Abstract Glioblastoma is an aggressive brain cancer characterized by complex intertumoral heterogeneity. Yet, no methods have successfully stratified patients for effective individualized treatments. We aimed to use a systems medicine approach to individualize treatments for patients. From 27 patients undergoing surgery for either a newly diagnosed or a progressive glioblastoma, we collected biopsies and cultured the cells as spheroids. Drug sensitivity testing for >500 drugs from the FDA-approved oncology collection was performed. Drug efficacy was quantified using a modified area under the dose-response curve (Drug Sensitivity Score, DSS), and individualized by normalizing the DSS to the drug response in bone marrow cells from healthy donors (selective DSS, sDSS). All samples were assayed by WES, WGS, and RNA-sequencing. We found that glioblastomas can be stratified into three DSS groups. These were associated with expression of genes both known (e.g. ALDH1A1), and not known (e.g. VRK2) to be linked to drug resistance. By unsupervised clustering of sDSS, we found a robust patient stratification primarily driven by drug sensitivity to four groups of targets: i) EGFR, ii) FGFR/PDGFR/VEGFR, iii) MDM2 and iv) MEK/ERK. We found differentially expressed genes (N = 20-90 genes, varying by group) between the responders and non-responders in each group. We explored differential pathway activity between the responders and non-responders in each group to correlate drug responses to transcriptional networks to inform underlying disease mechanisms. The transcriptional networks further provide a molecular basis for drug predictions for additional treatment prioritizations. We conclude that our platform can stratify glioblastomas into subgroups, inform the underlying biology of drug sensitivity, and prioritize treatments for patients. Clinical utility is currently under evaluation in a formal trial.
Read full abstract