The immunophenotypic landscape of newly diagnosed acute myeloid leukemia: Associations with genotype and outcomes

Despite widespread immunophenotyping in acute myeloid leukemia (AML), its immunophenotypic heterogeneity remains incompletely understood. Here, we applied state-of-the-art computational modeling to >50 million single cells from diagnostic bone marrow aspirates of 502 AML patients. Jointly modeling leukemic and non-malignant cells, we identified seven patient clusters with distinct immunophenotypic signatures. These clusters showed strong concordance with recurrent mutations, including NPM1 , IDH1/IDH2 , CEBPA , and t(8;21)(q22;q22); RUNX1-RUNX1T1 fusions, and enabled accurate prediction of genomic subgroups. Although these immunophenotypic signatures were not associated with therapy responses, loss of HLA-DR expression on leukemic blasts at diagnosis was independently associated with inferior outcomes in patients with chromatin-spliceosome mutations. In addition, we identified an immunophenotypic patient cluster comprising ∼11% of patients characterized by significant bone marrow lymphocyte enrichment where outcomes diverged depending on genetic risk. Together, these findings provide a comprehensive single-cell characterization of AML bone marrow and reveal clinically relevant genotype-phenotype relationships.