3128 - Integrating Functional and Computational Approaches to Precision Medicine for Medulloblastoma

Purpose/Objective(s):

Medulloblastoma (MB), the most common malignant pediatric brain tumor, is a heterogeneous disease comprising four molecularly-defined subgroups with distinct clinical outcomes – WNT, Sonic hedgehog (SHH), Group 3, and Group 4 – yet most MB patients receive similar multi-modal treatment. Even with aggressive therapy, many patients (particularly those with Group 3 MB) relapse and die of their disease. Mutation-based precision oncology approaches have limited utility due to lack of targetable mutations and significant genetic heterogeneity. We leverage a validated, transcriptome-based precision oncology computational platform for inference of regulatory protein activity and drug prioritization.

Materials/Methods:

We reverse engineered a MB-specific gene regulatory network (ARACNe-AP) from publicly available RNAseq profiles of 115 MB patient samples from the Children’s Brain Tumor Network. We then inferred the activity of 5,896 transcriptional regulators from each patient’s RNAseq profile, as well as from 20 MB patient-derived xenograft (PDX) RNAseq profiles, based on enrichment of their target genes in the differential gene expression signature (VIPER).

Results:

Unsupervised k-medoid clustering on protein activity across the 115 patient samples identified four clusters that segregated strongly with the known MB molecular subgroups (chi-squared p < 2.2^-16). We identified the tumor checkpoint comprised of the most dysregulated proteins—Master Regulators (MRs)—for each group and validated these findings in an independent cohort of 94 samples. These included established regulators such as GAB1 in SHH patients and MYC in Group 3 patients, as well as novel regulators that may inform future treatment. Specifically, we evaluated the activity of 180 directly targetable MRs (OncoTarget) in each sample, identifying an average of 14 (range 4-32) significant (p=10^-5) candidate targetable MRs per sample. For example, AURKA was a top MR among Group 3 patients, prioritizing aurora kinase inhibitors as a therapeutic opportunity. Importantly, MB patient MRs were significantly enriched in the MB PDX protein activity signatures, suggesting they are robust models for drug validation.

Conclusion:

These cognate PDX model cell lines will be used to generate drug-perturbed RNAseq profiles, allowing drugs to be prioritized for individual MB samples based on ability to invert the activity of the sample’s entire tumor checkpoint (OncoTreat). Drugs prioritized by OncoTarget/OncoTreat will be compared to predictions generated from high-throughput functional drug screening, an approach currently utilized in a clinical trial for MB. Comparison of candidate drugs targeting high-risk molecular subgroups derived from both computational and functional approaches will nominate novel therapies for preclinical validation, and guide future implementation of these CLIA-compliant tools into clinical precision oncology trials for MB patients.