MDS and secondary AML display unique patterns and abundance of aberrant DNA methylation

Abstract

Increasing evidence shows aberrant hypermethylation of genes occurring in and potentially contributing to pathogenesis of myeloid malignancies. Several of these diseases, such as myelodysplastic syndromes (MDSs), are responsive to DNA methyltransferase inhibitors. To determine the extent of promoter hypermethylation in such tumors, we compared the distribution of DNA methylation of 14 000 promoters in MDS and secondary acute myeloid leukemia (AML) patients enrolled in a phase 1 trial of 5-azacytidine and the histone deacetylase inhibitor entinostat against de novo AML patients and normal CD34(+) bone marrow cells. The MDS and secondary AML patients displayed more extensive aberrant DNA methylation involving thousands of genes than did the normal CD34(+) bone marrow cells or de novo AML blasts. Aberrant methylation in MDS and secondary AML tended to affect particular chromosomal regions, occurred more frequently in Alu-poor genes, and included prominent involvement of genes involved in the WNT and MAPK signaling pathways. DNA methylation was also measured at days 15 and 29 after the first treatment cycle. DNA methylation was reversed at day 15 in a uniform manner throughout the genome, and this effect persisted through day 29, even without continuous administration of the study drugs. This trial was registered at www.clinicaltrials.gov as J0443.

Trial registration: ClinicalTrials.gov NCT00101179.

Figure 1

MDS displays marked aberrant promoter DNA methylation. (A-C) Unsupervised clustering analysis of 14 MDS samples and 8 CD34⁺ normal controls with the use of 3 different algorithms: COA, PCA, and hierarchical clustering. (D) A plot of methylation difference between MDS cases and normal CD34⁺ controls (x-axis) versus statistical significance (y-axis) shows the marked asymmetry of the 2 branches, illustrating the overall tendency to higher methylation levels in the MDS cases. Red points mark probe sets that reached both criteria for differential methylation on our analysis [P < .0005 and absolute fold change in log(HpaII/MspI) > 1.5), whereas blue points mark probe sets that reached statistical significance but did not have an absolute change in log(HpaII/MspI) greater than 1.5. (E) Two-dimensional hierarchical clustering of genes differentially methylated between MDS and normal CD34⁺ controls, illustrated by a heatmap. Cases are represented in the columns and probe sets in the rows.

Figure 2

De novo AML presents with a lesser extent of aberrant promoter DNA methylation and aberrant hypomethylation. (A-C) Unsupervised clustering analysis of 15 de novo AML samples and 8 CD34⁺ normal controls with COA, PCA, and hierarchical clustering. (D) Plot of DNA methylation difference between AML cases and normal CD34⁺ controls (x-axis) versus statistical significance (y-axis). As before, red points mark probe sets that reached both criteria for differential methylation on our analysis [P < .0005 and absolute fold change in log(HpaII/MspI) > 1.5], whereas blue points mark probe sets that reached statistical significance but did not have an absolute change in log(HpaII/MspI) greater than 1.5. (E) Two-dimensional hierarchical clustering of genes differentially methylated between de novo AML and normal CD34⁺ controls identified by a supervised analysis, illustrated using a heatmap. Cases are represented in the columns and probe sets in the rows.

Figure 3

Underlying differences in DNA methylation distinguish MDS from de novo AML. (A-C) Unsupervised clustering analysis of 14 MDS cases and 15 de novo AML samples with 3 different methods: COA, PCA, and hierarchical clustering. (D) Plot of DNA methylation difference between MDS and de novo AML cases (x-axis) versus statistical significance (y-axis). Red points mark probe sets that reached both criteria for differential methylation on our analysis [P < .0005 and absolute fold change in log(HpaII/MspI) > 1.5], whereas blue points mark probe sets that reached statistical significance but did not have a change in log(HpaII/MspI) greater than 1.5. (E) Two-dimensional hierarchical clustering of genes differentially methylated between MDS and de novo identified by supervised analysis, illustrated with a heatmap. Cases are represented in the columns and probe sets in the rows. (F) Three-dimensional representation of PCA analysis, including normal CD34⁺ cells, de novo AML, and MDS. MDS samples can all be found the furthest away from normal CD34 cells, whereas de novo AML tend to cluster in an intermediate position between the other 2 groups of samples.

Figure 4

Genes aberrantly methylated in MDS contain common motifs for DNA binding proteins. Motif analysis with the FIRE algorithm showed a significant overrepresentation of several DNA motifs in promoter regions differentially methylated between MDS and normal CD34⁺ controls. The yellow color in the heatmap reflects overrepresentation of a motif in a given group, and the blue color represents underrepresentation of the motif. For each motif, we also specify (1) location of the motif; (2) statistical significance of the motif expressed as z scores; (3) robustness of the findings, ranging from 0 of 10 to 10 of 10; (4) conservation index; (5) seed that gave rise to the motif; and (6) name of the closest known motif in our motif database.

Figure 5

Marked promoter DNA hypomethylation is achieved after 15 days of treatment with 5AC. (A-C) Unsupervised clustering analysis (COA, PCA, and hierarchical clustering) of 9 MDS cases for which we obtained pretreatment and posttreatment (day +15) paired samples. (D) Plot of DNA methylation difference between baseline MDS and day +15 after treatment (x-axis) versus statistical significance (y-axis). Red points mark probe sets that reached both criteria for differential methylation on our analysis [P < .0005 and absolute fold change in log(HpaII/MspI) > 1.5], whereas blue points mark probe sets that reached statistical significance but did not have a change in log(HpaII/MspI) greater than 1.5. (E) Two-dimensional hierarchical clustering of genes differentially methylated between pretreatment and day +15 paired samples identified by supervised analysis is shown with a heatmap. Cases are represented in the columns and probe sets in the rows. (F) Density plots representing the relative frequency (y-axis) of promoters with different degrees of methylation (x-axis). Baseline (black) distributions are compared with day +15 of treatment with 5AC (red) and day +29 (blue). For 8 of 9 cases a clear shift toward more hypomethylated values is observed at day 15, and this shift persisted at day 29 for the cases for which we also had samples at that time point.