Epigenetic Phenocopying Expands Molecular Risk Assessment in Acute Myeloid Leukemia (Alliance)

Genetic profiling of leukemic cells forms the basis for risk stratification in acute myeloid leukemia (AML). Genetic markers in AML are used to make recommendations for distinct treatment approaches, such as remission consolidation with chemotherapy rather than stem cell transplant for patients with favorable risk genetics as defined by the current guidelines from the European LeukemiaNet (ELN). Yet, several limitations remain, such as overall rarity of many mutations, hierarchical complexity in cases with multiple mutations, conflicting reports of associations with outcomes for some genetic markers, and the absence of markers with prognostic significance in some patients. We have recently described genome-wide DNA methylation signatures that underlie biological features of AML cells and their utility to classify patients [Giacopelli et al. Genome Res. 2021;31:747]. The additional value of epigenetic information for risk assessment in AML in the context of current genetic and other clinical prognostic markers remains largely unexplored.

In this study, we have first developed a targeted approach for assessment of DNA methylation-based signatures and employed it to classify 1,262 patients with de novo AML enrolled onto the Cancer and Leukemia Group B/Alliance for Clinical Trials in Oncology studies. We successfully classified 87.5% of patients into one of 13 DNA methylation subgroups, termed 'epitypes' (Figure 1A,B). We found that epitypes are composed of a majority of patients with a specific genetic alteration (or a unique combination of alterations) in 9 of 13 epitypes. However, we also identified subgroups of patients that lack these highly recurrent alterations, and, instead, represent an epigenetic phenocopy of the dominant genetic feature (epiphenocopy). Epiphenocopies within epitypes were often enriched in specific lower frequency mutations, suggesting convergence of biological function(s) for these rare mutations. Epiphenocopying was also exhibited by patients displaying a DNA methylation signature involving hypomethylation of STAT DNA sequence motifs (termed the STAT hypomethylation signature, SHS) that mimicked FLT3-ITD mutations. Epitype and SHS DNA methylation signatures affected clinical outcomes separately to ELN risk groups (P<0.0001; Figures 1C,D), and FLT3-ITD status (P<0.0001; Figure 1E), respectively.

To broadly examine the prognostic power of DNA methylation signatures, we combined methylation-based classifications into a knowledge bank containing a compendium of other prognostic markers. Using a recently developed machine-learning approach [Gerstung et al. Nat Genet. 2017;49(3):332], we found that DNA methylation retained a high degree of importance for clinical outcomes, including overall survival (Figure 1F). Specifically, SHS and 6 epitypes were the most significant features negatively associated with overall survival along with age (P<0.0001; Figure 1G). SHS and epitype were among the most significantly associated features for all other endpoints, such as early death, remission and relapse (P<0.0001) and improved concordance between all predicted to actual outcomes. Finally, we used DNA methylation to reconstruct all 4 genetic features that define the ELN Favorable risk group. We found that patients with epiphenocopies of t(8;21)/inv(16) (CBF-AML), and CEBPA-dm had favorable outcome indistinguishable from that of patients with the respective genetic markers. NPM1-mutated, FLT3-ITD-negative patients displaying SHS-positivity had adverse risk despite lacking FLT3-ITD. Re-classifying patients with CBF-AML and CEBPA-dm epiphenocopies from more unfavorable risk groups into favorable group and excluding KMT2A/MLL-like and SHS-positive patients substantially improves the definition of favorable risk AML (P<0.0001; Figure 1H).

Our study demonstrates that DNA methylation signatures advance our understanding of the biology of AML and improve risk stratification through the identification of patients with epiphenocopies that mimic genetic mutations and other biological features. Use of DNA methylation signatures may lead to more effective assignment of patients to existing and novel therapeutic approaches.

Support: U10CA180821, U10CA180882, U24CA196171; https://acknowledgments.alliancefound.org; ClinicalTrials.gov Identifiers: NCT00048958 (8461), NCT00899223 (9665), and NCT00900224 (20202)

Figure 1

View largeDownload slide

Figure 1

View largeDownload slide

Close modal