Interpretation of genome-wide infinium methylation data from ligated DNA in formalin-fixed, paraffin-embedded paired tumor and normal tissue

Formalin-fixed, paraffin-embedded (FFPE) samples are a highly desirable resource for epigenetic studies, but there is no suitable platform to assay genome-wide methylation in these widely available resources. Recently, Thirlwell et al. (2010) have reported a modified ligation-based DNA repair protocol to prepare FFPE DNA for the Infinium methylation assay. In this study, we have tested the accuracy of methylation data obtained with this modification by comparing paired fresh-frozen (FF) and FFPE colon tissue (normal and tumor) from colorectal cancer patients. We report locus-specific correlation and concordance of tumor-specific differentially methylated loci (DML), both of which were not previously assessed.

Ligation of FFPE DNA prior to the Infinium genome-wide methylation assay may detect a reasonable number of loci, but the numbers of detected loci are much fewer than in FF samples. More importantly, the concordance of DML detected between FF and FFPE DNA is suboptimal, and DML from FFPE tissues should be interpreted with great caution.

We used Illumina's Infinium Methylation 27K chip for 12 pairs of FF and 12 pairs of FFPE tissue from tumor and surrounding healthy tissue from the resected colon of the same individual, after repairing the FFPE DNA using Thirlwell's modified protocol.

For both tumor and normal tissue, overall correlation of β values between all loci in paired FF and FFPE was comparable to previous studies. Tissue storage type (FF or FFPE) was found to be the most significant source of variation rather than tissue type (normal or tumor). We found a large number of DML between FF and FFPE DNA. Using ANOVA, we also identified DML in tumor compared to normal tissue in both FF and FFPE samples, and out of the top 50 loci in both groups only 7 were common, indicating poor concordance. Likewise, while looking at the correlation of individual loci between FFPE and FF across the patients, less than 10% of loci showed strong correlation (r ≥ 0.6). Finally, we checked the effect of the ligation-based modification on the Infinium chemistry for SNP genotyping on an independent set of samples, which also showed poor performance.