Genome‐wide DNA methylation analysis of colorectal adenomas with and without recurrence reveals an association between cytosine‐phosphate‐guanine methylation and histological subtypes

Aberrant methylation of DNA is supposed to be a major and early driver of colonic adenoma development, which may result in colorectal cancer (CRC). Although gene methylation assays are used already for CRC screening, differential epigenetic alterations of recurring and nonrecurring colorectal adenomas have yet not been systematically investigated. Here, we collected a sample set of formalin‐fixed paraffin‐embedded colorectal low‐grade adenomas (n = 72) consisting of primary adenomas without and with recurrence (n = 59), recurrent adenomas (n = 10), and normal mucosa specimens (n = 3). We aimed to unveil differentially methylated CpG positions (DMPs) across the methylome comparing not only primary adenomas without recurrence vs primary adenomas with recurrence but also primary adenomas vs recurrent adenomas using the Illumina Human Methylation 450K BeadChip array. Unsupervised hierarchical clustering exhibited a significant association of methylation patterns with histological adenoma subtypes. No significant DMPs were identified comparing primary adenomas with and without recurrence. Despite that, a total of 5094 DMPs (false discovery rate <0.05; fold change >10%) were identified in the comparisons of recurrent adenomas vs primary adenomas with recurrence (674; 98% hypermethylated), recurrent adenomas vs primary adenomas with and without recurrence (241; 99% hypermethylated) and colorectal adenomas vs normal mucosa (4179; 46% hypermethylated). DMPs in cytosine‐phosphate‐guanine (CpG) islands were frequently hypermethylated, whereas open sea‐ and shelf‐regions exhibited hypomethylation. Gene ontology analysis revealed enrichment of genes associated with the immune system, inflammatory processes, and cancer pathways. In conclusion, our methylation data could assist in establishing a more robust and reproducible histological adenoma classification, which is a prerequisite for improving surveillance guidelines.


| INTRODUCTION
Colorectal cancer (CRC) has the third highest cancer incidence worldwide with approximately 1.8 million new cases in 2018. 1 As CRC develops mainly via the adenoma-carcinoma sequence, endoscopic resection of adenomas reduces CRC risk significantly. 2 However, recurrences of initially resected polyps are common and contribute to the risk of developing CRC. 3 Many guidelines define postpolypectomy colonoscopy surveillance intervals, 4-6 but the scientific rationale for the different intervals is vague. Therefore, biomarkers for the prediction of recurrences could help to identify patients at risk and to define improved surveillance intervals.
As CRC develops over several years, molecular evidence has accumulated that the disease is driven by the acquisition of genetic and epigenetic alterations. Several distinct molecular pathways have been described: (a) the chromosomal instability (CIN) pathway that is associated with aneuploidy, copy number alterations (CNAs), and mutations most frequently affecting APC, TP53, and KRAS 7,8 ; (b) the microsatellite instability pathway that is caused by DNA hypermutation 9 ; (c) the CpG island mismatch repair phenotype (CIMP), which is defined by extensive methylation of CpG islands. 10 Abnormal DNA methylation is supposed to happen very early and affects up to 85% of adenomas and CRCs. 11 These tumors exhibit pathological methylation patterns predominantly affecting canonical and noncanonical WNT-pathway genes (eg, AXIN2, CTNNB1, SFRP1, and SFRP2) and Polycomb group genes. 12,13 Thus, it is reasonable to assume that adenoma recurrence after polypectomy might at least partially be triggered by methylation pattern changes.
The evolution from colorectal adenoma to CRC is associated with increasing hypermethylation of CpG islands (CGI) in promoter regions of tumor suppressor genes and concurrent global hypomethylation of CpG sites. 14 DNA hypomethylation might contribute to CIN because of the reactivation of transposable elements and chromosomal rearrangements. 15,16 Additionally, it is known that aberrant methylation of CpGs in gene bodies contributes to dysregulation of gene expression. 17,18 Recent advances in analytical technologies have permitted the assessment of genome-wide methylation patterns. 19 Consequently, multiple studies have shed light on the methylome landscapes of CRCs and determined the contribution of altered methylation to the development and progression of CRC. 13,20,21 The aberrant methylation of DNA also affects adenomas. 22 However, differential epigenetic alterations of recurring and nonrecurring colorectal adenomas have not yet been systematically and comprehensively investigated. In this study, we aimed to unveil differentially methylated CpG positions (DMPs) across the genome of formalin-fixed paraffin-embedded (FFPE) colorectal specimens comparing primary adenomas with recurrence vs primary adenomas without recurrence, primary adenomas vs recurrent adenomas, and colorectal adenomas vs normal mucosa. We utilized the Illumina Infinium Human Methylation 450K BeadChip array (HM450K), covering more than 485 000 genome-wide distributed CpG sites, to analyze the distribution of DMPs across the genome. A pathway enrichment analysis was conducted to uncover the most frequently altered pathways and hence to assess the biological functions of the DMPs. Selected markers were subsequently quantitatively analyzed by pyrosequencing to validate the array results.  (Table 1): (a) primary adenomas that did not unveil a tumor in the observation period (n = 30; primary adenoma without recurrence); (b) primary adenomas that exhibited a documented recurrence of the adenoma at the same location during the follow-up period (n = 29, primary adenoma with recurrence); (c) recurrent adenoma (n = 10; matched pairs corresponding to 10 of the primary adenomas with recurrence); and (d) normal colonic mucosa (n = 3).

| Clinical samples
Pathological classification was done, in accordance with the current WHO classification from 2010, 23 by two board-certified pathologists blinded to all data (TG/JR). Histological classification discerned tubular, tubulovillous, or villous adenomas ( Table 2). The adenomas exhibited cytological and histological alterations fitting the WHO definition of low-grade dysplasia (LGD).

| Tumor sample preparation
Consecutive sectioning was performed on archived FFPE tissue blocks.
The first tissue slide was stained by hematoxylin and eosin (H&E) and the region of interest (>70% tumor content) was marked. Two consecutive FFPE sections (10 μm thick) were deparaffinized in xylene for 10 minutes and rehydrated in an ethanol series for 10 minutes. The tumor area was macrodissected with a scalpel under the guidance of the marked H&E slide. Genomic DNA was extracted using the Gentra Puregene Tissue Kit (Qiagen, Hilden, Germany) and subsequently spectrophotometrically quantified by NanoDrop ND-1000 (Thermo Fisher Scientific, Waltham, MA). Eluted DNAs were purified by applying the DNA Clean & Concentrator Kit (Zymo Research, Irvine, CA) before a quality check was performed using the Infinium HD FFPE QC Real-time PCR assay (Illumina, San Diego, CA) as indicated in the manufacturer's protocol.

| DNA methylation microarray analysis
DNAs (250 ng) of 69 colorectal adenomas and three normal mucosae were subjected to bisulfite conversion using the EZ-96 DNA Methylation Kit (Zymo Research). Converted DNAs were hybridized to six HM450K arrays (Illumina) by sequential whole-genome amplification, enzymatic fragmentation, precipitation, and resuspension according to the manufacturer's instructions. Samples were randomly distributed on the arrays and hybridized for 20 hours at 48 C in a hybridization oven. Subsequently, arrays were treated with a primer extension, an immunohistochemical staining, and a coating as indicated by the supplier's protocol (Illumina). 19 Probe signals were detected by an iScan array scanner (Illumina) and microarray data (NCBI GEO Accession No. GSE129364) were exported as idat files. Quality control metrics for samples and probes were assessed by the GenomeStudio software

| Microarray data analysis
Principal component analysis (PCA) and unsupervised hierarchical clustering of the samples were conducted using P < .05 in a two-way cluster analysis (Euclidean distance and Ward's group linkage). Clinicopathological data were correlated via Spearman's correlation coefficient, whereas categorical variables were tested using the Freeman-Halton test.
Methylation levels expressed as β values were evaluated to identify significant DMPs applying the Benjamini Hochberg-corrected false discovery rates (FDR adjusted) of q < 0.05 and q < 0.01 using the R limma package. Significant DMPs were sought with a fold change of >10% by comparing sample sets of primary adenomas without recurrence, primary adenomas with recurrence, recurrent adenomas, and normal mucosa. Preselected CpG sites were filtered for top genes, that is, the most differentially hyper-or hypomethylated CpG positions that harbor the greatest potential as informative biomarkers. We considered probes with Δβ ≥ |0.1| and which are not located on sex chromosomes to eliminate a gender-specific bias.

| Bisulfite pyrosequencing of selected CpG loci
The obtained methylation array results were orthogonally validated via bisulfite pyrosequencing targeting eight CpG loci within a distinct gene body region of GREM2 ( Figure S1). The gene was selected for validation as two of these CpG loci were listed among the most signif-

| Identification of DMPs
The filtered probes of the HM450K were used to identify DMPs via the FDR across the adenoma subgroups (Table 3). Neither the comparison of primary adenomas without recurrence vs primary adenomas with recurrence, nor the comparison of primary adenomas vs the corresponding recurrent adenomas (matched pairs), nor the comparison of primary adenomas without recurrence vs adenomas associated with recurrence (primary and recurrent) showed any significant DMPs (  Table 3). Interestingly, 99% of DMPs were hypermethylated in recurrent adenomas compared with the primary adenomas with recurrence ( Figure S2A). Moreover, comparison of recurrent adenomas vs all primary adenomas exhibited 575 (0.2%) DMPs (comparison B, FDR≤0.05; Table 3). The DMPs present in the recurrent adenomas were nearly completely (99%) hypermethylated ( Figure S2B). As expected, colorectal adenomas displayed 9266 (2.8%) DMPs in comparison with normal mucosa samples (comparison C, FDR≤0.05; Table 3). More than two-thirds of the DMPs (73%) were hypermethylated in colorectal adenomas compared with normal mucosa ( Figure S2C).

| CpG methylation by pyrosequencing
Locus-specific methylation alterations of eight CpG dinucleotides in the gene body of the validation candidate GREM2 were evaluated by pyrosequencing ( Figure S1). Bisulfite-converted DNAs of primary adenomas with (n = 15) and without (n = 9) recurrence, and recurrent adenomas (n = 4) were analyzed for GREM2 gene body methylation frequencies ( Figure 6A). Detected GREM2 methylation frequencies The median methylation frequency by individual CpG sites 1-4 sorted by adenoma groups did not show a difference (P = .590) ( Figure 6D). However, recurrent adenomas tended to be slightly more hypermethylated compared with primary adenomas with (P = .312) and without (P = .406) recurrence. Comparing the median methylation levels of the individual CpG sites 12-15 sorted by groups demonstrated a highly significant difference among the adenoma groups (P = .002) ( Figure 6E). Moreover, recurrent adenomas were hypomethylated compared with primary adenomas with (P = .003) and without (P = .004) recurrence.

| DISCUSSION
Methylation of CGI is a mechanism for suppressing gene transcription in cancer including colorectal neoplasia. 33 Many tumor suppressor genes are hypermethylated at CpG-dense promoters, which are, therefore, silenced in colorectal tumors. 16,34 It is also now accepted that methylation assays can be applied for CRC screening. The FDA approved a SEPT9 gene methylation assay in 2016 as an additional CRC screening option to be considered for patients who have a history of an intermittent completion of colonoscopy and fecal occult blood tests. 35 Thus, this study sought to discover aberrantly methyl- proximal promoter regions were more frequently affected by hypermethylation compared with gene body and 3 0 UTR, which were found to be targeted by hypomethylation. This fact is linked to the current understanding of alterations in tumors suggesting an association of gene silencing with promoter hypermethylation, whereas gene transcription is associated with gene body hypomethylation. 16 However, DNA methylation of gene bodies might also be involved in differential promoter usage, alternative splicing, or the prevention of transcription initiation. 17 Thus, hypermethylation within the gene body of GREM2 in recurrent adenomas vs primary adenomas with recurrence could indicate a gene silencing leading to deregulation of the BMP axis and WNTpathway disruption. 44 We discovered a discrepancy of the mean β values of recurrent adenomas in probe cg02577267 when comparing the results obtained via pyrosequencing and HM450K. This is because of the less representative number of samples analyzed by pyrosequencing.
Furthermore, we could detect that the hierarchical clustering (methylation β values) of specimens showed a significant difference in the distribution of adenomas with tubular histology compared with adenomas with tubulovillous/villous histology. It is known that the frequency of methylated genes is higher in histologically advanced adenomas compared with hyperplastic polyps and low-grade adenomas. 45 However, the differentiation of histological subgroups based on their methylation pattern is novel. A Dutch register study demonstrated a considerable interlaboratory variation in evaluating colorectal adenomas. 46 Specialized gastrointestinal histopathologists showed only moderate (kappa = 0.49) concordance in discriminating the different polyp forms of the group of serrated adenomas. 47 Thus, an expert panel proved that morphological criteria alone are not sufficient but can be improved by including molecular data like BRAF or KRAS mutation status. 48 Additionally, a meta-analysis indicated that hypermethylation of p16 might be an unfavorable biomarker for CRC patients. 49 Recent studies about brain tumor classification showed that the analysis of methylation patterns could be highly relevant for precise diagnostics and treatment suggestions. 50,51 The same holds true for sarcoma diagnostics and allows a more precise classification of undifferentiated and small blue round cell tumors with array-based DNA-methylation profiling. 52 The here presented finding that adenoma morphology differs by DNA methylation might provide the opportunity for enhancing the accuracy of the up to now mainly morphology-based classification of colonic adenomas.

DATA ACCESSIBILITY
The array data that support the findings of this study are openly available in NCBI Gene Expression Omnibus (GEO) at https://www.ncbi.