MEDB-43. Development of a bioinformatics pipeline for identification of differential DNA methylation events associated with medulloblastoma relapse

Relapsed medulloblastoma (rMB) is treatment-resistant and fatal in ~95% of cases. The epigenetic features of rMB, and any role as drivers of disease relapse/treatment-resistance have yet to be investigated. We therefore developed a pipeline to identify differentially methylated CpGs (DM-CpGs) and regions (DMRs) in a paired-rMB cohort. Our paired-rMB cohort (n=61, relapsed tumours matched with diagnosis counterparts) with available Illumina Methylation 450K/850K microarray data was processed in R-Studio. The packages Limma and DMRcate were used to perform a paired differential methylation analysis on a filtered selection of array probes (n=335,767), identifying DM-CpGs and DMRs with a 5% FDR. DMRs were further retained if they had a maximum-Δβ of >0.2 and correlated with locus-specific gene expression in a separate paired DNA-methylation array/RNA-seq cohort from medulloblastoma diagnosis samples (n=202). Finally, we created univariable Cox models to assess the prognostic potential of DM-CpGs/DMRs in an independent survival cohort of medulloblastoma diagnosis samples (n=498). Across the paired-rMB cohort, there were few significant differential methylation events initially identified at relapse (n=258 DM-CpGs, n=32 DMRs). Upon sub-analysis by molecular group, MBGroup4 (n=18 pairs) alone yielded significant findings (n=189 DM-CpGs, n=26 DMRs). Most changes involved hypermethylation events detected at relapse. Multiple DM-CpGs identified at relapse were prognostic for both overall and event-free survival when assessed in our independent cohort (n=22 whole cohort, n=13 Group 4, BH-adjusted p<0.05). When applying the DMR filters, only the MBGroup4 DMRs passed the Δβ filter (n=18/26), with few correlating with gene expression (n=2, p<0.001), and none demonstrating prognostic significance. This pipeline facilitates exploration of the clinical relevance of epigenome-wide changes in a paired-rMB cohort. We highlight the potential prognostic significance of DM-CpGs, and future work will explore the potential functional role of candidate-genes associated with our DMRs, as novel drivers of rMB.