DNA methylation is strongly associated with cancer, and studies on DNA methylation are likely to shed light on the process of tumorigenesis as well as identify biomarkers. In this study, we used MethylCap-seq, a genome-wide DNA methylation capture method, to analyze differential DNA methylation in 24 CRCs and matched normal colon tissues. We show that the tumor and normal profiles can be clearly distinguished from one another (), and that DMRs are detected for each tumor compared with its matched normal sample. Applying a stringent cut-off, we identified 2,678 and 468 high-confidence hyper- and hypomethylated DMRs, respectively (). However, this approach and the limited cohort size did not allow us to identify statistically significant methylation signatures for subgroups of the analyzed tumor set.
The high-confidence hypermethylated DMRs overlap almost exclusively with CGIs and promoters (; Fig. S6 and S9
), whereas the high-confidence hypomethylated DMRs are mainly focused in intragenic regions and not CGI shores (Fig. S7
Classical DNA methylation analyses have shown overall hypomethylation in (colorectal) cancers, which occur at unique sequences, but especially at repetitive elements.29-31
To detect differentially methylated regions we focused in this study at regions for which we detect DNA methylation above the peak-calling threshold in at least one sample. We did not analyze global hypomethylation because MethylCap-seq is not the best method to study this. Mainly because MethylCap-seq, as well as other capture assays by means of MBD proteins, is biased toward regions with high methylated CpGs content, and hence regions with low methylated CpG density are covered less well.9
The analysis of epigenetic marks in the human embryonic stem cells showed that hypermethylated CRC DMRs overlap for 53% with regions with elevated H3K27me3 (). Notably, hESC bivalent regions are enriched for developmental regulators,24,32
and silencing of these regulators by aberrant DNA methylation could contribute to cancer formation. Possibly, methylation factors are preferentially targeted to bivalent regions in cancer, and in this way reversible gene silencing in stem cells is replaced by permanent gene silencing in cancer.33,34
We performed an integrated analysis of our data with publically available expression data. Generally, we did not observe strong changes in expression between tumor and normal for the genes associated with the high-confidence hypermethylated DMRs (). Indeed, hypermethylation frequently occurred at genes not or lowly expressed in normal colon (). Other studies also reported marginal effects of aberrant DNA methylation on overall gene expression in cancer.5,33,35
The question arises whether these hypermethylated regions are bystander/passengers or causative. It seems safe to assume that lowly expressed genes are “easier” to become methylated than actively transcribed genes, and this can result in sites of non-functional DNA methylation. However, it is also possible that methylation is an important mechanism that ensures a permanent inactive state (a hard switch). In line with this hypothesis, a small fraction of the genes that become hypermethylated in tumors and do not show differences in expression between tumor and normal, do gain expression in the HCT116 DKO compared with the corresponding WT HCT116 cells (). In the DKO cells maintenance of methylation is lost, and this suggests that DNA methylation is important to suppress the expression at these sites.
The majority of the identified high-confidence hypermethylated DMRs (~95%) co-localize with genes that are not known to be cancer-related. However, at least 9% of these regions are at the promoters of transcription factors. For example, seven different members of the Sox protein family (SOX1, SOX2, SOX5, SOX7, SOX11, SOX14 and SOX21) are targeted. It is well known that these and many other transcription factors are involved in development; they turn on/off genes that induce changes in cell morphology, proliferation, interaction, and movement. It is conceivable that deregulation of some of these transcription factors by hypermethylation could contribute to tumorigenesis. In addition, we also find frequent DMRs positioned at genes associated with the extracellular matrix, such as IGFBP3, MMP9, FBN2, NRCAM, NTNG1, GPNMB, GPC6, RELN, COL4 and ITGA4. Since the ECM affects adhesion, motility, viability, and proliferation, deregulation of this pathway could contribute to malignant transformation of colon cells. Functional assays for these genes, such as siRNA-mediated knockdown or overexpression in model cell lines, will be required to clarify their significance and could provide new insights in tumorigenesis.
The 184 DMRs occurring in over 80% of the tumors (support > 19) constitute potential biomarkers for CRC detection. The fact that hypermethylation of these regions was found in different subtypes of CRC suggests that hypermethylation at these sites may occur relatively early in tumorigenesis. Examples of genes that coincide with these DMRs are SDC2 and KCNK12. SDC2 is a cell surface receptor and appears to play diverse and possibly conflicting roles in cancer: SDC2 downregulation was associated with poor prognosis in esophageal squamous cell carcinoma,36
whereas SDC2 overexpression was associated with poor prognosis in prostate cancer tissue.37
KCNK12 encodes a member of the potassium channel superfamily, and was shown to be hypermethylated in 32 out of 77 colorectal tumors.7
Analysis of the DMRs in a large independent cohort of CRC and preferably using an orthogonal analysis method should be performed to evaluate their frequency of hypermethylation in CRC, as well as their clinical value.
In summary, we have performed a comparative and comprehensive genome-wide DNA methylation analysis of 24 CRC and matched normal colon samples. We provide a resource containing genome-wide DNA methylation maps and comprehensive lists of DMRs. From our analyses, it is evident that hypermethylation in CRC does not only occur at tumor suppressor-like genes, but affects a wide spectrum of coding loci. This suggests that only a small fraction of the DMRs plays a direct role in tumorigenesis. Furthermore, this study corroborates and extends the intriguing link between promoter region hypermethylated in tumors and H3K27me3 in human embryonic stem cells.