Decoding macrophage polarization heterogeneity in post-chemotherapy AML microenvironments

Acute myeloid leukemia (AML) treatment faces significant chemoresistance and high relapse rates. Macrophage polarization dynamics within the bone marrow microenvironment critically influence therapeutic response, yet their precise roles in chemoresistant AML remain largely unexplored. This study aimed to define the macrophage polarization landscape in chemoresistant AML, elucidate the functional roles and molecular regulators of specific subtypes, and identify prognostic biomarkers and therapeutic targets. Primary AML bone marrow samples obtained after standard 7 + 3 chemotherapy (GSE198681; n = 5 patients: three complete remission and two non-complete remission) underwent Scupa-based single-cell RNA sequencing analysis to classify five macrophage polarization states (Mac.a–e). hdWGCNA module hub genes were evaluated for prognostic relevance in independent bulk AML cohorts (GEO-AML training, n = 943; TCGA-AML validation, n = 132). TCGA-LAML bulk data were analyzed using CIBERSORTx deconvolution with a custom Mac.a–e signature matrix derived from scRNA-seq. Five macrophage polarization states (Mac.a–e) were identified in post-chemotherapy AML scRNA-seq (GSE198681). At the cell level, mean Scupa polarization scores for Mac.a, Mac.b, and Mac.c were higher in non-complete remission (non-CR) than complete remission (CR) samples. GO/KEGG enrichment revealed distinct functional programs: Mac.a (FoxO/PI3K-Akt signaling), Mac.b (NF-κB and inflammatory pathways), and Mac.c (oxidative phosphorylation), with Mac.b showing the highest pro-inflammatory Hallmark activity. hdWGCNA identified ten co-expression modules toward Mac.a-e whose hub genes were used to construct a 32-gene prognostic signature validated in GEO-AML (HR = 2.51, p < 0.001) and TCGA-AML (HR = 1.66, p < 0.05). In TCGA-LAML bulk data, CIBERSORTx-inferred Mac.d proportions showed the strongest association with LAG-3 expression ( r = 0.581, p < 0.001), and Mac.e correlated with hypermutator signatures (mutSig2: r = 0.86, p < 0.001). This study defines Scupa-based Mac.a–e polarization states in post-chemotherapy AML and links Mac.a/b/c to chemotherapy response in exploratory analyses. Integrative analyses yielded prognostic biomarkers and candidate macrophage polarization–anti–LAG-3 combination strategies.