Intermediately Methylated Regions in Normal Cells Are Epimutation Hotspots in Cancer

Abstract

DNA methylation is altered in all cancers, but the mechanisms responsible for these changes are not well understood. Using data from 100 primary samples, we show that regions with intermediate methylation in normal hematopoietic cells are hotspots for stable, clonal epimutations in acute myeloid leukemia. Analysis of hematopoietic stem cell clones demonstrated that intermediate methylation at epimutation hotspots reflects random allele-specific methylation and expression at the single cell level, representing a previously unrecognized form of somatically acquired imprinting. We identified somatically imprinted regions in other tissues and show they are epimutation hotspots in other cancer types. This demonstrates that random allele-specific methylation is both a general property of normal cells, and a vulnerability that renders them susceptible to cooperating events and clonal selection during cancer.

Fig. 1.. Recurrent differentially methylated regions in AML are enriched for intermediately methylated regions in normal hematopoietic cells.

(A) Analysis of differentially methylated regions (DMRs) in 100 primary AML samples. WGBS data from each AML was compared to a pooled dataset of 16 hematopoietic cell populations purified from bone marrow aspirates from 7 distinct healthy donors. (B) Scatter plot showing the methylation level of each DMR identified in all AML samples versus mean methylation in normal hematopoietic cells. (C) Recurrence of DMRs across AMLs after merging overlapping regions from all samples. Private DMRs are regions with differential methylation in one AML sample, while DMR hotspots are regions overlapping highly recurrent DMRs identified in at least 25 AML samples. (D) DNA methylation at a DMR hotspot in the HOXA3 locus from normal hematopoietic cells and selected AML samples. (E) Heatmap showing methylation in normal hematopoietic stem/progenitors (HSPCs) at regions overlapping private DMRs identified in only one AML sample (left) and DMR hotspots (25+ samples) (right). DMRs are length-normalized and shown with 1 kb of flanking sequence. (F) Methylation levels at intermediately methylated regions in normal hematopoietic cells that are hotspots for differential methylation in AML samples (IMR hotspots). Each point shows the mean methylation of IMR hotspot in the indicated normal hematopoietic cell population (green), AMLs with no DMR called at the hotspot (blue), and AMLs in which a DMR was called (red). AML samples with a DMR at the hotspots are grouped by the presence of mutations in selected genes (minimum of 2 AML samples with the DMR hotspot per mutation group).

Fig. 2.. Intermediately methylated regions in normal hematopoietic cells display allelic methylation patterns in AML.

(A) Models of possible fragment-level methylation patterns at intermediately methylated regions (IMRs). (B) Comparison of WGBS fragment-level methylation metrics at IMRs overlapping DMR hotspots vs. non-hotspot DMRs in normal hematopoietic cells. Each point represents a different metric, with the x-axis indicating the effect size (standardized difference in means between hotspots and non-hotspots) and the y-axis indicating statistical significance as –log_₁₀(p-value) from a t-test comparing the two groups. Metrics include PDR (Proportion of Discordant Reads), Entropy (Shannon entropy), CHALM (Cell Heterogeneity-Adjusted cLoNal Methylation), R^² (linkage disequilibrium R squared), MHL (Methylation Haplotype Load), MCR (Methylation Concordance Ratio), and MBS (Methylation Block Score). (C) Fraction of IMR hotspots exhibiting allele-specific methylation (ASM) per sample from WGBS in normal hematopoietic cells and AML samples, calculated among all IMR hotspots with a heterozygous SNP and sufficient coverage. Only samples with at least 5 informative IMR hotspots are included. (D) ONT sequencing and haplotype-resolved methylation analysis of normal HSPCs (N=4) and AML samples (N=10). (E) Example of haplotype-resolved methylation in an IMR hotspot (at the promoter of CD1D), showing mosaic methylation in normal hematopoietic cells, allele-specific methylation (ASM) in AML sample 335640, and full methylation (epimutation) in AML sample 740266. Tracks show the DMR region identified in each AML sample vs. normal cells using DSS R package and the “IMR block” defined by correlation of fragment-level CpG methylation using ONT data(34). (F) ONT haplotype-level methylation at IMR blocks in all normal HSPCs (N=4) and AML samples (N=10). Top panels show the distribution haplotype-level methylation at IMR blocks. Bottom panels show the average methylation of haplotype 1 versus haplotype 2 for each IMR block. (G and H) Aggregated H3K4me3 CUT&Tag signal (G) and chromatin accessibility (Fiber-seq) (H) at IMR blocks in AML samples, with blocks grouped by methylation status (hypomethylated, intermediately methylated, or hypermethylated). (I) Chromatin accessibility (Fiber-seq) at IMR blocks with intermediate methylation (green in panel H) and exhibiting ASM in AML samples, comparing accessibility between the methylated and unmethylated haplotypes. (J) Example of haplotype-resolved methylation and chromatin accessibility at an IMR block in an intergenic region on chromosome 10 (chr10:61,866,122–61,869,606).

Fig. 3.. IMR blocks are stable, clonal events in hematopoietic cell populations.

(A) Longitudinal analysis of methylation at IMR blocks in 3 AML patients (112200, 895870, and 627311) who achieved a complete morphologic remission with standard “7+3” induction chemotherapy. ONT sequencing was performed on the samples at the timepoints indicated by the open circles. (B) Heatmap representation of haplotype-level methylation from ONT sequencing of paired presentation/relapse sample pairs. IMR blocks were classified as either having ASM (left) or a fully methylated/unmethylated epimutation (right) via Fisher’s exact test on methylation read counts either between haplotypes (for ASM) or compared to normal HSPCs (for epimutations). Methylated and unmethylated IMR blocks are shown in red and blue, respectively. Only IMR blocks with statistically significant classifications at both presentation and relapse are shown. The percent of concordant ASM and epimutation blocks between presentation and relapse is shown on the right. (C) Haplotype-level methylation at IMR blocks from AML patient 895870 at presentation, complete morphologic and molecular remission (i.e., blasts <1% and no leukemia-associated mutations detected), and relapse. Points show the methylation of IMR blocks in haplotype 2 vs. haplotype 1 and are colored by their methylation status in the presentation sample. Bar plots show the variant allele fraction (VAF) of selected recurrent mutations in each sample. Bottom panel shows haplotype-level methylation at the same IMR blocks from a representative CD34+ HSPC sample and are colored based on the methylation status of the block in the presentation AML sample. (D) Haplotype-level methylation in AML patient 627311 at presentation, complete morphologic remission (bone marrow blasts <1%), but with a persistent DNMT3A, and relapse. Scatter plots are shown as in (C). Bar plots show mutation VAFs in each sample and indicate the presence of a DNMT3A^R729W variant in the remission sample, consistent with persistent clonal hematopoiesis. (E) Correlation between clonal hematopoiesis in AML remission and the presence of ASM and epimutations at IMR blocks from ONT sequencing. Top panels show mutation VAFs at presentation and complete remission for 6 patients, 3 with clonal hematopoiesis in remission (left) and 3 with nonclonal hematopoiesis. Bottom panel shows the fraction of all IMR blocks (N=1,575) that met criteria for being ASM or fully methylated/unmethylated epimutations compared to normal HSPCs in each sample using a Fisher’s exact test on haplotype-level methylation read counts. All remission samples were bone marrow aspirates with normal trilineage hematopoiesis and <1% blasts. See reference (36).

Fig. 4.. Random allele-specific methylation at IMR blocks is established in single hematopoietic stem cells and is associated with allele-specific gene expression.

(A) HSC expansion and sequencing to analyze haplotype-level methylation in single-cell-derived HSC clones. (B) Haplotype-level methylation at IMR blocks in normal donor HSPCs, bulk cord blood HSCs, and single-cell-derived HSC clones. Each point corresponds to the average methylation of one IMR block within a single clone. Only IMR blocks with intermediate methylation (0.3–0.7) in the bulk HSCs were included. (C) Haplotype-level methylation patterns in single-cell-derived HSC clones at IMR blocks exhibiting intermediate methylation in bulk HSCs (top, labelled “IMR Blocks”) and regions showing allele-specific methylation (ASM) in bulk HSCs (bottom left, “Bulk ASM”). In both panels, line plots display the average methylation of haplotype 1 (orange) and haplotype 2 (purple) across all clones for each region. The dot plots below indicate the number of individual clones in which Haplotype 1 or Haplotype 2 was methylated, which is indicated by the Y-axis and the dot color. The bar plot (bottom right) quantifies the percentage of regions where the methylated allele is concordant across clones versus switched (i.e., random) for both IMR blocks and bulk ASM regions. Only regions with ASM observed in at least four clones were included in this analysis. (D) Example of an IMR block exhibiting random methylation patterns in HSC clones. The tracks display haplotype-level methylation at the block that shows intermediate methylation in bulk HSCs. Individual clones demonstrate random allele-specific methylation (ASM) on either haplotype 1 (orange) or haplotype 2 (purple), or an epimutation showing either fully hypomethylated or hypermethylated states. (E) Allele-specific methylation is associated with allele-specific expression in HSC clones at an IMR block located within an internal promoter of the SCHIP1 gene. For several representative clones, tracks display DNA coverage, haplotype 1 methylation (orange), haplotype 2 methylation (purple), and RNA coverage. The bar plots quantify normalized, haplotype-specific expression in Transcripts Per Million (TPM). RNA Reads are assigned to haplotypes based on a heterozygous SNP, which is indicated by the vertical blue/grey (haplotype 1) and red (haplotype 2) lines in the coverage tracks. While DNA coverage is consistently biallelic, the level of RNA expression from each haplotype is inversely correlated with its methylation status, demonstrating that higher methylation is linked to transcriptional repression.

Fig. 5.. Tissue-specific IMRs in normal cells are hotspots for epimutations in cancer.

(A) Identifying methylation changes at tissue-specific intermediately methylated regions (IMRs) in multiple cancers. IMRs for hematopoietic cells were identified using ONT data from this study, while IMRs for other tissues were defined using a normal tissue WGBS atlas(39). For each tissue, epimutations (hyper- or hypo-methylation) were identified by comparing tumor samples to their normal counterparts. Allele-specific methylation (ASM) was called using haplotype-phased ONT data. This analysis integrated data for AML generated in this study with published datasets for other cancers, including ONT(41) and WGBS(42, 44) data. For details, see the Methods section. (B) Heatmap showing the fold enrichment of tissue-specific IMRs within tissue-specific chromatin states. For each tissue (columns), the enrichment of its specific IMRs (n = number of blocks) was calculated for each chromatin state (rows) relative to a genomic background. Values displayed represent the fold enrichment for regions with an enrichment of ≥5. Chromatin state annotations were obtained from the Roadmap Epigenomics Consortium(45). (C) Comparison of Haplotype 1 versus Haplotype 2 methylation at tissue-specific IMRs in cancer samples using ONT data. (D) Percentage of tissue-specific IMRs with retained intermediate methylation (0.3–0.7) that exhibit allele-specific methylation (ASM) per sample in various cancer types, using phased ONT data. Each point represents an individual tumor sample. The box plots summarize the distribution of ASM percentages for each cancer type shown. (E) Distribution of average methylation levels at tissue-specific IMRs in normal tissues (blue) and their corresponding primary tumors (red) from ONT data. The numbers at the top of the plot indicate the number of samples included in each group. Each point represents the average methylation of a single IMR block within a sample, with data from multiple samples pooled for each tissue type. For visualization, 1,000 points were randomly sampled for each group. Data for tumor samples and normal blood were derived from ONT, while data for other normal tissues were from WGBS. (F) Percentage of various genomic features exhibiting epimutations across different cancer types from ONT data for individual primary tumor samples. Epimutations were defined as a significant difference in aggregate methylation via a Fisher’s exact test on methylation counts for each region between tumor and its respective normal. Boxplots show the percentage of tissue-specific IMRs (IMR Blocks), CpG islands, tissue-specific transcriptional start sites (TSS), and tissue-specific enhancers that have significant count-based differential methylation in each tumor sample compared to its corresponding normal tissue.