Dissecting the biology and clinical implications of aberrant DNA methylation in acute myelogenous leukemia

Abstract

Acute myeloid leukemia (AML) is a highly lethal malignancy characterized by unchecked expansion of immature myeloid blasts. While certain genetic and cytogenetic aberrations have been associated with chemotherapy response and disease risk, clinical outcomes remain heterogeneous. AML harbors relatively few somatic mutations compared to other cancers, however, it shows marked enrichment for epigenetic regulator alterations, and has been shown to harbor DNA methylation defects. My focus has been to dissect these epigenetic defects using high-throughput DNA methylation data. I first characterized two genome-wide hypermethylation signatures in AML: AML-CpG island methylator phenotype (A-CIMP+), and IDH-associated CIMP (I-CIMP+). While I-CIMP+ leukemias showed significant enrichments for mutations in IDH1 or IDH2, A-CIMP+ cases were mutation independent, and were best defined by their epigenetic defects, and associated transcriptomic changes. Importantly, A-CIMP+ leukemias had relatively favorable clinical outcomes, while I-CIMP+ patients did not. I next sought to characterize epigenetic defects involving demethylation of normally methylated genomic regions. I identified two distinct demethylator phenotypes (DMPs): DMP.1+ and DMP.2+. DMP.1+ AML was largely defined by mutations in DNMT3A, FLT3, and NPM1, while DMP.2+ leukemias harbored favorable-risk genomic rearrangements and a distinct gene expression profile. Both DMPs also carried prognostic information in AML; DMP.1+ cases had poor outcomes, while DMP.2+ patients tended to have favorable survival. Using both CIMP and DMP signatures, I then built an integrated epigenetic model for AML prognosis I termed MethylScore. The MethylScore algorithm was prognostic independent of age and cytogenetic risk in multivariate Cox regression models, suggesting that DNA methylation defects may augment existing clinical tools for risk stratification, and/or treatment selection. Finally, I explored whether DNA methylation signatures and genetic mutations could serve as biomarkers of response to epigenetic therapy, and found that DNA hypermethylation correlated with poor overall survival, and a gene mutation profile was associated with lack of complete remission after treatment with a DNA methylation inhibitor. These data provide evidence of distinct epigenetic signatures in AML that define transcriptionally, genetically, and clinically distinct populations that should be evaluated in future translational/clinical studies.