Abstract
Acute myeloid leukemia (AML) is characterized by a low 5-year survival rate. Despite having many clinical metrics to assess patient prognosis, there remain opportunities to improve risk stratification. We hypothesized that an underexplored resource to examine the prognosis of patients with AML is plasma metabolome. Circulating metabolites are influenced by patients' clinical status and can serve as accessible cancer biomarkers. To establish a resource of circulating metabolites in genetically diverse patients with AML, we performed an unbiased metabolomic and lipidomic analysis of 231 diagnostic AML plasma samples before treatment with intensive chemotherapy. Intriguingly, circulating metabolites were highly associated with the mutation status within the AML cells. Furthermore, lipids were associated with refractory status. We established a machine learning algorithm trained on chemotherapy-refractory-associated lipids to predict patient survival. Cox regression and Kaplan-Meier analysis demonstrated that the high-risk lipid signature predicted overall survival in this patient cohort. Impressively, the top lipid in the high-risk lipid signature, sphingomyelin (d44:1), was sufficient to predict overall survival in both the original data set and an independent validation data set. Overall, this research underscores the potential of circulating metabolites to capture AML heterogeneity and lipids to be used as potential AML biomarkers.