MDB-23. Multi-Omic Characterization of Pediatric Medulloblastoma: Uncovering Novel Prognostic Biologic Entities for Precision Medicine

While advances have been made in the molecular subclassification of medulloblastoma, significant molecular heterogeneity remains within subgroups with unrealized ramifications for therapeutic design. Multi-modal characterization holds promise to improve precision medicine approaches and more effectively disambiguate disease risk. Leveraging multi-omic data from 152 primary medulloblastoma tumors from the Children’s Brain Tumor Network (CBTN), we derived a novel, 13-cluster model of pediatric medulloblastoma using non-negative matrix factorization incorporating coding DNA mutations, copy number variations, functional alternative splicing events, transcript, and methylation profiling data. Wnt-driven medulloblastoma remained a relatively pure subtype, and Sonic Hedgehog (SHH), Group 3, and Group 4 medulloblastoma subdivided into additional subgroups. Kaplan-Meier analysis found that the 13 clusters showed highly distinct overall survival outcomes (p < 0.0001), uncovering multi-omic clusters comprised of Group 3 (p = 0.00024) and Group 4 (p = 0.039) subgroups that were highly distinguishable. Cox regression adjusted for age at diagnosis and extent of tumor resection found that Group 3 (p < 0.001) and Group 4 (p < 0.001) multi-omic subgroups also differed in terms of event-free survival. Known medulloblastoma driver mutations enriched in a subset of multi-omic subgroups, and highly distinct molecular events that were correlated across the expression, methylation, and splicing layers were observed within the multi-omic clusters, among them putative regulatory methylation sites impacting expression targets known to induce oncogenesis. Among prognostically distinct Group 3 and 4 multi-omic subgroups, we found unique pathways inferred to be druggable when querying the NIH-LINCS drug response library, including VEGR2, DAG-IP3, GPCR, FLT3-STAT, PIP3-AKT, and ERBB2 signaling. When predicting multi-omic risk group membership using MR- and pathology-based imaging data, we found that intermediate-high and high-risk groups were more effectively predicted by MRI features related to edema, cellularity, and contrast enhancement (AUC-ROC: 0.9 and 0.83, respectively). Our study demonstrates that multi-omic molecular features underlie novel risk groups in medulloblastoma with potentially actionable biology and that may be effectively predicted by foundational, clinically-relevant imaging data.