Abstract 4385: Proteogenomics-driven synthetic lethality discovery to predict targetable protein dependencies induced by somatic deletions

Abstract

Abstract Computational prediction of synthetic lethality has been shown as a promising approach to identify therapeutic vulnerabilities in many cancer types but has been limited to using genomics and transcriptomics data for discovery. Recent advancements on integrating mass spectrometry (MS)-based proteomics with genomics and transcriptomics in cancer research have produced proteogenomics datasets on large tumor cohorts, enabling the integration of proteomics data into synthetic lethality discovery. This is particularly attractive for tumors driven by loss-of-function alterations such as somatic deletions. Although deletions are challenging to target themselves, they may induce targetable protein dependencies, and this relationship can be identified through computational prediction of synthetic lethality. Here, we performed an unbiased discovery of such synthetic lethal interactions by identifying protein targets of drugs for which under-expression is observed significantly less than expected by chance with gene deletions using proteogenomics data from a breast cancer cohort with 105 tumors. We found many significant pairs uniquely driven by either protein under-expression (18,897 pairs) or mRNA under-expression (42,754 pairs) while only 5% (3,042 pairs) were supported by both protein and mRNA measurements. To systematically evaluate candidate synthetic lethal pairs, an AUROC analysis was performed using publicly available drug response data from breast cancer cells. Results showed synthetic lethal relationships supported by both protein and mRNA-based analyses were substantially more predictive of drug response (AUC = 81%) than protein alone (AUC = 65%) or mRNA alone (AUC = 61%). Despite having comparable AUCs, protein-based analysis showed much higher sensitivity than mRNA-based analysis when limiting false positive rate to less than 10%. An example of a synthetic lethal pair supported by both protein and mRNA-based analyses include tumors with low AURKA protein/mRNA and PARP2 deletions, which were significantly under-represented in the proteogenomics dataset (p = 0.003). Interestingly, these tumors were enriched for triple-negative breast tumors (p = 0.001), a type of breast cancer that lacks biomarkers to guide clinical treatment. This finding suggests AURKA as a putative dependency in triple-negative tumors with PARP2 deletion as a biomarker. Taken together, these results demonstrate that a multi-omics approach using proteomics to complement transcriptomics along with genomic changes may better predict therapeutic vulnerabilities in breast cancer with driver gene deletions than using proteomics or transcriptomics alone. As genomic testing has become more widely implemented as part of precision oncology programs, our approach to identifying protein dependencies induced by loss-of-function genomic alterations may provide new treatment opportunities for many patients with breast cancer or other tumor types. Citation Format: Jonathan T. Lei, Eric Jaehnig, Bing Zhang. Proteogenomics-driven synthetic lethality discovery to predict targetable protein dependencies induced by somatic deletions [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 4385.