Abstract
Azacitidine (AZA) is a DNA methyltransferase inhibitor and epigenetic modulator that can be an effective agent in combination with chemotherapy for patients with high-risk acute myeloid leukemia (AML). However, biological factors driving the therapeutic response of such hypomethylating agent (HMA)-based therapies remain unknown. Herein, the transcriptome and/or genome-wide 5-hydroxymethylcytosine (5hmC) is characterized for 41 patients with high-risk AML from a phase 1 clinical trial treated with AZA epigenetic priming followed by high-dose cytarabine and mitoxantrone (AZA-HiDAC-Mito). Digital cytometry reveals that responders have elevated Granulocyte-macrophage-progenitor-like (GMP-like) malignant cells displaying an active cell cycle program. Moreover, the enrichment of natural killer (NK) cells predicts a favorable outcome in patients receiving AZA-HiDAC-Mito therapy or other AZA-based therapies. Comparing 5hmC profiles before and after five-day treatment of AZA shows that AZA exposure induces dose-dependent 5hmC changes, in which the magnitude correlates with overall survival (p = 0.015). An extreme gradient boosting (XGBoost) machine learning model is developed to predict the treatment response based on 5hmC levels of 11 genes, achieving an area under the curve (AUC) of 0.860. These results suggest that cellular composition markedly impacts the treatment response, and showcase the prospect of 5hmC signatures in predicting the outcomes of HMA-based therapies in AML.