Metabolic Subtypes Predict Treatment Response in Acute Myeloid Leukemia: A Pilot Study

To establish a dynamic metabolic subtyping system for acute myeloid leukemia (AML) based on longitudinal metabolomics and multi-omics integration, and to evaluate its ability to predict treatment response.

We enrolled 29 AML patients and performed untargeted metabolomics on pre‑ and post-treatment serum samples. Based on finite cyclic combinations of metabolic pathways, pre-treatment patients were classified into G1 (glycolysis/gluconeogenesis/TCA cycle) and G2 (fatty acid/folate biosynthesis). Post‑treatment patients were categorized into TG1 (α-linolenic acid metabolism, pantothenate/CoA biosynthesis) and TG2 (purine metabolism, cysteine/methionine metabolism). Baseline transcriptome data were integrated and validated in three external cohorts (Beat2, GSE6891, GSE37642; total n=994). Single cell and spatial transcriptomics were used to investigate cellular heterogeneity and resistance mechanisms.

The complete remission (CR) rate was significantly higher in G2 (71%) than in G1 (29%). After treatment, TG2 showed an 83% CR rate versus 17% in TG1. All patients transitioning from G2 to TG2 achieved CR (100%), whereas those from G1 to TG2 maintained poor response. Baseline metabolic subtype was an independent predictor of treatment response (p<0.05). To explore candidate cellular niches linking G1 metabolic features to chemotherapy resistance, we further integrated single-cell and spatial transcriptomic analyses. The results revealed co-localization of CA2-high immature erythrocytes and neutrophils in non-CR patients.

Dynamic metabolic subtyping (G1/G2, TG1/TG2) strongly correlates with AML treatment response. The co-localization of CA2-high immature erythrocytes and neutrophils in the bone marrow microenvironment of non-CR patients provides a candidate cellular and microenvironmental correlate of the G1 metabolic phenotype, suggesting a mechanistic link between systemic metabolic reprogramming and chemotherapy resistance. Collectively, these findings highlight the potential of combining dynamic metabolic subtyping with identified microenvironmental features to guide precision therapy.