Proteomics Reveals 8 AML Subtypes and an Aging Signature That Predicts Survival

Acute myeloid leukemia disproportionately strikes and kills older adults—patients over 65 account for over 60% of cases and nearly 76% of deaths—yet the protein-level biology driving this age-related vulnerability has remained largely unexplored. This study addresses that gap by conducting one of the largest AML multiomics analyses to date, integrating proteomics, phosphoproteomics, genomics, transcriptomics, and ex vivo drug screening from 374 newly diagnosed patients.

Using similarity network fusion (SNF) clustering on 10,016 quantified proteins, the team identified 8 proteomic subtypes (S1–S8) with distinct clinical and molecular properties. Several subtypes mapped cleanly onto WHO-defined genetic entities: S1 correlated with CEBPA mutations, S3 with myelodysplasia-related AML (AML-MR), S4 and S6 with PML::RARA fusions, S5 with NPM1 mutations, and S8 with CBFB::MYH11 fusions. Subtypes S2 and S7 were mixed and harder to classify genomically, yet showed proteomic coherence. UMAP visualization confirmed that major genomic subtypes cluster distinctly at the protein level, validating the biological relevance of the classification.

A central finding was the identification of age-associated protein signatures in AML hematopoiesis. Locally weighted regression modeling of protein abundance across patient ages, followed by fuzzy c-means clustering, revealed that megakaryocyte/platelet-related and immune-related protein networks show the most striking age-associated changes—consistent with the group's prior RNA-level findings. Phosphosite data similarly showed age-related patterns, adding a layer of post-translational regulatory information. Subtypes S2, S3, and S7 showed elevated megakaryocytic protein signatures and were enriched in older patients with worse outcomes.

Building on these observations, the researchers constructed a Hematopoietic Aging Score (HAS) based on 19 proteins that were both age-associated and prognostically significant in Cox regression analysis. Higher HAS correlated with AML-MR, NPM1 mutations, and clonal hematopoiesis-related mutations (e.g., DNMT3A, TET2, ASXL1), and independently predicted inferior overall and event-free survival. This score offers a quantitative, proteomic readout of biological aging in AML beyond chronological age.

The study also incorporated ex vivo drug sensitivity screening of 41 drugs and 39 combinations across 102 non-APL samples, linking proteomic subtypes to differential drug responses and suggesting subtype-specific therapeutic vulnerabilities. Overall, this work enriches the current AML classification framework, establishes protein-level aging hallmarks with prognostic utility, and underscores the value of proteomics as a complementary dimension to genomic and transcriptomic classification in hematologic malignancies.