Methylome profiling reveals context-dependent chemo-resistance mechanisms and enhances risk stratification in AML

Acute Myeloid Leukemia (AML) is the second most lethal hematologic malignancy, with approximately 30% of patients being refractory to first-line induction therapy. While current risk stratification systems, such as European LeukemiaNet (ELN), use cytogenetic and molecular markers to guide treatment decisions, their predictive accuracy remains suboptimal, particularly in forecasting treatment response. In this study, we performed integrated methylome and transcriptome analysis of diagnostic samples from 146 AML patients, including methylome data on 25 patients, to investigate the molecular basis underlying the discrepancy between predicted and actual treatment outcomes. We utilized MSP-PCR, cell proliferation assays, and colony formation assays to evaluate the effects of DNA methylation on PTX4 expression and function in AML. A comprehensive machine-learning pipeline was implemented to develop a methylome-based classifier predicting primary refractory disease. Unsupervised analysis revealed that while genomic backgrounds strongly influence molecular profiles, treatment response patterns frequently diverge from predictions based on cytogenetic and mutational risk classifications. We identified PTX4 as a novel tumor suppressor gene (TSG) silenced by DNA hypermethylation in patients with adverse-risk AML. Within individual AML subtypes, comparison of refractory/relapsed (RR) cases versus those achieving complete remission (CR) uncovered distinct resistance mechanisms. Furthermore, analysis of hematopoietic developmental trajectories revealed that RR cases exhibit altered epigenetic programming, characterized by preferential methylation changes in regulatory regions, such as polycomb-repressed chromatin states. Based on these insights, we developed a methylome-based classifier (AUROC = 0.86) that addresses 32.2% misclassification rate observed with ELN criteria. These results represent both context-dependent and unifying mechanism of treatment resistance that is independent of genetic background. These findings highlight the limitations of current genetic-based risk assessments and underscore the potential of incorporating epigenetic profiling, such as methylome analysis, to better understand more accurately, predict treatment response and guide therapeutic strategies in AML.