Bivalent Regions of Cytosine Methylation and H3K27 Acetylation Suggest an Active Role for DNA Methylation at Enhancers

Abstract

The role of cytosine methylation in the structure and function of enhancers is not well understood. In this study, we investigate the role of DNA methylation at enhancers by comparing the epigenomes of the HCT116 cell line and its highly demethylated derivative, DKO1. Unlike promoters, a portion of regular and super- or stretch enhancers show active H3K27ac marks co-existing with extensive DNA methylation, demonstrating the unexpected presence of bivalent chromatin in both cultured and uncultured cells. Furthermore, our findings also show that bivalent regions have fewer nucleosome-depleted regions and transcription factor-binding sites than monovalent regions. Reduction of DNA methylation genetically or pharmacologically leads to a decrease of the H3K27ac mark. Thus, DNA methylation plays an unexpected dual role at enhancer regions, being anti-correlated focally at transcription factor-binding sites but positively correlated globally with the active H3K27ac mark to ensure structural enhancer integrity.

Fig. 1. H3K27ac co-exists with DNA methylation at regular and super-enhancers but not at promoters

Smooth scatter plots of H3K27ac-enriched region quantifications (log₂ of read count) against the average DNA methylation level of the CpG sites within the same regions in HMEC (A), MCF7 (B), PREC (C), PC3 (D), normal colon (E) and HCT116 cells (F). Promoters are defined by H3K27ac peaks within +-1500bp of TSSs, while regular enhancers comprise H3K27ac peaks outside the +-1500bp TSSs but not part of a super-enhancer region. Super-enhancers are defined by H3K27ac peaks or constituents that are not part of regular enhancers or TSSs. White dotted threshold lines have been drawn at 30% and 70% for DNA methylation cutoffs. The cold to warm color scale indicates a low to high density in the scatter plot. Percentages in the squares reflect the percentage of H3K27ac regions that fall between the indicated boundaries. Panels (G, H & I) show genome browser shots of regular (G&H) and super-enhancer (I) regions. The first track shows binding sites of TCF4 in purple. This is followed by a gene, CGI and H3K27ac peak tracks in HCT116 cells. Wiggle plots of H3K27ac ChIP-seq enrichment in HCT116 cells and NOMe-seq data are shown below. The first track of the NOMe-seq heatmap shows HCT116 CpG methylation from low to high (green to red), followed by accessibility from low to high (yellow to blue). Black boxes highlight regions of focal DNA demethylation with high H3K27ac enrichment, chromatin accessibility and TCF4 binding. Pink boxes highlight regions with H3K27ac enrichment and DNA methylation in HCT116 cells. Orange boxes indicate transcription start sites. See also Fig. S1.

Fig. 2. Bivalent chromatin exists at enhancers but not at promoters

(A) Schematic representation of the CDKN2A promoter. H3K27ac enrichment in HCT116 cells is shown in light blue, followed by the CDKN2A gene track in dark blue, a CpG island in green and the region amplified by PCR in black. Input and H3K27ac ChIP DNA was bisulfite sequenced. Every row shows a different clone and every circle a CpG site. White circles indicate unmethylated sites, while black circles are methylated CpG sites. (B) Three consecutive methylated and H3K27ac-marked regions from our genome-wide data within a 2.4kb super-enhancer containing a CGI (green) in the middle were sequenced to eliminate a possible ChIP bias of pulling down methylated regions. Both input and H3K27ac pull-down show nearly 100% methylation in these regions. See also Fig. S2.

Fig. 3. TCF4 binding is reduced at bivalent enhancers

(A) Bar graphs of the percentage of H3K27ac peaks bound by TCF4 in regular and super-enhancers in HCT116 cells. H3K27ac peaks are categorized into three methylation groups according to their average methylation level (<30, 30–70, >70). Peaks bound once by TCF4 are shown in blue, while those bound multiple times, are shown in green. (B) DNA methylation level at +−500bp of the TCF4 binding center in regular and super-enhancers. Every line indicates a TCF4 binding site in HCT116 cells that is maintained in the same order in (C). Every CpG site is indicated on a 0–100% methylation, green to red scale. The order shows the highest average DNA methylation at the top of the scatter plots. (C) Accessibility plots indicate GpC methylation levels from 0–100% accessible on a yellow to blue color scale. See also Fig. S3.

Fig. 4. DNA methylation maintains the H3K27ac mark at regular and super-enhancers

(A) Bar graphs of the percentage of HCT116 H3K27ac peaks maintained in demethylated DKO1 cells for the category of lowly (<30%) and highly (>70%) methylated peaks. Dark blue percentages indicate regular enhancers in DKO1 cells, while light blue ones indicate super-enhancer constituents in DKO1 cells. A significant difference (***, p<0.001) in maintenance between lowly and highly methylated peaks was established in DKO1 cells for regular and super-enhancers. (B) Genome browser shots of regular (left) and super-enhancer (right) regions lost in DKO1 cells. The first track shows binding sites of TCF4 in purple. This is followed by a gene, CGI and called peaks (grey boxes) track in HCT116 and DKO1 cells. Wiggle plots of H3K27ac ChIP-seq enrichment in HCT116 and DKO1 cells with their respective NOMe-seq data are shown below. The DNA methylation NOMe-seq heatmap shows CpG methylation from low to high (green to red). See also Fig. S4B, C, S5F

Fig. 5. H3K27ac remains on the methylated strand of bivalent enhancers and they regain DNA methylation after 5-Aza-CdR treatment

(A–B) Input and H3K27ac ChIP DNA of control and 5-Aza-CdR-treated samples after 14 days of recovery were bisulfite converted and sequenced to show HCT116 regular (A) and super-enhancer (B) regions that are close to 100% methylation. At D14, the input samples have considerably lost their initial DNA methylation and are not marked by H3K27ac in DKO1 cells in our genome-wide data sets. The H3K27ac pull-down samples have barely changed. This shows that H3K27ac only remains on the methylated strand and is lost on the demethylated daughter strand. Every row shows a different clone and every circle a CpG site. White circles indicate unmethylated sites, while black circles are methylated CpG sites. (C) Boxplots of 7460 regular enhancers and 1124 super-enhancer constituents with an initial methylation level >70%. Their average CpG methylation level is shown at day 0 (control sample), 5, 25 and 42 after 5-Aza-CdR treatment. The original DNA methylation level is greatly re-gained at day 25. See also Fig. S5.