Abstract 5723: An in vivo pharmacogenomics platform replicates and extends biomarkers of therapy response identified via causal inference analysis of clinical data

Abstract Recent development of therapies targeting oncogenes have dramatically improved cancer care for subsets of patients and generated a wave of interest in cancer precision medicine. However, effective targeted therapies are scarce as we have a limited understanding of how drug responses are modulated by tumor genotype. Here, we utilize both human data and in vivo models to identify genetic drivers of therapy response in KRAS-driven non-small cell lung cancer patients treated with chemotherapy. We first present an integrative causal inference analysis of three clinical data resources: (1) the recently released Genomics Evidence Neoplasia Information Exchange, Biopharma Collaboration dataset (GENIE-BPC; n=197), (2) a selected set of patients from the Tempus Clinico-genomic Database (n=330), and (3) an additional cohort from Memorial Sloan Kettering Cancer Center (n=218). Each dataset was first analyzed separately using a causal inference pipeline. Doubly robust estimators for each variant were inferred within a counting process survival analysis model accounting for time-varying treatments and immortal time bias. Meta-analysis of the results from all three cohorts identified three commonly mutated and highly replicable genes with a significant effect on overall survival: KEAP1, SMARCA4, and CDKN2A. As further validation, we used our murine in vivo pharmacogenomics (PGx) platform that can quantify the effects of therapies across thousands of tumors of diverse genotypes. These tumors are initiated de novo in the native microenvironmental context in mice with an intact adaptive immune system. We tested a chemotherapy combination of carboplatin and pemetrexed in mice with KrasG12D-driven lung tumors and inactivation of each of 60 putative tumor suppressors. Treatment led to >75% reduction in tumor sizes relative to tumor suppressor inactivated (matched) vehicle-treated controls. These models identified causal effects for two out of the three candidates above: KEAP1 (resistance) and CDKN2A (sensitive). Moreover, our PGx platform identified additional candidate genes beyond those found using the clinical data, which had insufficient sample size for these rarely mutated genes. Together, we demonstrate how leveraging our PGx platform together with human data within a causal inference framework may improve the stratification of patients by their clinical outcomes, profoundly advancing the promise of precision medicine. Citation Format: David Amar, Erick Scott, Ian P. Winters, Gregory D. Wall, Dmitri A. Petrov, Monte M. Winslow, Joseph Juan, Ian K. Lai, Lafia Sebastian, Edwin A. Apilado, Gabriel Grenot, Vy B. Tran, Charles Rudin, Michael J. Rosen. An in vivo pharmacogenomics platform replicates and extends biomarkers of therapy response identified via causal inference analysis of clinical data. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5723.