the -keepOne option53. The tag directories of the mapped files were analysed using the analyzeRepeats.pl script of HOMER with the option ‘repeat’ and -noadj. This script adds the reads that map multiple loci to the expression of the repeat class they represent, which were summarized to 1,221 repeat class. The total read counts of each sample were normalized to 1 million, and replicates were averaged for comparison.
To analyse the methylation level and histone modification level of repeat elements, we downloaded the repeat annotations from the UCSC table browser. DNA methylation level and normalized histone modification signals were calculated for each repeat annotations and the values of the same repeat class were summed and then averaged by the number of copies in the genome.
Association analysis of histone modification and methylation level on repeats
We performed association analysis between the DNA methylation level and the input normalized H3K9me3 signal on samples from the oocyte stage to the ICM stage. Pearson’s correlation was calculated for DNA methylation and the input normalized H3K9me3 signal on genes and LTRs, and association tests were performed based on weighted Pearson’s correlation coefficient54. Genes or LTRs with a negative correlation and significant associations (P ≤ 0.05) have increased H3K9me3 signal and decreased DNA methylation level along development and were defined as H3K9me3-increased genes or LTRs; genes or LTRs with a positive correlation and significant associations have both decreased H3K9me3 signal and DNA methylation level along development and were defined as H3K9me3-decreased genes or LTRs. We filtered the genes and LTRs with extremely low H3K9me3 signals (log2-transformed H3K9me3/input ratio of <0.1). Similar calculations were made between the DNA methylation level and the input normalized H3K27me3 signal.
Enrichment of transcription factor-binding sites in LTRs
To calculate the enrichment of transcription factor-binding sites in LTR regions, we obtained the ChIP–seq data of Chaf1a, Sumo2, Trim28, Zfp809, Setdb1, Suz12 and Rest in mESCs from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/)18. All the ChIP–seq data were processed as previously described. The Jaccard index of the transcription factor ChIP–seq peaks and LTRs were calculated using bedtools (v2.20.1)55. The Jaccard index of the same repeat class were summed and averaged by the copy of the class. Enrichment of the transcription factor-binding sites in LTRs was evaluated based on the Jaccard index of different groups.
Identification of LTRs with differential H3K9me3 and expression levels
The H3K9me3 level and RNA-seq read counts of all KD experiments were first normalized by the sequence depth and then compared with control experiments to identify LTRs with differential H3K9me3 and expression levels. We used the Combat function from the R sva package to remove the potential batch effect for the H3K9me3 and RNA-seq samples from different batches56. LTRs with an input normalized H3K9me3 signal of >0 (log2 transformed) in both control and KD experiments and an absolute fold change >0.25 (log2 transformed) were defined as differential H3K9me3 LTRs. LTRs with normalized RNA-seq read counts greater than –4 (log2 transformed) in both control and KD experiments and an absolute fold change greater than 1 (log2 transformed) were defined as differentially expressed LTRs.
Identification of genes with lineage-specific H3K9me3 level
To identify the genes with lineage-specific H3K9me3 level, we calculated the input normalized H3K9me3 level on all RefSeq gene promoters and enhancers (defined as ±10 kb around the TSS). We compared the H3K9me3 signal of the ICM and trophectoderm stage, the E6.5 epiblast and extraembryonic stage, as well as the E7.5 epiblast and extraembryonic stage. Genes with lineage-specific H3K9me3 level were defined as genes with an input normalized H3K9me3 signal of >0 (log2-transformed scaled) in both stages and an absolute fold change of >0.5 (log2 transformed).
Differential gene expression analysis
To perform differential gene expression analysis, we first calculated the read counts of each RNA-seq sample using HTSeq (v0.6.0). Then, the results were fed into edgeR to perform differential analysis. Genes with a Benjamini and Hochberg-adjusted P ≤ 0.05 and a fold change > 1 were defined as differentially expressed between compared stages.
Transcription factor-binding analysis on lineage-specific H3K9me3 peaks around genes
H3K9me3 peaks from epiblast samples (E6.5 and E7.5) that overlapped with previously defined epiblast-specific genes but not presented in extraembryonic-specific genes were defined as epiblast-specific H3K9me3 peaks; H3K9me3 peaks from extraembryonic tissues (E6.5 and E7.5) that overlapped with extraembryonic-specific genes but not presented in epiblast-specific genes were defined as extraembryonic-specific H3K9me3 peaks. To investigate the potential factors responsible for the establishment of lineage-specific H3K9me3, we collected the publicly available ChIP–seq data in mESCs from the GEO database for 168 factors, including multiple transcription factors, histone enzymes and chromatin remodellers. We identified ChIP–seq peaks using MACS2, and for each factor, we removed the peaks with <5-fold enrichment over control and merged the peaks from different data sets to generate the final binding sites. For each factor, we calculated the number of binding sites that overlapped with epiblast-specific H3K9me3 peaks and extraembryonic-specific H3K9me3 peaks separately, the preference of binding between epiblast-specific and extraembryonic-specific peaks was evaluated using fold change and significance was calculated using the Fisher’s exact test, using the total number of epiblast-specific and extraembryonic-specific H3K9me3 peaks as background. To enlarge the potential transcription factor list, we collected 335 mouse transcription factor motifs from the Cistrome Data Collection57 and performed genome-wide motif scanning using BINOCh. The identified motif sites were treated as potential binding sites of transcription factors and similar analyses were made to evaluate the preference of transcription factor-binding sites in lineage-specific H3K9me3 peaks.
Statistics and reproducibility
Error bars in the graphical data represent the s.d. For all the box plots presented in the analysis, the centre represents the median value and the lower and upper lines represent the 5% and 95% quantile, respectively. The number (n) for the box plots has been indicated in the corresponding figure legends. Significance between different groups was determined using one-sided Wilcoxon test, with P < 0.05 considered to be statistical significant. For the association analysis between H3K9me3/H3K27me3 and DNA methylation signal, we choose eight time points (n = 8, MII oocyte, zygote, 2 cell, 4 cell, 8 cell, morula, ICM and trophectoderm) during development, significance was determined using the two-sided association test. We used DAVID online tools to determine the enrichment of genes in each functional term, and the P value was determined based on a modified Fisher’s exact test. The enrichment of transcription factor-binding sites in epiblast-specific/extraembryonic-specific peaks were evaluated using Fisher’s exact test, and the transcription factors with P < 1 × 10–10 were defined as significant candidates.
ChIP–seq for H3K9me3, H3K4me3 and H3K27me3 and RNA-seq were performed two to four times and WGBS was performed once, with the precise number of replicates and mapping quality for these data shown in Supplementary Table 1. For all experiments described above, all attempts at replication were successful, with similar results.
Reporting Summary
Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.
Codes availability
All the analysis was made based on custom python and R codes and can be available upon request.
Data availability
All the ChIP–seq, RNA-seq and WGBS data generated in this study are summarized in Supplementary Table 1 and have been deposited in the GEO database under the accession number GSE97778. H3K4me3 and H3K27me3 ChIP–seq data for pre-implantation embryos were from our previous publication (GSE73952)20. Allelic H3K4me3, H3K27me3 and ATAC-seq data were downloaded from previous publications (GSE66390, GSE71434 and GSE76687)22,27,34. ENCODE mESC H3K9me3 data were downloaded from the ENCODE website (https://www.encodeproject.org/experiments/ENCSR000CFZ/). All other data supporting the findings of this study are available from the corresponding author on reasonable request.