Pan-cancer analysis of frequent DNA co-methylation patterns reveals consistent epigenetic landscape changes in multiple cancers

Abstract

Background: DNA methylation is the major form of epigenetic modifications through which the cell regulates the gene expression and silencing. There have been extensive studies on the roles of DNA methylation in cancers, and several cancer drugs were developed targeting this process. However, DNA co-methylation cluster has not been examined in depth, and co-methylation in multiple cancer types has never been studied previously.

Results: In this study, we applied newly developed lmQCM algorithm to mine co-methylation clusters using methylome data from 11 cancer types in TCGA database, and found frequent co-methylated gene clusters exist in these cancer types. Among the four identified frequent clusters, two of them separate the tumor sample from normal sample in 10 out of 11 cancer types, which indicates that consistent epigenetic landscape changes exist in multiple cancer types.

Conclusion: This discovery provides new insight on the epigenetic regulation in cancers and leads to potential new direction for epigenetic biomarker and cancer drug discovery. We also found that genes commonly believed to be silenced via hypermethylation in cancers may still display highly variable methylation levels among cancer cells, and should be considered while using them as epigenetic biomarkers.

Keywords: DNA co-methylation; Epigenetics; Frequent network mining; Pan-cancer methylation.

Fig. 1

Similarity of co-methylation clusters among multiple cancer types, cluster edge frequency distribution among all cancer types, and an example of correlated methylation level within one cluster in COAD. a Clustered heatmap of Jaccard indices among co-methylation clusters in 11 cancer types, for the cancer types with two different methylation data platforms, only the common cluster probe pairs were used for comparison, as indicated by common after cancer type names. b The frequency distribution of the cluster edge (probe-pairs) in all 17 cancer datasets. c The centralized methylation beta values of Cluster 4 probes for all COAD-450 cohort. Colored lines represent different probes

Fig. 2

Top enriched biological functions for Cluster 1, 3 and 4 marked genes using Ingenuity Pathway Analysis (IPA). a Cluster 1 genes enriched biological functions. b Cluster 3 genes enriched biological functions. c Cluster 4 genes enriched biological functions

Fig. 3

Top networks identified with IPA for cluster 3 and cluster 4 marked genes. a Cluster 3 cell signaling, molecular transport, vitamin and mineral metabolism network. Names in red: genes involved in cell signaling and cancer. b Cluster 4 cell signaling and interaction, nervous system development and function, neurological disease network. Names in red: genes involved in cell signaling and neural signal transmission. Grey: genes from Cluster 3 or 4. White: molecules not present in Cluster 3 or 4

Fig. 4

Contributions of edges to the four frequent co-methylation clusters from each of the 11 cancer types. The percentages of shared edges with respect to the total number of edges in each of the four clusters are plotted in the heatmap

Fig. 5

Samples differentiated by the transformed methylation level of Cluster 3 and 4 eigengene. Green: normal samples; Red: tumor samples. a sorted by Cluster 3 eigengenes for the methylation beta value from high to low. b sorted with Cluster 4 eigengenes for the methylation beta value from high to low

Fig. 6

Age average in multiple cancer datasets used for co-methylation cluster mining. Blue: age average for all samples. Red: age average for normal samples

Fig. 7

Patient ages sorted according to the transformed methylation values of the eigengenes for Cluster 1 and Cluster 4 of BRCA and THCA. a patient age sorted according to methylation levels of the eigengene for Cluster 1 from high to low in BRCA. b patient age sorted according to Cluster 4 eigengenes' methylation level high to low in BRCA. c patient age sorted according to Cluster 4 eigengenes' methylation level high to low in THCA