Machine learning-driven glycolytic subtyping and exosome-based PKM splicing modulation overcome drug resistance in hyper-glycolytic myeloid leukemia.

This study comprehensively investigates the role of glycolysis in acute myeloid leukemia (AML) pathogenesis. Elevated glycolysis correlated significantly with poor prognosis. Bioinformatics identified HIF1A and MIF as key regulators and revealed two robust molecular subtypes: a high-glycolysis subtype (C1) associated with increased malignant cell proportion, activated oncogenic pathways, genomic instability, and inferior survival, and a low-glycolysis subtype (C2). These subtypes exhibited distinct drug sensitivities (C1 sensitive to panobinostat, MK-2206, 17-AAG; C2 sensitive to venetoclax) and predicted immunotherapy responses (C1 potentially benefiting more from anti-PD-1). An optimized 9-gene prognostic signature was developed using CoxBoost and StepCox algorithms, demonstrating accurate survival prediction across cohorts. Crucially, aberrant PKM2 overexpression was linked to imatinib (IM) resistance. A vivo-morpholino antisense oligomer (vMO) targeting the PKM exon 9-10 splice junction effectively converted PKM splicing from PKM2 to PKM1, inhibiting leukemia growth and reversing IM resistance in vitro and in vivo. To mitigate vMO toxicity, IL3-Lamp2b-engineered exosomes were developed, demonstrating efficient vMO loading, targeted delivery to leukemia cells, potent PKM splicing correction, significant IM resistance reversal, and minimal stromal cell toxicity. This work defines glycolysis-based AML subtypes with therapeutic implications and establishes engineered exosome-delivered vMO as a promising strategy to overcome drug resistance in hyper-glycolytic myeloid leukemia.