Context-dependent gene essentiality in glioblastoma

Abstract

Glioblastoma (GBM) is a heterogeneous and aggressive brain tumour that is invariably fatal despite maximal treatment. Genetic or transcriptomic ‘biomarkers’ could be used to stratify patients for treatments, however, pairing biomarkers with appropriate therapeutic ‘targets’ is challenging. Consequently, therapeutics have not yet been optimised for specific GBM patient subsets. Here, I integrate genome-wide CRISPR/Cas9 knockout screening and genetic-annotation data for 60 distinct patient-derived, IDH-wildtype, adult GBM cell lines, quantifying the essentiality of 15,145 genes. I describe a novel method that uses Targeted Learning to estimate the effect size of GBM-relevant biomarkers on context-dependent gene essentiality (GBM-CoDE) and use it to derive multiple target-biomarker pair hypotheses. Two of the target-biomarker pairs that I have identified (EGFR mutation/amplification as a biomarker of WWTR1 essentiality, and low VRK2 expression as a biomarker of VRK1 essentiality) have been validated in GBM, implying that my additional novel findings may be valid. I have identified several other novel target-biomarker pairs that require further experimental validation. I have developed a web application that I will release in an open, accessible format for other researchers to explore the results from GBM-CoDE. I hope dissemination of these results will accelerate translation to biomarker-stratified clinical trials. To my knowledge, framing the pursuit of CoDE as an estimation problem solved with targeted machine learning has not been described before in GBM or any other type of cancer. My method is readily translatable to other cancers of unmet need.