Abstract
The proteogenomic characterization of human tumors has begun to uncover molecular attributes enriched in different tumor subtypes and proposed a panoply of subtype-specific targets. As these studies lacked experimental testing of potential targets in high-quality cell models accurately reflecting individual tumor subtypes, it remains unclear whether any of the proposed targets retained value for precision medicine. The absence of clinically useful classifiers further hampered translation of proteogenomic information to better diagnose and treat cancer patients. In the case of the brain tumor glioblastoma (GBM), such challenges are exacerbated by tumor heterogeneity and broad therapeutic resistance. Here, we analyzed a large dataset of human GBM to determine the multi-omics features that characterize four GBM subtypes we recently identified through single cell RNAseq analysis. Each subtype exhibits activation of unique hallmark functional traits including distinct sensitivity to inhibitors of mitochondrial respiration for the mitochondrial subtype. The inspection of proteomics, phosphoproteomics, metabolomics, lipidomics and acetylomics data revealed that each GBM subtype has a coherent molecular structure that drives the dominant function and can be extracted by each analytical platform. Stratification of tumors in four functional classes is not a specific attribute of GBM as we were able to unbiasedly identify the same subtypes in breast and lung cancer. To identify actionable targets in GBM we developed an unbiased protein kinase signaling network approach for the selection of master kinases aberrantly activated in each GBM subtype. The experimental follow-up using a library of annotated GBM patient derived organoids (PDOs) established that inhibition of DNA-PK with the clinically tested compound nedisertib radiosensitizes proliferative/progenitor PDOs by exacerbating replication stress-induced DNA damage in this GBM subtype. Furthermore, genetic and pharmacological tools qualified PKCδ as vulnerable target for the broadly resistant glycolytic/plurimetabolic GBM subtype. Thus, together with the targeting of mitochondrial GBM with OXPHOS inhibitors, the functional classification delivers experimentally validated actionable targets for three GBM subtypes. To provide rapid translation of the classifier for precision medicine in GBM, we developed a probabilistic classification tool which determines the probability that a patient’s GBM belongs to one of the four subtypes based on transcriptomic features and exhibits optimal performance when using RNA extracted from either frozen and paraffin embedded tissues. The algorithm is publicly accessible. It can be used in retrospective studies to evaluate the association of therapeutic response with GBM subtypes and as tool for selection criteria in prospective clinical trials.
Citation Format: Simona Migliozzi, Young Taek Oh, Luciano Garofano, Hasanain Mohammad, Fulvio D'Angelo, Ryan D. Najac, Franck Bielle, Karima Mokhtari, Jann N. Sarkaria, Michele Ceccarelli, Marc Sanson, Anna Lasorella, Antonio Iavarone. Pathway based analysis of glioblastoma by multi-omics with applications in targeted therapy, prognosis and probabilistic classification [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 4014.