MODL-10. COMPUTATIONAL DRUG SENSITIVITY PREDICTS TAILORED THERAPY FOR MEDULLOBLASTOMA SUBGROUPS

Abstract

Abstract BACKGROUND Medulloblastoma is the most common malignant pediatric brain tumor. Tumors are typically characterized as SHH α- δ, WNT α-β, Group 3α- γ, or Group 4 α- γ. Current standard-of-care includes surgery, radiation, and chemotherapy; however, treatment response and prognosis vary widely between subgroups. Despite extensive characterization of medulloblastoma tumors into four molecular subgroups, few subgroup specific therapies have advanced to the clinic and a universal approach is unlikely to be effective due to the considerable intertumor heterogeneity of medulloblastoma. Hence, there is an urgent demand for innovative therapeutic strategies. METHODS To address heterogeneity between patients, we have developed a novel platform called DrugSeq, which predicts drug sensitivities in patients and is able to stratify drug response based on subgroup. We first calculated disease signatures for each patient by normalizing patient gene expression to median gene expression across the Cavalli dataset. Using previously developed transcriptional consensus signatures (TCSs), which contain genes that are consistently dysregulated following drug treatment, we then calculated the discordance between each patient disease signature and drug TCS. In doing so, we were able to identify drugs to target specific molecular subtypes based on Cavalli molecular classifications. RESULTS As a proof of concept we identified vismodegib and curcumin as selective for SHH-MB. Furthermore, we correlated predicted drug response with target protein expression. For example, we found that predicted sensitivity to vismodegib correlates closely with GLI1 protein expression. CONCLUSIONS Collectively, we show that DrugSeq may identify novel therapies and facilitate patient stratification in clinical trials, leading to more successful targeted medulloblastoma therapies.