Epigenetic Signatures Identify New Clinically Relevant Subtypes and Define Gene Regulatory Patterns in Patients with Acute Myeloid Leukemia (AML)

AML is heterogeneous group of diseases with variable clinical outcomes. While cytogenetics, molecular markers and gene expression profiling can help to classify these patients, they still cannot fully explain the biology and clinical outcomes of the disease. Epigenetic gene deregulation is a hallmark of cancer and our preliminary data suggest that epigenetic signatures are critical determinants of cellular phenotype in AML. Therefore, we hypothesized that aberrant epigenetic regulation of genes would provide critical insight into the biological complexity of AML and identify new and clinically relevant disease subtypes. We studied genome wide DNA methylation in a cohort of 295 patients from HOVON multicenter clinical trials using the HELP assay, which measures with >95% accuracy the abundance of DNA methylation at ~50,000 CpG sites covering ~13,000 promoter regions. Median follow-up was 18.2 months (range=0.1–214.5); median age: 48.1 years (range=15.8–75). Unsupervised analysis using hierarchical clustering (Pearson correlation distance with Ward’s clustering method) segregated the AMLs into 16 well-defined epigenetic clusters. Cluster 1 consisted 100% of patients with acute promyelocytic leukemia (n=6); 100% of cluster 4 harbored CEBPA mutations (n=14); 21/23 patients in cluster 6 carried an inv(16); clusters 7 and 9 were enriched for cases carrying the NPM1 mutation (#7: 80% NPM1+ and #9: 96%) and cluster 12 was enriched for t(8;21) AMLs (18/23). Most of the clusters however define previously unknown biological entities. Next we used a supervised analysis and identified the differentially methylated genes and gene networks that define each cluster, which revealed previously unknown biological differences among these patients. Moreover, Kaplan-Meier survival analysis revealed significant differences in event-free survival (EFS) and overall survival (OS) for the 8 clusters that consisted of >20 patients (clusters 5, 6, 7, 8, 9, 11, 12, and 14), which includes clusters that represent previously unidentified AML subtypes. The inv(16) and t(8;21) containing clusters (i.e. #6 and #12) demonstrated a 2-year EFS of 48% and 58%, respectively, compared to 2-year EFS ranging from 19%–44% for all other clusters (p=0.002 by log-rank test) and a 2-year OS of 70% and 61%, respectively, compared to 2-year OS ranging from 25%–50% for all other clusters (p=0.008 by log-rank test). After adjustment for age, cytogenetic risk, NPM1 mutation, and FLT3-itd status in a multivariate cox proportional hazards regression model, differences in EFS and OS remained between clusters i.e. multivariate analysis (utilizing cluster 12 as reference) showed that clusters 9, 5, 8 and 11 demonstrated hazard ratios for poor events of 3.2 (95% CI=1.0, 10.6; p=0.06), 3.2 (95% CI=1.1, 9.1; p=0.03), 3.4 (95% CI=1.2, 9.8; p=0.03) and 3.6 (95% CI=1.2, 10.3; p=0.02), respectively. Similarly, clusters 9, 8 and 11 demonstrated hazard ratios for mortality of 4.7 (95% CI=1.1, 19.8; p=0.03), 4.1 (95% CI=1.1, 15.2; p=0.03) and 3.7 (95% CI=1.0, 13.6; p=0.05), respectively. Interestingly none of these clusters could be entirely explained by any of the known molecular or cytogenetic markers. Clusters 9 and 5 consisted mainly of cases with normal karyotypes, while #8 and #11 grouped cases with a variety of karyotypes. Furthermore, cluster 9 was associated with a worse outcome despite the fact that 24/25 cases were NPM1+, only 11 of which also presented the poor risk association with FLT3-itd. An analysis restricted to the 125 cases with normal karyotype (NK-AML) segregated them into 2 main clusters, one enriched for NPM1+ cases (81.9%) and the other not (29.6% NPM1+) (Fisher exact test: p-value <2.6e-9). 13/14 cases with CEBPA mutations grouped in the NPM1 cluster, while the 52 FLT3-itd cases were equally distributed among the two clusters. A supervised analysis of NPM1+ NK-AMLs vs. cases without the mutation revealed a 167-gene signature of genes that were uniformly hypermethylated in NK-AML that did not carry the NPM1 mutation, suggesting that non-NPM1 NK-AML patients display common features indicative of a new AML subtype. These data show that rigorous analysis of epigenetic gene regulation in AML identifies novel and biologically relevant subgroups of AML with prognostic significance, and establishes the capture of epigenetic signatures as a new paradigm to improve understanding of disease pathogenesis and clinical behavior.