ATAC-Me Captures Prolonged DNA Methylation of Dynamic Chromatin Accessibility Loci during Cell Fate Transitions

Abstract

DNA methylation of enhancers is dynamic, cell-type specific, and vital for cell fate progression. However, current models inadequately define its role within the hierarchy of gene regulation. Analysis of independent datasets shows an unanticipated overlap between DNA methylation and chromatin accessibility at enhancers of steady-state stem cells, suggesting that these two opposing features might exist concurrently. To define their temporal relationship, we developed ATAC-Me, which probes accessibility and methylation from single DNA library preparations. We identified waves of accessibility occurring rapidly across thousands of myeloid enhancers in a monocyte-to-macrophage cell fate model. Prolonged methylation states were observed at a majority of these sites, while transcription of nearby genes tracked closely with accessibility. ATAC-Me uncovers a significant disconnect between chromatin accessibility, DNA methylation status, and gene activity. This unexpected observation highlights the value of ATAC-Me in constructing precise molecular timelines for understanding the role of DNA methylation in gene regulation.

Keywords: DNA methylation; cellular differentiation; chromatin accessibility; enhancers; epigenetics.

Figure 1.. ATAC-Me accurately measures the chromatin accessible methylome.

A. Methodology overview of ATAC-Me library construction and experimental design. B. Spearman-rank comparison of standard ATAC and ATAC-Me read counts at 0hrs or 24hrs PMA stimulation within a common set of broadpeak accessible regions. C. Genome browser profile of the IL8 locus displaying ATAC-seq accessibility signal (gray), ATAC-Me accessibility signal (blue), CpG methylation frequency (green), and CpG read abundance (black). All data shown have been read depth normalized. D. Scatterplot of mCpG fraction in THP1 cells calculated with WGBS or ATAC-Me data for non-allelic CpGs (Pearson = 0.92, p < 0.0001, 0hr; Pearson = 0.91, p < 0.0001, 24hr). E. CpGs within validated allele-specific methylated (Pearson = 0.80, p < 0.0001, 0hr; Pearson = 0.80, p < 0.0001, 24hr) regions. The CpG subset that displayed the expected intermediate methylation state for AMR CpGs as calculated from WGBS data was not correlated with ATAC-Me CpG methylation (Pearson = 0.02, p = 0.59, 0hr; Pearson = 0.02, p = 0.64, 24hr). See also Figure S1,S2,S3.

Figure 2.. Dynamic chromatin accessibility and transcription factor motif enrichment during PMA-stimulated gene activation.

A. C-means clustering of ATAC-Me time course accessibility data reveals multiple categorizations of dynamic accessibility behavior. Membership indicates the goodness of fit for a region in a particular cluster. B. Heatmap of ATAC-Me accessibility signal at each time point for the dynamically accessible regions identified with corresponding C-means clustering. C. UCSC genome browser non-contiguous, “Multi-region”, view of multiple dynamic loci across time. Colored arrows indicate dynamic peak behaviors: Transient Response (yellow), Late Response (blue), Gradual Opening Response (red), Gradual Closing Response (gray). D. P-value significance scores for select transcription factor motif enrichment within accessibility clusters. Motif enrichment p-values were calculated with HOMER. Sequence logos for enriched motifs are depicted to the right of motif enrichment plots. See also Figure S4, S5, S6, S7.

Figure 3.. Genomic regions with rapid chromatin accessibility dynamics exhibit prolonged methylation.

A. Density histogram of average methylation fraction for static accessible regions (no membership to a TCseq cluster) across the time course, regions with decreasing accessibility signal across time (Gradual Closing Response), and regions with increasing accessibility signal (Late + Gradual Opening + Transient + Early Persistent Responses identified with TCseq). B. Heatmaps display mCpG fraction across individual peak regions in different accessibility groups (Static, Gradual Closing Response, Transient Response, Early Persistent Response, Gradual Opening Response, and Late Response). mCpG fraction is calculated in 50bp bins across regions scaled to 600bp with a flanking region of +/− 300bp. C. CpG density and GC content of TCseq cluster regions dynamic for chromatin accessibility across time. CpG density was calculated as observed/expected occurrence of CpG dinucleotides within TCseq cluster regions defined in Figure 2. Dashed lines represent CpG island thresholds defined by Gardiner-Garden and Frommer (1987) for CpG density/GC content. See also Figure S7.

Figure 4.. Transcriptional changes track with chromatin accessibility behaviors independently of methylation.

A. KEGG pathway enrichment of genes proximal to dynamic peaks in specific TC-seq clusters. B. K-means clustering of the top 20% most variable genes identified with RNA-seq. C. Read count plots across time for select TFs known to be related to the monocyte to macrophage transition. Depicted values are a replicate average of DESeq2 normalized read counts. D. Hierarchical clustering of the top 20% most variable genes in nearby proximity to genomic loci identified as dynamic for chromatin accessibility. The mean mCpG fraction of all timepoints for neighboring ATAC-Me peaks is indicated to the right of each plot. Black bars below the plots indicate replicate pairs for each time point.

Figure 5.. Integrating the dynamics of chromatin accessibility, DNA methylation and transcription.

A. Scatterplot of RNA-seq read abundance for the top 20% most variable genes associated with a dynamic chromatin accessible locus plotted against neighboring dynamic chromatin accessible locus read counts (left, Pearson = 0.35, p-value < 0.0001) or average CpG DNA methylation (right, Pearson = −0.11, p-value < 0.0001). Scatterplots represent all time points (0hr, 0.5hr, 1hr, 2hr, 24hr) plotted simultaneously. B. Boxplot distributions of standardized difference across time for all dynamic loci defined by TC-seq. Standardized difference across time was calculated for normalized ATAC read counts, rlog(read counts) of neighboring transcripts (top 20% most variable), and the DNA methylation state of the chromatin accessible DNA fragments. Lines represent median values.

Figure 6.. Loss of DNA methylation is delayed in nascent open chromatin regions

A. Heatmap of ATAC-Me accessibility signal for extended time points across all genomic exhibiting accessibility by 24hrs (Early Persistent Response + Gradual Opening Response + Late Response). B. Heatmaps display CpG methylation levels in 50bp bins across all regions in (A) C. Boxplot distribution comparisons of methylated CpG fractions across extended time points and treatments. Wilcoxon rank sum test was used for statistical comparison (ns = not significant). D. Boxplot distributions of standardized difference across early and late time points for all genomic regions opening at 24hrs (A and B). Standardized difference across time was calculated for normalized ATAC read counts and CpG methylation fraction within chromatin accessible fragments. Lines represent median values. See also Figure S7.