Abstract 3172: Mapping the molecular landscape of acute myeloid leukemia enables prediction of drug response from proteogenomic data

Abstract Acute myeloid leukemia (AML) is a deadly blood cancer that remains largely classified by genetic aberrations, which inform therapy stratification. However, therapeutic response cannot be predicted or explained by genetic abnormalities alone. The integration of multiple omics, consisting of genomic, transcriptomic, proteomic, and phosphoproteomic measurements, offers a holistic view to resolve the underlying pathophysiology of AML that influences response to therapy. In this work, we pair multi-omic characterization together with ex vivo drug sensitivity assays accrued with 145 small molecule inhibitors for 210 AML patient samples (Bottomly et al., Cancer Cell 2022). We showcase how the integration of these data can guide drug sensitivity exploration and prediction. We first expanded the dataset by generating matching comprehensive proteomics and phosphoproteomics data for the Beat AML samples and integrated these data using non-negative matrix factorization. This analysis identified four distinct proteogenomic subtypes of AML, each representing distinct clinical and biological features including differences in survival and biological pathway activation. We then sought distinct patterns of drug sensitivity across the subtypes of the patient cohort and found one pair of drugs, venetoclax and panobinostat, to be sensitive in complementary sets of patients, suggesting that they could be more effective in combination. Lastly, we further enhanced the proteogenomic subtypes by a building machine learning based model of distinct drug response that we then evaluated in vitro. Our results show that the four proteogenomic subtypes are independent yet complementary to existing mutational profiles, and can be used to improve drug treatment stratification. We tested the combination of panobinostat and venetoclax in patient samples and show that they are more effective in combination than as single agents. We then tested drug-specific machine learning models to predict drug response on AML cell lines that were in varying stages of resistance to the FLT3 inhibitor quizartinib. The models predicted a change across the proteogenomic landscape as quizartinib resistance evolves, resulting in a shift in drug sensitivities that we experimentally validated. This work represents a seminal effort in the integration of proteogenomic and ex vivo drug sensitivity datasets. In summary, we show how multi-omic characterization of AML maps a proteogenomic landscape that enables improved exploration of patient drug response and ultimately patient treatment. Citation Format: James C. Pino, Camilo Posso, Sunil K. Joshi, Michael Nestor, Jamie Moon, Joshua R. Hansen, Marina A. Gritsenko, Chelsea Hutchinson-Bunch, Karl K. Weitz, Kevin Watanabe-Smith, Jason E. McDermott, Brian J. Druker, Tao Liu, Jeffrey W. Tyner, Anupriya Agarwal, Elie Traer, Paul D. Piehowski, Cristina E. Tognon, Karin D. Rodland, Sara J. Gosline. Mapping the molecular landscape of acute myeloid leukemia enables prediction of drug response from proteogenomic 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 3172.