We performed whole genome bisulfite sequencing (WGBS) to obtain unique DNA methylation data sets for five normal tissues and eight associated cancer samples (Table 1). Normal samples (n = 5) included brain, blood (CD19+), breast, lung and colon specimens. In order to enable the analysis of DNA methylation variance from different perspectives, we produced references data sets for cancer samples that involved both primary tumors (n = 2) and cancer cell lines (n = 6). These included a donor-matched primary colon triplet (normal tissue, primary cancer, liver metastasis) and matched primary and metastasis breast cancer cell lines, enabling us to analyze changes during tumor progression. The epigenetic peculiarities that could be present in cancer cell lines were addressed through replication experiments in an additional set of 78 normal tissue samples and 714 primary tumors using the HumanMethylation450 BeadChip (Table 2). The obtained data were also validated using the DNA methylation microarray profiles available for 208 normal samples and 675 primary tumor samples in The Cancer Genome Atlas (TCGA) projects (Table 2) [16–18].
Aligning uniquely mapping bisulfite sequencing reads (mean ~480 million reads per sample) of the original 13 samples undergoing whole genome single-nucleotide resolution analysis resulted in a median genomic coverage of 11.1× (14.1× CpG coverage) per sample. Consistent with previous reported results, apart from bimodal DNA methylation levels at promoter sites, the genomes presented high methylation levels, which were globally reduced in cancer samples (Table S1 and Figure S1 in Additional file 1) [5, 6]. To estimate the relationship between super-enhancers and DNA methylation levels, we determined DNA methylation profiles for enhancer regions within their respective tissue types. From the super-enhancers previously described in our normal tissue types through the histone modification H3K27ac (identified as a superior and sufficient mark for the identification of super-enhancers ), we could examine 99.3 % (5128 of 5163; >50 % CpGs covered; Table S1 in Additional file 1) using our WGBS data. We found significant enrichment of unmethylated DNA sequences within the super-enhancers compared with the flanking genomic regions (Fisher’s exact test, odds ratio (OR) 5.6, p < 0.001), supporting the relevance of the features in the here interrogated context. In particular, the edges of the enhancers were CpG-unmethylated, clearly marking the boundaries of the regulatory regions (Fig. 1a, b), a phenomenon that was consistent throughout the analyzed tissue types (Figure S2 in Additional file 1) and that could not be observed in traditional enhancers (Figure S3a, b in Additional file 1) . Moreover, super-enhancers were significantly more hypomethylated than traditional enhancers (Fisher’s exact test, OR 1.8, p < 0.001), further supporting DNA methylation to specifically indicate functionality in this enhancer subtype.
The fact that super-enhancer edges show lower DNA methylation levels compared with their center could be related to an enrichment of transcription factor binding sites at the extreme parts of the regions (Fisher’s exact test, OR 5.33, p = 1.0 × 10−11; Figure S3c in Additional file 1) . Indeed, DNA hypomethylation and transcription factor occupancy revealed a significant relationship (Fisher’s exact test, OR 11.3, p = 2.2 × 10−16; Figure S3d in Additional file 1), consistent with previous reports describing a co-dependency of both regulatory mechanisms [20, 21].
The extent of tissue-specific DNA methylation differences in the super-enhancer regions was low, with only 12.6 % (644 out of 5111) of them showing CpG methylation differences from different normal tissues (δ hypomethylated regions (HMRs) occupancy >10 %; Supplementary methods, Figure S4a and Table S2 in Additional file 1). We assessed variance in super-enhancer DNA methylation profiles by differential analysis of HMRs, focal sites of low DNA methylation levels that mark active regulatory loci [22–24], to account for the high heterogeneity at the large genomic regions represented by super-enhancers. Remarkably, tissue-specific HMRs at breast and blood super-enhancers were significantly enriched in specific transcription factor binding within the respective tissues, as measured by the occupancy of ten commonly profiled factors determined in CD19+ (GM12878; Fisher’s exact test, OR = 2.81, p < 0.001) and breast cells (MCF7; Fisher’s exact test, OR = 1.64, p = 0.007) . Moreover, super-enhancers with tissue-specific DNA methylation levels in breast and brain samples were enriched at promoter regions compared with non-specific super-enhancers, in contrast to previous results that suggest tissue-specific DNA methylation to be enriched in cis-elements (Fisher’s exact test, OR 6.64, p < 0.001 and OR 1.74, p = 0.018, respectively; Figure S4b in Additional file 1) . The sample with the greatest DNA methylation difference compared with normal tissues was that of the CD19+ cell-related super-enhancers (ANOVA, p < 0.001; Figure S4c in Additional file 1), which was the only representative of a non-solid tissue type. It is of note that the presence of tissue-specific DNA methylation in this minor fraction of super-enhancers could be validated by genome-scale analysis using DNA methylation microarrays (HumanMethylation450 BeadChip). Of the normal tissue-derived super-enhancers, 75.5 % (486 of 644) were represented by at least three probes, in a unique set of 78 normal samples (Table 2), representing the analyzed tissue types, of which 71.4 % (347 of 486) showed significant difference between the respective tissue types (Student’s t-test, false discovery rate (FDR) < 0.05; Figure S4d and Table S3 in in Additional file 1). As examples of super-enhancer tissue-specific DNA methylation we can cite the genes encoding the RNA-binding protein QKI (involved in myelinization and oligodendrocyte differentiation), which is unmethylated in white brain matter but heavily methylated in all other normal tissues (Figure S5a in Additional file 1), and lymphoblastic leukemia-associated hematopoiesis regulator 1 (LYL1; plays a role in blood vessel maturation and hematopoiesis), which is unmethylated in CD19+ cells but hypermethylated in all other normal tissues (Figure S5b in Additional file 1).
From the 5111 super-enhancers studied we established four categories based on their average DNA methylation levels (Fig. 1b, c). Remarkably, we determined striking differences between DNA methylation profiles at super-enhancers, ranging from fully hypermethylated to completely unmethylated (Fig. 1d). Moreover, focal hypomethylated regions pointed to spatial differences in DNA methylation within super-enhancers, suggesting local variability in their activity. Accordingly and in contrast to previous assumptions, the focal variability of the here studied epigenetic mark supports the action of independent regulatory units and challenges the conjoint activity of enhancer clusters for this subset of super-enhancer regions.
From an epigenetic perspective, the CpG unmethylated status was significantly correlated with H3K27ac occupancy (Spearman’s correlation test, rho 0.535, p < 0.001; Fig. 1e) and, to a lesser extent, with H3K4me1 (Spearman’s correlation test, rho 0.278, p < 0.001), further supporting the former mark as sufficiently bookmarking super-enhancer functionality. This association was independent of the local CpG density, suggesting a sequence-independent connection between the two epigenetic marks (multivariate linear model, p < 0.001; Figure S6 in Additional file 1). Most importantly, unmethylated status was significantly associated with increased transcriptional activity of the regulated target genes, indicating that DNA methylation levels at these sequences may be of value as surrogate marks of super-enhancer functionality (Spearman’s correlation test, rho −0.77, p < 0.001; Fig. 1f). Although, functional DNA methylation variance at enhancer sites has been reported previously [25–28], we observed a stronger effect of differential DNA methylation on gene expression levels of super-enhancer-related targets (Figure S7a in Additional file 1). It is of note that the increased correlation between DNA methylation and gene expression at super-enhancers compared with traditional enhancers was observed for enhancer sites overlapping promoter regions and those distal to the target gene transcription start site (TSS), suggesting an elevated effect of differential super-enhancer DNA methylation independent of the distance to its target (Figure S7a in Additional file 1). Moreover, DNA methylation levels at super-enhancers overlapping promoters showed significantly higher correlation at regions flanking the proximal (±2 kb of the TSS) promoter (Spearman’s correlation test, rho 0.26 versus 0.18), further suggesting that enhancer-specific dynamics drive gene regulation. It is noteworthy that we did not observe a correlation between super-enhancers and target promoter-related CpG island DNA methylation levels (Spearman’s correlation test, rho 0.0001, p = 0.99), although both genomic features independently correlated significantly with gene expression (Spearman’s correlation test, rho 0.31, p < 0.001 and rho 0.16, p < 0.001, respectively), suggesting an independent function of both regulatory elements. Furthermore, the effect of enhancers on gene expression was closely related to the enhancer size, with DNA methylation levels at super-enhancers presenting the highest correlation with target gene expression compared with smaller sized counterparts (Figure S7b in Additional file 1).
For cis-acting super-enhancers, we observed that the assignment of the closest gene as target resulted in better correlations between super-enhancer DNA methylation and gene expression than a chromatin conformation-based method (ChIA-PET Pol2 in MCF-7 cells, Spearman’s correlation test, rho −0.048, p = 0.4; Figure S7c in Additional file 1) . However, both strategies clearly include falsely assigned enhancer–target pairs and more suitable methodologies have yet to be defined.
Aberrant DNA methylation profiles of super-enhancers in human cancer
Considering the association between DNA methylation status and super-enhancer activity in normal tissues, we wondered whether the observed epigenetic pattern was significantly altered in human cancer. We observed that 14 % (727 out of 5111) of the super-enhancers studied underwent CpG methylation changes in their respective human tumor types, e.g., normal breast versus breast cancer cell lines (Fig. 2a). The most common DNA methylation shift was the loss of CpG methylation in the cancer sample, which was noted in 75.4 % (548 of 727) of cases, whilst 24.6 % (179 of 727) of super-enhancers gained DNA methylation across the eight tissue-matched cancer samples (δ HMR occupancy >25 %; Fig. 2a; Figure S8a and Tables S4 and S5 in Additional file 1). Interestingly, the hypomethylation events were rather unspecific, as they were associated with the global loss of DNA methylation usually observed in cancer samples (paired t-test, p > 0.05) [5, 6, 30], the only notable exception being colorectal tumors, in which they were significantly super-enhancer locus-specific (average flanking regions versus super-enhancer reduction 29.8 % [tumor] and 33.9 % [metastasis], paired t-test, p < 0.001; Figure S8b and Table S5 in Additional file 1). Thus, to determine functional epigenetic alterations, we decided to initially focus on the hypermethylated events, which were enriched in genes associated with transcriptional and metabolic processes and angiogenesis (FDR < 0.01; Table S6 in Additional file 1). Importantly, hypermethylation events were also replicated using DNA methylation microarray analyses in a unique cohort of 714 primary cancer samples (Table 2 and Fig. 2b), where 58.1 % (68 of 117) of the interrogated DNA hypermethylation events at super-enhancers were confirmed (Student’s t-test, FDR < 0.05; Fig. 2c; Table S7 in Additional file 1). These results further suggest that the hypermethylation events observed in the cancer cell line models are mirroring altered DNA methylation profiles at super-enhancer regions in primary tumors. Hypermethylated super-enhancers in cancer included genes previously related to cellular transformation (e.g., CIC, FOXA2, FOXP1, RUNX1 and TBX3) . Importantly, we excluded that copy number variations (CNVs) have confounded our analysis of the primary cancer samples by detecting significant differences in DNA methylation levels between normal and CNV samples in only a very minor fraction of the super-enhancers (4.3 %, 5/117; Student’s t-test, FDR < 0.05; Table S7 in Additional file 1).
It is of note that, using oxidative bisulfite (ox-BS) treatment coupled with DNA methylation microarray analyses, we could exclude the gain of DNA methylation observed in cancer to be due to an increase of 5-hydroxy methylation (5-hmC), a specific cytosine modification that confounds with 5-methylation (5-mC) in bisulfite (BS)-based analyses and found to be enriched in traditional enhancer regions (Figure S9 in Additional file 1) . In order to test a significant contribution of the 5-hmC to the methylation gain in super-enhancers, we compared the methylation values obtained from BS-treated against ox-BS-treated cancer samples, enabling us to estimate the 5-hmC levels . With the alternative hypothesis being that the ox-BS values were greater than 0, we did not observe a significant presence of 5-hmC in any cancer sample (paired one-tailed Wilcoxon test).
To further elucidate the functional consequences associated with the identified cancer-specific super-enhancer DNA methylation shifts, we investigated the impact of the tumor-associated gains of super-enhancer DNA methylation on gene expression. We first used a breast cancer model that included the paired breast cancer cell lines MDA-MB-468PT (derived from the primary tumor) and MDA-MB-468LN (derived from a lymph node metastasis) and the untransformed immortalized breast epithelial cell line MCF10A, associating differential gene expression (RNA sequencing, RNA-seq) with super-enhancer DNA methylation levels. As has been observed for the proximal regulatory gene regions, where a general repressive effect of DNA methylation is widely recognized , we found an association between DNA methylation gain in breast super-enhancer regions and gene repression of the associated genes for both MDA-MB-468PT (Spearman’s correlation test, rho −0.25, p = 0.026) and MDA-MB-468LN (Spearman’s correlation test, rho −0.3, p = 0.002; Fig. 2d) cell lines.
We extended these observations to primary breast tumors from the TCGA , whose expression patterns have also been determined by RNA-seq. We confirmed the significant association between the DNA methylation gains of super-enhancers identified in our breast cancer cell line data set and gene repression observed in the matched TCGA breast cancer samples (Spearman’s correlation test, rho −0.24, p = 0.01; Fig. 2e). Interestingly, the super-enhancers that became hypermethylated in breast cancer were those that, in normal breast epithelial cells, were the most enriched in the H3K27ac histone mark (Spearman’s correlation test, rho 0.2, p < 0.001; Fig. 2f), which defines these particular distal regulatory regions [8, 11, 13], and the H3K4me1 enhancer mark (Spearman’s correlation test, rho 0.2, p < 0.001). Remarkably, the most hypermethylated super-enhancers had also the highest level of expression for the respective associated genes in normal breast epithelial cells (linear slope 1.23, p < 0.001; Fig. 2g).
We were able to validate the link between cancer-specific super-enhancer hypermethylation and the transcriptional inactivation of the corresponding genes beyond the breast tumor type. In the lung tumorigenesis samples from the H1437 (lung adenocarcinoma) and H157 (lung squamous cell carcinoma) cancer cell lines, we found evidence that lung super-enhancer gain of DNA methylation was associated with the downregulation of the target genes (linear slope −3.06, p < 0.001 and −2.09, p = 0.004, respectively; Figure S10a, b in Additional file 1) determined by publically available expression microarrays . We also extended these findings to primary lung adenocarcinoma and lung squamous cell carcinoma tumors from the TCGA , in which expression of the candidate genes originates from RNA-seq experiments. In this setting, we observed a significant association between lung super-enhancer hypermethylation identified in our lung cancer cell lines and gene downregulation found in the matched primary lung cancer samples (Spearman’s correlation test, rho −0.19, p = 0.012 and rho −0.25, p < 0.001, respectively; Figure S10c, d in Additional file 1). The significant association between cancer-specific DNA methylation of super-enhancers and gene repression was also noted in the glioblastoma cell line U87MG (Spearman correlation test, rho −0.26, p < 0.001; Figure S10e in Additional file 1), in which we performed an expression microarray experiment. Thus, the results overall suggest that a tumor-related gain of DNA methylation in super-enhancers has a transcriptionally repressive effect on the corresponding related genes.
We next considered the commonality among different tumor types within super-enhancer DNA methylation changes, and the type of genes and pathways affected by these aberrant epigenetic shifts. We first observed that within regions of commonly hypomethylated super-enhancers in normal contexts, the cancer samples (Table 2) clustered by tumor type (Fig. 3a), a phenomenon we previously identified for DNA methylation events in proximal promoters among distinct human tumors . Interestingly, despite the clear presence of super-enhancer DNA methylation that is associated with the cancer type, there are hypermethylated super-enhancers shared by common epithelial tumors such as the breast and lung samples (Figure S11a in Additional file 1). This is the case for the super-enhancer of the tumor suppressor microRNA MIRLET7, where hypomethylation of the super-enhancer was diminished by a gain of CpG methylation in a fraction of the regulatory region (Fig. 3b, c; Figure S11b, c in Additional file 1). It is of note that the large highly hypomethylated super-enhancer regions displayed focal gains in DNA methylation in cancer, suggesting that distinct segments might exhibit specific functions in healthy and cancer contexts. Consistent with the suspected regulatory function, hypermethylation of the MIRLET7-associated super-enhancer region was associated with transcriptional silencing of MIRLET7B and MIRLET7A3, two family members coded within the affected pri-microRNA (Figure S11d in Additional file 1). Moreover, microRNAs MIRLET7B and MIRLET7A3 were repressed in primary breast carcinomas (TCGA ; Wilcoxon test, p = 0.001 and p = 0.033, respectively) and lung adenocarcinomas (TCGA ; Wilcoxon test, p < 0.001 and p < 0.001, respectively) (Figure S11e, f in Additional file 1) and hypermethylation at super-enhancers was significantly correlated with microRNA repression in breast carcinomas (Spearman correlation test, rho −0.4 and −0.42, p < 0.001 and p < 0.001, respectively) and lung adenocarcinomas (Spearman correlation test, rho −0.47 and − 0.3, p < 0.001 and p < 0.001, respectively) (Fig. 3d, e).
Cancer-specific super-enhancers coincide with regional hypomethylation
Until now we have focused our attention on those sequences described as being super-enhancers that ensure cell and tissue identity in normal tissues [8, 11]. However, a new class of super-enhancer sequences has recently been described that only play this de novo regulatory role in transformed cells to drive the cancer phenotype and its associated hallmarks [11, 13, 37]. We examined the DNA methylation changes occurring in the super-enhancers of colorectal cancer (HCT-116, n = 387), in which we obtained 99 % coverage using our WGBS approach. We observed that these newly developed tumor-related super-enhancers were associated with DNA hypomethylation events (n = 23, δ HMR occupancy >25 %) at these sequences in the transformed cells compared with normal colorectal mucosa (Fig. 4a; Table S8 in Additional file 1). Most notably, the super-enhancer hypomethylation shift was independent of the global loss of DNA methylation generally found in cancer cells (paired t-test, p < 0.001) [5, 6, 30] and rather represented a focal DNA demethylation event within the super-enhancer regions (Figure S12 in Additional file 1). As we did with the aforementioned normal tissue super-enhancers, we validated the DNA hypomethylation changes in these de novo cancer super-enhancers using a cohort of matched normal colon and primary colorectal tumors (TCGA , n = 41) analyzed by DNA methylation microarrays (Fig. 4a; Table S8 in Additional file 1). Noteworthy, we again excluded potential biases included by CNV in these regions (Table S8 in Additional file 1). In this setting, we further confirmed that the loss of DNA methylation in these emerging cancer super-enhancers was significantly associated with an increase in expression of the corresponding regulated genes in the primary colon tumors in comparison with the matched normal colon mucosa (TCGA ; Spearman’s correlation test, rho −0.18, p = 0.009; Fig. 4b). Examples within the most hypomethylated cancer super-enhancers include those sequences regulating the MYC and RNF43  oncogenes (Fig. 4c; Figure S13a, b in Additional file 1), regions not affected by CNV in the primary colorectal cancer sample analyzed by WGBS (Table S8 in Additional file 1). Importantly, DNA methylation changes affected solely regions specifically marked by H3K27ac in colon cancer and widely excluded H3K4me3, further indicating that alterations in super-enhancers occur predominantly distal to the core promoter regions (Fig. 4c).
An interesting matter arising from these results is their value for identifying putative mechanisms that create such specific patterns of oncogenic super-enhancer hypomethylation. It has been proposed that the availability and binding of transcription factors (TFs) to regulatory regions might be able to impact on the DNA methylome and that it is not the transcriptional activity per se that alters the DNA methylation profile of regulatory elements [20, 21]. Herein, we have studied the putative enrichment of TF binding sites in these colorectal cancer-specific hypomethylated enhancers and we observed a significant enrichment for specific TF binding motifs (Figure S14a in Additional file 1). From these factors, specifically FOXQ1 (forkhead box Q1; p = 0.013), a member of the FOX gene family that is involved in tumorigenesis , was the most overexpressed TF in primary colorectal cancer samples and showed multiple binding sites (Table S8 in Additional file 1) and a significant enrichment at hypomethylated super-enhancer loci (Figure S14b in Additional file 1). In relation to this point, FOXQ1 had a 73-fold greater expression in primary colorectal cancer samples than in matched control samples (TCGA ; Wilcoxon test, p < 0.001; Fig. 4d). Furthermore, the stronger FOXQ1 expression was significantly associated with hypomethylation of the previously defined super-enhancers (linear slope −3.74, p = 0.008; Fig. 4e) and the activation of associated target genes (linear slope 0.14, p < 0.001; Fig. 4f), such as the well-known oncogenes MYC and RNF43 (Figures. S15a, b and S16a, b in Additional file 1). Interestingly, the presence of cancer-specific super-enhancer hypomethylation and the tumorigenic effect mediated by the presence of FOXQ1 binding sites could be useful for identifying new candidate oncogenes, such as GPRC5A (G protein-coupled receptor, class C, group 5, member A; Figures. S13c, d, S15c and S16c in Additional file 1), which, by mediating between retinoid acid and G protein signaling pathways, has a role in epithelial cell differentiation .
Importantly, we experimentally validated the association between FOXQ1 expression and target gene regulation in a colorectal cancer cell line model system (HCT116 and SW1116 cancer cell lines). Initially, we confirmed the occupancy of FOXQ1 at binding sites within the super-enhancer regions of the previous described target genes MYC, RNF43 and GPRC5A (Figure S17a in Additional file 1). Furthermore, following small hairpin RNA (shRNA)-mediated knockdown of the TF, we observed significant downregulation of MYC, RNF43 and GPRC5A, suggesting a direct regulatory role of FOXQ1 (Figure S17b in Additional file 1). In line with the oncogenic role of FOXQ1 targets in colorectal cancer settings, knockdown of the TF reduced cell proliferation of the colorectal cancer cell line (Figure S17c in Additional file 1). Remarkably, in addition to FOXQ1, we could also experimentally confirm the regulatory effect of other enriched TFs, whose expression correlated significantly with super-enhancer hypomethylation level (p < 0.05; Figure S14b in Additional file 1). Specifically, we experimentally confirmed the regulatory effect of the TFs HNF4A and PPARG on RNF43 and GPRC5A expression (Figure S18a, b in Additional file 1). Herein, knockdown of the TFs repressed RNF43 and GPRC5A expression (Figure S18c in Additional file 1) and resulted in reduced cell viability (Figure S18d in Additional file 1), further supporting the accuracy of the functional prediction based on super-enhancer DNA methylation levels (Figure S14b in Additional file 1).
Further, we were interested if disruption of the super-enhancer structure would interfere with the DNA methylation levels in the respective regions. Therefore, we treated the colorectal cancer cell lines HCT116 and SW1116 at sub-lethal concentrations with the BET-bromodomain inhibitor JQ1, a small molecule targeting BRD4, a key component of the secondary super-enhancer structure (Figure S19a, b in Additional file 1) . Interestingly, although the treatment with JQ1 decreased the expression of super-enhancer gene targets, such as MYC, RNF43 or GPRC5A, we could not detect an effect on DNA methylation levels at super-enhancer-related CpG sites (Figure S19c, d in Additional file 1). The lack of DNA methylation variance following JQ1 treatment suggests that the secondary super-enhancer structure per se is not a determinant of DNA methylation profiles, but that it is the binding of TFs to the DNA that locally establishes CpG methylation levels.
Large-scale hypomethylation marks potential cancer drivers
Finally, we wondered whether DNA methylation data obtained from WGBS could be used to identify new candidate cancer regulatory regions beyond the histone-based super-enhancer loci [8, 11]. In line, extended hypomethylated regions were previously established as important regulatory elements in hematopoietic cells with a function in leukemogenesis . To test this hypothesis, we ranked all the de novo formed hypomethylated DNA regions (<20 % average DNA methylation) in our colorectal cancer samples by size, having shown above that HMRs in colorectal tumorigenesis presented locus-specific properties (Figure S8b and Table S5 in Additional file 1). In this setting, we did observe an unequal distribution of HMR sizes, as previously reported for the super-enhancer-defining mark H3K27ac (Fig. 5a). Importantly, these large HMRs were mutually exclusive to the presence of super-enhancers in the respective regions, suggesting they represent an independent epigenetic feature to histone defined regulatory elements. Intriguingly, large HMRs mainly spanned gene promoter regions (22/26; Table S9 in Additional file 1), a phenomenon previously described for genes activated in medulloblastoma patients, where an extensive expanded hypomethylation beyond the proximal promoter was observed, which might be a general feature of cancer-related gene activation . Further, most of the HMRs that were present only in the metastatic cancer samples presented features suggesting a role in tumorigenesis. For example, the largest observed HMR (34.1 kb) in the metastatic colorectal cancer sample corresponded to beta-catenin (CTNNB1), a key component of the WNT pathway and driver of epithelial–mesenchymal transition (Fig. 5b) . AXIN2, another key member of the WNT signaling pathway , was also among the top identified HMRs and is, together with an additional illustrative example, displayed in Fig. 5c, d. Importantly, these findings were validated in an independent cohort of colorectal metastasis samples (n = 24) using DNA methylation microarray analysis (Student’s t-test, p < 0.05; Fig. 5e, f). Thus, these findings suggest that large cancer-specific HMRs are likely candidate markers for identifying sequences that could act as de novo activators in a super-enhancer-like manner.