GE-07DNA METHYLATION IN MEDULLOBLASTOMA AND MOLECULAR DIAGNOSTICS OF BRAIN TUMORS

We recently performed an in-depth analysis of the DNA methylation landscape of medulloblastoma, the most common malignant pediatric brain tumor (Hovestadt et al., 2014). By integrating whole-genome bisulfite sequencing and gene expression data of 50 primary human and murine samples, we identified profound differences between medulloblastoma subgroups. Most notable were extended regions of differential methylation downstream of the promoter of more than 1,000 genes that strongly correlated with gene expression (i.e. promoter downstream correlated regions, pdCRs). Enrichment of transcription factor binding motifs within focal lowly-methylated regions (LMRs) shed light on differential regulatory networks between subgroups. Some differences between subgroups were indicative of the cell-of-origin, while others were somatically acquired. Motivated by these findings, we expanded our study cohort to include an exhaustive set of more than 2,000 common and rare brain tumor samples from both pediatric and adult patients profiled by DNA methylation arrays. Unsupervised clustering refined histological entities into more than 70 molecular entities, often resulting in splitting or combining of morphology-based groups. To explore the diagnostic potential of DNA methylation profiling, we created a random forest-based classifier that is capable of predicting the molecular class of clinical specimens to complement histological diagnosis. Early clinical application frequently provided a refined diagnosis (e.g. molecular subtyping) or suggested re-classification, and was often associated with an expected benefit for patients.