Search

Additional file 2: Contains all supplementary tables.

Additional file 3: Review history.

Additional file 1: Contains all supplementary figures and supplementary methods.

Nongenetic predisposition to colorectal cancer continues to be difficult to measure precisely, hampering efforts in targeted prevention and screening. Epigenetic changes in the normal mucosa of patients with colorectal cancer can serve as a tool in predicting colorectal cancer outcomes. We identified epigenetic changes affecting the normal mucosa of patients with colorectal cancer. DNA methylation profiling on normal colon mucosa from 77 patients with colorectal cancer and 68 controls identified a distinct subgroup of normally-appearing mucosa with markedly disrupted DNA methylation at a large number of CpGs, termed as "Outlier Methylation Phenotype" (OMP) and are present in 15 of 77 patients with cancer versus 0 of 68 controls (P0.001). Similar findings were also seen in publicly available datasets. Comparison of normal colon mucosa transcription profiles of patients with OMP cancer with those of patients with non-OMP cancer indicates genes whose promoters are hypermethylated in the OMP patients are also transcriptionally downregulated, and that many of the genes most affected are involved in interactions between epithelial cells, the mucus layer, and the microbiome. Analysis of 16S rRNA profiles suggests that normal colon mucosa of OMPs are enriched in bacterial genera associated with colorectal cancer risk, advanced tumor stage, chronic intestinal inflammation, malignant transformation, nosocomial infections, and KRAS mutations. In conclusion, our study identifies an epigenetically distinct OMP group in the normal mucosa of patients with colorectal cancer that is characterized by a disrupted methylome, altered gene expression, and microbial dysbiosis. Prospective studies are needed to determine whether OMP could serve as a biomarker for an elevated epigenetic risk for colorectal cancer development.Our study identifies an epigenetically distinct OMP group in the normal mucosa of patients with colorectal cancer that is characterized by a disrupted methylome, altered gene expression, and microbial dysbiosis. Identification of OMPs in healthy controls and patients with colorectal cancer will lead to prevention and better prognosis, respectively.

Additional file 6. Genomic context specificity of different methylation platforms. Bar plots of the odds ratios (y-axis) of all sites (left) and of all aging sites (right) in DREAM to 450K array. X-axis is the genomic context of all comparisons. All comparisons were tested for significance by a chi-square test and stars indicate p-values < 0.0001.

Additional file 3. PCA analysis of age-related methylation sites. a) Summary of PCA analysis of the methylation values of 146 age-dependent probes in 6 publicly available datasets. Proportion of variance explained by each of the first 10 principal components (PC). b) Scatter plots of PC components 1 and 2, c) components 2 and 3, and d) components 1 and 3.

Additional file 1. Identification and validation of age-related DNA methylation sites in TCGA breast normal adjacent tissue. a) TCGA 97 normal-adjacent samples were used as validation dataset for the aging sites. One thousand permutations of the data were performed, and empirical p-values were computed. The distribution of the Spearman correlation r values of the actual dataset is shown in pink while the distribution of the correlation values obtained by random permutation analysis of the same data is shown in green. b) The age-dependent methylation changes were selected based on a cutoff of permutation empirical p-value (p < 0.05) and based on Spearman correlation of r ≥ 0.3 (gain of methylation with age) and r ≤ −0.3 (loss of methylation with age). The age-dependent sites from the discovery dataset (DREAM) and the validation dataset (450K array), were aligned and restricted to < 250 bp distance between SmaI sites and the 450K probes.

Additional file 5. Comparison of TCGA clinical parameters. Comparison of different variables was done between the not-outlier and outlier samples and accelerated outliers. The outlier status in this analysis was based off the aging methylation data of 427 patients. Chron Mean Age and Pred Mean Age are chronological and predicted average ages respectively. Mixed ductal lobular carcinoma (MDLC), Invasive ductal carcinoma (IDC), Invasive lobular carcinoma (ILC). TN is triple negative. The significant p-value was generated by ANOVA testing followed by Tukey’s HSD post hoc test.

Additional file 2. Genomic distribution of validated aging sites. a) Distribution of the 146 aging sites within the promoter CpG islands (pCGI), promoter non CpG islands (pnCGI), non-promoter CpG islands (npCGI) and non-promoter non CpG islands (npnCGI). b) Distribution of all sites (45,135) within the same genomic context as in (a) in the discovery dataset. c) Summary of odds ratios of the genomic region specificity of age-related sites (146) compared to all sites (45,135) in the discovery dataset. Dots represent the point estimates of odds ratio with lines representing 95% confidence intervals around the estimates. A chi-square test was used to test for statistical significance for each comparison; p-values for all comparisons were significant. d) Contingency tables of the shared 146 aging sites between the DREAM and array platforms. Each table represents a genomic context, and the numbers indicate the number of CpG for hyper (hypermethylated) or hypo (hypomethylated) sites. All comparisons were significant with p-values < 0.0001.

Additional file 4. Outlier analysis using Horvath’s multi-tissue estimator clock’s 353 CpG probes. a) Flow chart of the identification of age-related sites in TCGA. Bold numbers in parentheses indicate the number of Horvath clock’s CpGs present based on the cut-off. b) Venn diagram showing no overlap between the clock’s 32 CpG sites and the validated 146 aging sites. c) Venn diagram showing the overlap between outlier samples identified by Horvath clock’s CpG sites and the outliers identified by our 146 aging sites. the significance of overlap was tested by the hypergeometric test and found to be insignificant (p= 0.06). d) Overlap of the accelerated outliers identified by the clock’s CpG sites in the TCGA dataset and the accelerated outliers identified by our aging sites in the same dataset.