Genome-wide CpG island methylation and intergenic demethylation propensities vary among different tumor sites

Abstract

The epigenetic landscape of cancer includes both focal hypermethylation and broader hypomethylation in a genome-wide manner. By means of a comprehensive genomic analysis on 6637 tissues of 21 tumor types, we here show that the degrees of overall methylation in CpG island (CGI) and demethylation in intergenic regions, defined as 'backbone', largely vary among different tumors. Depending on tumor type, both CGI methylation and backbone demethylation are often associated with clinical, epidemiological and biological features such as age, sex, smoking history, anatomic location, histological type and grade, stage, molecular subtype and biological pathways. We found connections between CGI methylation and hypermutability, microsatellite instability, IDH1 mutation, 19p gain and polycomb features, and backbone demethylation with chromosomal instability, NSD1 and TP53 mutations, 5q and 19p loss and long repressive domains. These broad epigenetic patterns add a new dimension to our understanding of tumor biology and its clinical implications.

Figure 1.

Distribution of genome-wide average CGI and backbone methylation in normal and tumor tissues. (A) Normal adjacent tissues have average CGI methylation around 0.2 and average backbone methylation around 0.8 within narrow limits. (B) Tumor cells have variable degrees of CGI methylation and backbone demethylation. (C) The distributions of abnormal methylations according to tumor types. Tumor names are presented at median levels of CGI and backbone methylation in each tumor type. (D) THCA cases are distributed mostly in normal ranges (NC-NB). Abbreviations: NC, normal CGI methylation (≤0.24); HC, high CGI methylation (>0.24); NB, normal backbone methylation (≥0.78); LB, low backbone methylation (<0.78). (E) LGG cases are distributed mostly in the CGI-methylated zone (HC-LB). (F) Most of COADREAD cases are both CGI-methylated and backbone-demethylated with variable degrees.

Figure 2.

Exemplary tumors showing remarkable association with clinical, pathological and molecular parameters. (A) KICH tumors higher CGI methylation are related with advanced pathological stage, high grade, chromosome 9 loss, high mutation rate, SETD2 and BAP1 mutations, and low activities of FoxM1, Aurora B and PLK1 pathways. (B) STAD tumors with higher CGI methylation are related with high histological grade, high microsatellite instability (MSI-H), gains of 9p and 19p, high mutation rate, PIK3CA and ARID1A mutations, and low activities of p75^NTR, Reelin and Ret pathways. (C) STAD tumors with lower backbone methylation are associated with low histological grade, non-diffuse type, low pathological stage, higher number of somatic copy number alterations (SCNAs), loss of 19p and 5q11, gain of 1q21, TP53 mutation, and low activities of c-Kit, FoxM1 and p53 pathways. (D) In LGG, backbone demethylation is associated high histological grade, astrocytoma histology, temporal lobe location, high number of SCNAs, loss of chromosome 10, gain of chromosome 7, EGFR and PTEN mutations, and low activities of BMP receptor, Ret and EGFR pathways.

Figure 3.

Relation of copy number and somatic mutation with methylation change. (A) By concatenating gene-level associations, copy change at 5q is found to be the most recurrently associated region with backbone demethylation. Color gradients are according to the percentiles of 21 tumor types. Chromatin modifier genes in chromosome 5 are presented. (B) Copy loss of NSD1 at 5q35.2 is significantly correlated with backbone demethylation. All types of tumors analyzed together and copy-neutral tumors were removed from the plot. (C) NSD1-mutated tumors show significant genome-wide backbone demethylation (P = 1.9 × 10⁻¹¹). (D) Bi-allelic NSD1 aberration by mutation and/or copy loss shows more profound demethylation (P < 2.2 × 10⁻¹⁶). (E) With increasing number of chromatin modifier mutation, more backbone demethylation is apparent (P < 2.2 × 10⁻¹⁶).

Figure 4.

Backbone demethylation in large genomic regions. (A) In an exemplary chromosome (chr10) demethylation is profound in large genomic regions that overlap with partially-methylated domains (PMDs), lamina-associated domains (LADs) and late-replicating regions. Values are averaged within each sliding window (2 Mb bin and 1 Mb step). (B and C) When all tumors are analyzed together, high degrees of backbone demethylation and CGI methylation are observed inside LADs.

Figure 5.

Comprehensive model for CGI methylation and backbone demethylation.