Genome-wide methylation profiling in decitabine-treated patients with acute myeloid leukemia

Abstract

The outcome of older (≥ 60 years) acute myeloid leukemia (AML) patients is poor, and novel treatments are needed. In a phase 2 trial for older AML patients, low-dose (20 mg/m(2) per day for 10 days) decitabine, a DNA hypomethylating azanucleoside, produced 47% complete response rate with an excellent toxicity profile. To assess the genome-wide activity of decitabine, we profiled pretreatment and post treatment (day 25/course 1) methylomes of marrow samples from patients (n = 16) participating in the trial using deep-sequencing analysis of methylated DNA captured by methyl-binding protein (MBD2). Decitabine significantly reduced global methylation compared with pretreatment baseline (P = .001). Percent marrow blasts did not correlate with global methylation levels, suggesting that hypomethylation was related to the activity of decitabine rather than to a mere decrease in leukemia burden. Hypomethylation occurred predominantly in CpG islands and CpG island-associated regions (P ranged from .03 to .04) A significant concentration (P < .001) of the hypomehtylated CpG islands was found in chromosome subtelomeric regions, suggesting a differential activity of decitabine in distinct chromosome regions. Hypermethylation occurred much less frequently than hypomethylation and was associated with low CpG content regions. Decitabine-related methylation changes were concordant with those previously reported in distinct genes. In summary, our study supports the feasibility of methylome analyses as a pharmacodynamic endpoint for hypomethylating therapies.

Figure 1

Global methylation levels for all pre- and postdecitabine treatment samples. Global methylation levels were measured for each sample by the global methylation indicator (GMI; see “Global DNA methylation comparison”). The histograms represent the mean GMI values in the sample groups (n = 16). Error bars represent SD within the sample group. The P value for the pretreatment versus posttreatment comparison was calculated with a paired sample nonparametric test (Wilcoxon signed rank test).

Figure 2

Pretreatment and posttreatment methylation changes by genomic features. Error bars represent SEM. *P < .05 using a paired Wilcoxon signed sum.

Figure 3

Chromosomal locations of DMRs in pre- versus postdecitabine treatment AML samples. The 4 lanes above each chromosome symbol show the number of DMRs in 400-kb bins as a heat map for all genomic features (ALL), CpG islands (CGI), Gene deserts (GD), and RefSeq gene bodies (RG), respectively (intensity in each lane is normalized to the bin with most DMRs in that lane). These 4 features were selected as they depict genomic features with diverse CpG densities (see Table 1 for other features evaluated in this study but not shown here). Comparing the 4 heat map tracks shown here (ALL vs CGI, or GD, or RG), the most significant number of DMRs induced by decitabine treatment is contributed by the CpG islands. This effect of decitabine is most pronounced at the chromosome ends (exception: chromosome 3, chromosome 15, and chromosome Y). The green track below the chromosome symbol shows the distribution of CpG islands (CGIs) across the chromosome (independently of their methylation status). Because CpG islands are also enriched in chromosome ends, we performed Pearson χ² analysis of observed DMRs versus expected DMRs for every 400-kb bin in the genome (“Chromosomal localization of DMRs”) to test whether enrichment of DMRs within the chromosome ends is related to the higher density of CpG islands in that region or as a nonrandom effect of decitabine treatment. The resulting P values are plotted in the χ² track in blue. For all chromosomes except 2, 3, 15, 19, and 21, the P values were significant (P ≤ .05) at least at one end of the chromosome, indicating significant clustering of DMRs not accounted for by the dense distribution of CpG islands in that region of the chromosome.