Heatmap representation of the aberrant DNA methylation signatures of specific clusters compared with a cohort of normal CD34+ hematopoietic cells obtained from healthy donors. Each row of the heatmap represents one probe set of the HELP array, and each column represents an AML patient (denoted by light brown bars) or a healthy donor (denoted by dark brown bars). (A) DNA methylation signatures for clusters with recurrent translocations, (B) DNA methylation signatures associated with abnormalities of CEBPA, (C) DNA methylation signatures for clusters presenting NPM1 mutations, (D) DNA methylation signatures for the five epigenetically defined clusters. The complete information for each cluster is contained in Table S3.

(A) Top scoring aberrantly expressed gene networks for each of the five epigenetically defined clusters. Genes overexpressed compared with normal CD34+ cells are colored in red, whereas downregulated genes appear in green.
(B) Comparative analysis of the most significantly deregulated canonical pathways of the five epigenetically defined clusters as captured by an integration of the aberrant epigenetic and gene expression signatures.

(A) Outline describing the steps for building the DNA methylation classifier. In a first step, 200 randomly selected patients were used to identify HELP probe sets that best predicted survival. The model was then tested on a different cohort of 95 patients (test set). Once the final model was selected, its performance in predicting survival was tested in an independent validation set consisting of 49 randomly selected cases.
(B) Kaplan-Meier curves for overall survival for the predicted groups in the independent validation set. Overall survival was compared between patients in an independent validation set that were predicted either alive or deceased by the DNA methylation classifier. (Cox proportional hazards p < 0.005, hazard ratio = 1.39, 95% CI = 1.10, 1.75).
(C) Kaplan-Meier curves for event-free survival for the predicted groups in the independent validation set. Event-free survival was compared between patients in an independent validation set that were predicted either alive or deceased by the DNA methylation classifier (Cox proportional hazards p < 0.0002, hazard ratio: 1.53, 95% CI = 1.21, 1.93).
(D) Kaplan-Meier curves for overall survival for the predicted groups in the combined test and independent validation sets. Overall survival was compared between patients in the combined test and independent validation sets that were predicted either alive or deceased by the DNA methylation classifier. (Cox Proportional hazards p < 0.000003, hazard ratio: 1.34, 95% CI = 1.18, 1.51).
(E) Multivariate Cox proportional hazards regression model for the DNA methylation predictor, age, cytogenetic risk, NPM1 mutation, FLT3-ITD and CEBPA mutations. For additional information please see Tables S5, S6, S7, and S8, as well as supplementary R scripts.

Heatmap representation of a correlation matrix in which each patient’s DNA methylation profile is correlated with that of the other patients in the data set. Patients are ordered according to the unsupervised analysis (hierarchical clustering) results, so that highly correlated patients are located next to each other. Parallel bars on the right of the heatmap have been used to indicate the principal cytogenetic and molecular findings for each patient. Cluster membership and cluster feature summaries are described on the left of the heatmap. Additional information is shown in Figure S1.

(A) Left: Kaplan-Meier curves for overall survival for the favorable risk clusters 1 (inv(16)) and 4 (CEBPA-dm), and the novel epigenetically defined clusters. For plotting simplicity curves for clusters 3 (t[8;21]), cluster 5 and cluster 15 were not included in the plot. Figure S3 shows a Kaplan-Meier plot including all the clusters in the overall survival analysis. Right: Kaplan-Meier curves for overall survival for the favorable risk clusters 1 (inv[16]) and 4 (CEBPA-dm), and the NPM1 clusters. For plotting simplicity, curves for clusters 3 (t[8;21]) and NPM1 cluster 14 were not included in the plot. Figure S3 shows a Kaplan-Meier plot including all the clusters in the overall survival analysis.
(B) Kaplan-Meier curves for overall survival (left) for the five novel clusters. Right: Table summarizing the multivariate Cox proportional hazards regression model, using cluster 1 (inv[16]) as the referent cluster. Additional Kaplan Meier plots are shown in Figure S3.

(A) Heatmap representation of the common 45 gene signature consistently aberrantly methylated in AML. Each row represents a probe set from the HELP microarray and each column represents a sample.
(B) Boxplots of gene expression levels in 4 representative genes from the 45 gene common epigenetic signature demonstrating downregulation of expression in the AML samples compared with normal CD34+ cells. The list of genes is shown in Table S4.