Bone marrow tumor microenvironment profiling predicts distinct immunosuppressive phenotypes and immunotherapy potential in AML patients

The immune landscape of acute myeloid leukemia (AML) is poorly understood, and its myeloid origins complicate the distinction between pathological and immunoregulatory components. We refined the AML microenvironment (AME) classification, demonstrating distinct immunophenotypes from AML lineages and maturation stages. We developed distinct 13-gene megakaryocytic/erythroid (MK/Ery) and 16-gene myelomonocytic/monocytic (ML/Mo) polygenic markers and validated them in adult and pediatric AML patient cohorts, including single-cell RNA-sequencing data. To identify immunoregulatory factors, the AME composition was predicted by the xCell algorithm, immune dysfunction was computed by TIDE, and differential gene expression analyses identified candidate genes. To validate our findings, we studied the effects of MK/Ery-like and ML/Mo-like AML cell lines on CD8⁺ T-cells/AML cells in co-culture assays, using models identified by Celligner matching AML patient blasts. Patients with high MK/Ery expression (MK/Ery^High) exhibited a dysfunctional microenvironment with increased pro-inflammatory cytokines, increased T-cell infiltration, and upregulated immune checkpoints, particularly CD274 (PD-L1). Single-cell RNA-seq confirmed that CD274 overexpression originated from malignant subclones with MK/Ery-like phenotypes. Conversely, AML with high ML/Mo expression (ML/Mo^High) displayed a T-cell-depleted niche enriched in myeloid-derived suppressive elements, including M2 macrophages, VISTA, and galectins. MK/Ery-like (e.g., OCI-M1, HEL) and ML/Mo-like (e.g., MUTZ3, MONO-MAC-1) cells suppressed T-cell proliferation in co-culture. Pharmacological PD-1/PD-L1 blockade with the small-molecule inhibitors BMS-1166 and BMS-1001 abrogated HEL-mediated T-cell proliferation inhibition. Transcriptomic data, single-cell analyses, and functional co-culture experiments reveal two AMEs: T–cell–rich yet dysfunctional vs. myeloid-driven and immunosuppressive. This refined categorization overcomes the traditional hot–cold classification to tailor future AML immunotherapies.