Epic methylation array

Bulk sample DNA methylation was measured using the EPIC methylation array (Illumina) for opposite sides of 16 tumors, gland pairs from opposite sides of 11 CRCs, and six normal colons. Idat files in each color channel (red and green) were processed with the ‘noob’ function in the minfi R program to extract the methylated and unmethylated probe signal intensities and to calculate the beta value for each data point after within-array normalization for background correction, dye-bias equalization and probe design. Beta values between samples were compared based on pairwise distances (PWDs) defined as the absolute differences in beta values. Only autosomal CpG sites were analyzed.
Methylation of four widely spaced samples from nine normal adult colons (ages 50 to 70 years, mucosal biopsies, not EDTA washout-purified) was also measured with the Infinium methylation arrays (Illumina). Raw intensity data was read, preprocessed and batch corrected using the minfi (v.1.16.0) in R (v3.2.2). Probes were mapped to the genome and probes showing mean intensity p-value > 0.05 across all samples were excluded. Excluded probes also included: probes with proximal SNPs; non-CpG probes; probes on sex chromosomes; probes with observed cross-reactivity between two or more genomic regions. After subsequent filtering, 426,718 probes were left for analysis.

Time-to-event outcomes, including the primary endpoint PFS and OS, were summarized using Kaplan–Meier estimates of the survival function. Cox proportional hazards models were used to compare the two treatment arms and to evaluate and adjust for potential prognostic factors (e.g., nodal status, microscopic margins, increase in CA19-9, previous SOC therapy). Logistic regression was used to compare the response rate between the groups and to identify risk factors associated with response. For exploratory correlative outcomes, summary statistics (e.g., means, standard errors) and plots were used to describe the pharmacodynamic endpoints at each time point (resection and relapse) as well as the change over time. T tests and Fisher’s exact tests were used to assess the differences in the change in pharmacodynamic outcomes between those with and without CC-486 exposure. We performed 2-sided, paired-sample, empirical Bayes moderated t tests on each CpG site evaluated on the Illumina Epic methylation array. Results were considered statistically significant for p values < 0.05 for clinical and pharmacodynamic analyses and for Benjamini–Hochberg adjusted p values < 0.1 for DNA methylation correlates.

Genomic DNA was isolated from the suction blister roofs using the QIAamp DNA Investigator Kit (Qiagen, Inc.) following the manufacturer’s instructions. Subsequently, the isolated genomic DNA was processed on EPIC Methylation arrays (Illumina, Inc.) to obtain the DNAm patterns.
We utilized the R Bioconductor minfi package (Aryee et al., 2014 (link); Fortin et al., 2017 (link)) to pre-process the raw.idat files from the EPIC arrays. A series of filtering steps were applied to ensure data quality. First, methylation loci (probes) were filtered based on high detection p-values (p > 0.01). Then, probes were filtered based on their self-hybridization ability and potential SNP contamination resulting in a total number of 794,441 CpGs. To normalize the methylation data, we performed matrix normalization using quantile normalization with the “preprocessQuantile” function provided by the minfi package (Aryee et al., 2014 (link); Fortin et al., 2017 (link)). Quality control checks were performed after each pre-processing step to monitor the integrity and reliability of the data.