分化状态与免疫相互作用促成AML中的药物反应

Introduction: We and others have linked efficacy of targeted therapies to cellular differentiation state in acute myeloid leukemia (AML). In this study, we refine this link by integrating ex vivo drug response, joint single-cell transcriptomic and proteomic, and cell-surface proteomics data that measures both myeloid differentiation and immune markers. Methods: A total of 90 AML patient samples were profiled by CyTOF (cytometry by time-of flight) with ~100 immune and myeloid differentiation makers to characterize the landscape of cell-surface protein expression. Matched drug response data was used from the largest assembled AML patient cohort BeatAML. Concurrently, five AML patient samples were profiled following ex vivo drug treatment using CITE-seq (cellular indexing of transcriptomes and epitopes by sequencing) with a total of 6 treatments (3 single agent, 2 pairwise and 1 triple combination treatment) and compared to DMSO treated cells. Custom computational pipelines were developed to enable data analysis. Topic modeling using the latent Dirichlet allocation (LDA) model was used to identify topics emerging in the space of both myeloid differentiation and immune markers. Topics were then stratified by matched drug response. Single-cell drug response prediction methods were trained on sc-RNA-seq to identify cell type-specific drug response signatures. Results: Our results recapitulate known biomarkers of response to single agents BCL2 inhibitor venetoclax (ven) and menin inhibitor revumenib. Further, we observe known and novel gene expression and cell population shifts when combined with hypomethylating azacytidine. For example, venetoclax eliminates primitive and B-cell populations. When ven is used in combination with revumenib, intermediate populations such as pro-monocytes mature into monocytes, yielding predominantly differentiated cells. Adding azacytidine to ven+revumenib further depletes intermediate populations, including pro-monocytes. Our topic modeling identified 5 (out of 10) topics that have a stark opposing pattern in drugs targeting primitive vs differentiated cells. We further identified the most prominent immune markers associated with each of these topics. Conclusion: Our results recapitulate previously established links between myeloid differentiation state and selective efficacy to targeted compounds in AML. Notably, our recent findings informed by topics compromised of immune signatures further deconvolute these trends, pointing to new immune programs associated with cell state driven drug response.