DNA methyltransferases (DNMTs) are epigenetic regulators targeted to the treatment of hematological malignancies.1, 2, 3, 4 Mutations in the DNA methyltransferase DNMT3A and high expression of its paralogue DNMT3B have been associated with inferior outcome in acute myeloid leukemia (AML) and other hematological malignancies.5, 6, 7, 8 Using a publicly available gene expression data set,9 we studied whether DNMT3B expression correlates with outcome in genetically well-defined AML subgroups. We first validated the expression data from the microarray by quantitative PCR, using 39 patient samples (Supplementary Figure S1A). DNMT3B micro-array analyses showed that the expression was not normally distributed among AML patients; in the quartile with highest expression, a larger variation in expression was observed compared to the three quartiles with relatively lower expression (Supplementary Figure S1B). The median DNMT3B expression in AML samples was significantly lower compared with that observed in normal bone marrow (NBM)-derived CD34+ cells (P<0.0001), while it was higher compared to NBM cells, but the latter difference was not statistically significant (Supplementary Figure S1B). Subsequently, we investigated the correlation between DNMT3B expression and overall survival (OS) and event-free survival (EFS). In univariate Cox regression analyses, continuous DNMT3B expression was significantly associated with poor survival (P<0.001 for both OS and EFS, data not shown). To visualize the prognostic impact, we performed Kaplan–Meier analyses on the four quartiles based on expression levels. The quartile including the patients with the highest DNMT3B expression, exhibiting the largest variation in expression, showed a significantly reduced OS and EFS compared to the other quartiles (Supplementary Fig. S2). As the survival between the lower three quartiles did not differ significantly, we grouped these patients together as having lower DNMT3B expression, whereas the remaining patients were ranked as having higher DNMT3B expression. Using these criteria, the 5-year OS and EFS were 17.2%±3.3% and 13.6%±3.0% for patients with higher DNMT3B expression compared to 43.8%±2.5% and 34.4%±2.4% for patients with lower DNMT3B levels (P<0.001, Figure 1a). We next performed a multivariate Cox regression analysis including known prognostic factors (including age >60 years; white blood cell counts >100 × 109/l; transplantation status; karyotypes t(8;21), t(15;17) and inv(16); nucleophosmin 1 (NPM1), FLT3-ITD, DNMT3A and double CEBPA mutations, and ecotropic viral integration site 1 (EVI1) overexpression), which revealed that higher DNMT3B expression carried an independent prognostic risk for both OS and EFS (hazard ratio (HR): 1.768, 95% confidence interval (CI): 1.384–2.260; P<0.001 and HR: 1.706, 95% CI: 1.342–2.168; P<0.001, respectively, Table 1), in line with a recently published study.7 In fact, higher DNMT3B expression showed a higher hazard ratio for OS than that of well-known adverse prognostic factors such as internal tandem duplications of the fms-related tyrosine kinase 3 (FLT3-ITD, HR: 1.675, 95% CI: 1.287–2.179; P<0.001) and overexpression of the EVI1 gene (HR: 1.430, 95% CI: 0.999–2.047; P=0.051).

Figure 1
figure 1

Higher DNMT3B expression correlates with inferior OS and EFS in AML. (a) Kaplan–Meier plots for OS and EFS showed that higher DNMT3B expression correlated significantly with a poor OS and EFS among AML patients (5-year OS: 17.2%±3.3% vs 43.8%±2.5% and 5-year EFS: 13.6%±3.0% vs 34.4%±2.4% for patients with higher and lower DNMT3B expression, respectively). (b) Higher DNMT3B expression predicted a very poor OS and EFS among patients with normal karyotype carrying NPM1 and FLT3-ITD mutations compared to patients with lower DNMT3B expression (5-year OS: 16.7±6.2% vs 47.6±8.8%, P=0.001 and 5-year EFS: 16.7±6.2% vs 39.0±8.6%, P=0.005 for patients with higher and lower DNMT3B expression, respectively). (c) Within the group of patients with normal karyotype that carries NPM1 mutations with high FLT3-ITD allelic burden, higher DNMT3B expression correlated with an extremely poor OS and EFS compared to patients with lower DNMT3B expression (5-year OS: 0%±0.0% vs 38.9%±12.9%, P<0.001 and 5-year EFS: 0%±0.0% vs 32.0%±12.4%, P<0.001 for patients with higher and lower DNMT3B expression, respectively). P values were determined with the log-rank test. In agreement with de Jonge et al.,13 high FLT3-ITD allelic burden was defined as an allelic FLT3-ITD/FLT3 ratio >1.

Full size image
Table 1 Data proving that DNMT3B expression is a prognostic factor in AML, particularly in NPM1+/ FLT3-ITD+ patients with normal karyotype
Full size table

We then studied the association of higher DNMT3B expression with specific AML subcategories. Higher DNMT3B expression was under-represented in FAB-M4 and mutually exclusive with the favorable karyotypes t(8;21) and inv(16), as shown in Supplementary Table S1. In contrast, higher DNMT3B expression was over-represented in the FAB-M1 subcategory (Supplementary Table S1), and predicted poor OS and a trend toward poor EFS in this group (data not shown). A significant association between higher DNMT3B expression and EVI1 overexpression and IDH2 mutations was also observed (Supplementary Table S1), but not with IDH1 mutations. DNMT3B expression did not predict clinical outcome in these subgroups (data not shown). We also observed that higher DNMT3B expression was associated with a normal karyotype (NK, P=0.017), and strongly associated with mutations in NPM1 (NPM1+, P<0.001) and FLT3-ITD (FLT3-ITD+, P<0.001, Supplementary Table S1). Patients with NK can be classified based on NPM1 and FLT3-ITD mutational status. NPM1 mutations, particularly in the absence of FLT3-ITD, display a favorable disease outcome, whereas the presence of a FLT3-ITD mutation is generally considered as an adverse prognostic factor.10 Remarkably, among patients with NK, higher DNMT3B expression did not significantly associate with NPM1−/FLT3-ITD− status and was under-represented in the NPM1+/FLT3-ITD− and NPM1−/FLT3-ITD+ groups, but was significantly over-represented in the NPM1+/FLT3-ITD+ group (Supplementary Table S2). Within the latter group, patients with higher DNMT3B expression showed a significantly worse outcome than patients with lower DNMT3B expression (Figure 1b). The 5-year OS and EFS of patients with higher DNMT3B expression were 16.7%±6.2% (P=0.001) and 16.7%±6.2% (P=0.005) compared to 47.6%±8.8% and 39.0%±8.6% in patients with lower DNMT3B expression. The survival of patients with co-occurring NPM1 and FLT3-ITD mutations is negatively influenced by high FLT3-ITD allelic burden.11, 12, 13 Higher DNMT3B expression was observed in both patients with low and those with high FLT3-ITD allelic burden (Supplementary Table S2). We next analyzed whether DNMT3B expression had an effect on survival among these subgroups. Higher DNMT3B expression did not exhibit a significant effect on OS and EFS among patients with low FLT3-ITD allelic burden (data not shown). However, patients with higher DNMT3B expression showed an extremely poor OS and EFS (5-year OS: 0.0%±0.0%, 5-year EFS: 0.0%±0.0%, P<0.001), compared to patients with lower DNMT3B expression (5-year OS: 38.9%±12.9%, 5-year EFS: 32.0%±12.4%) within the subgroup with high FLT3-ITD allelic burden (Figure 1c). In multivariate analysis, higher DNMT3B expression showed an independent prognostic value for OS and EFS, with a high hazard ratio both in the NPM1+/FLT3-ITD+ subgroup (HR: 4.850, 95% CI: 1.980–11.880) and among those patients with high FLT3-ITD allelic burden, indicating that the latter subgroup can be separated into two groups, one with an intermediate and the other with an extremely poor survival, based on DNMT3B expression (Table 1).

In conclusion, these data show that higher DNMT3B expression is a strong independent predictive factor for poor disease outcome in AML in general, and especially among NPM1+/FLT3-ITD+ patients with normal karyotype. Drugs that inhibit DNA methylation are tested for clinical efficacy in AML.1, 2, 3, 4 Because DNMT3B catalyzes DNA methylation, its expression level may affect the therapeutic sensitivity to these drugs. Thus, it will be interesting to investigate whether DNMT3B expression predicts therapy responses to treatments affecting DNA methylation.