DNMT3A mutations are recurrent in patients with AML and are associated with poor event-free and overall survival, independently of age and the presence of FLT3 or NPM1 mutations and regardless of the type of mutation or genetic location. This finding strongly suggests that DNMT3A mutations are probably relevant to the pathogenesis of AML. DNMT3A mutations do not cause genomic instability (since most genomes with mutations have a normal cytogenetic profile and an unaltered number of total mutations), do not alter total 5-methylcytosine content or global patterns of methylation, and do not dramatically alter gene expression. Currently, the only clue regarding the pathogenetic mechanism is strong selection against DNMT3A mutations (and also IDH1/2 and NPM1 mutations) in patients who have a favorable-risk cytogenetic profile, suggesting a biologic relationship that is not random ().
In patients with AML, there are two major classes of DNMT3A
mutations. The first class is the highly recurrent set of mutations at R882, as recently described in 3 of 74 AML samples tested by Yamashita et al.18
(The low prevalence of this variant in that study could reflect a different sample population or a lower sensitivity of mutation detection.) The second class is represented by all the other mutations in this gene. The locations of mutations in DNMT3A
are similar to those of DNMT3B
mutations associated with the immunodeficiency, centromere instability, and facial anomalies (ICF) syndrome.19,20
Recurrent mutations at a single amino acid position suggest a gain-of-function mechanism, although widely divergent mutations at many positions in a gene generally suggest loss of function, a pattern seen for many classic tumor-suppressor genes (e.g., TP53 and BRCA1). Indeed, several of the non-R882 mutations clearly cause loss of function in DNMT3A. A gain-of-function property that is induced by the R882 mutation is not yet apparent.
The MeDIP-chip experiments revealed a significant reduction in DNA methylation at 182 genomic locations, suggesting that R882 mutations may act in a dominant-negative fashion to reduce the methyltransferase activity of the enzyme. This hypothesis is supported by the fact that all R882 mutations are heterozygous and by observations that this mutation reduces methyltransferase activity in vitro.18,21
Both the R882H and R882C mutations are caused by a C-to-T transition at a CpG dinucleotide (R882H on the noncoding strand and R882C on the coding strand), suggesting that these mutations may be caused by the deamination of methylcytosine on either strand of this CpG dinucleotide.22
mutations are predicted to cause changes in the DNA binding groove of DNMT3A, and some are predicted to change its interaction with DNMT3L (Fig. 16 in the Supplementary Appendix
). However, it is possible that R882 mutations alter functions of DNMT3A that are not yet fully understood, including its ability to bind to other proteins involved in transcriptional regulation and localization to chromatin regions containing methylated DNA.23-26
In any case, all DNMT3A
mutations are associated with poor overall survival, suggesting that they have an important common effect on the potential of AML cells to cause lethal disease.
The association between DNMT3A
mutations and mutations in the four other most commonly mutated genes in AML (FLT3, NPM1, IDH1
, and IDH2
) is evident, as shown in . A large proportion of patients with an intermediate-risk cytogenetic profile had mutations in one or more of these genes (mutation group B in ). However, many patients with an intermediate- or adverse-risk cytogenetic profile had no mutations in any of these genes (mutation group C), which was not a random finding (P<0.001). Patients in mutation group C had outcomes similar to those of patients in mutation group B with one or two mutations (Fig. 17 in the Supplementary Appendix
) but may have a unique set of driver mutations. These may include some of the less commonly mutated AML genes (e.g., CEBPA, RUNX1, NRAS
, and KRAS
), although whole-genome sequencing may ultimately be required to identify the key mutations in this group.
Remarkably, no DNMT3A
mutations were found in the group of 79 patients with a favorable-risk cytogenetic profile ( and ), which includes patients with t(15;17), t(8;21), and inv(16). Mutations in NPM1, IDH1
, and IDH2
were not detected in these patients either, a finding that is consistent with data reported in other studies.4,27-30
The virtual exclusion of mutations in these four genes in patients with a favorable-risk profile is not random and may reflect the leukemogenic properties of the fusion proteins created by these chromosomal rearrangements. The PML-RARA fusion protein, which is created by t(15;17), physically interacts with DNMT3A
, and AML-ETO, which is created by t(8;21), interacts with DNMT1;
both fusion proteins alter the methylation of specific promoters.31-33
regulate telomere function, and all-trans
retinoic acid, which is part of the therapy for patients with t(15;17), down-regulates DNMT3A
Together, these data suggest that DNMT3A
mutations and the favorable-risk fusion oncogenes (e.g., PML-RARA and AML-ETO) may not be found in the same AML genomes because they both act to alter the function of DNA methyltransferases and are therefore redundant. However, the outcomes for patients with DNMT3A
mutations and those with a favorable-risk cytogenetic profile are dramatically different, for reasons that are currently unclear.
mutations do not change 5-methylcytosine content in AML genomes, and the R882H mutation appears to minimally perturb the methylation of CpG islands. We did not detect changes in DNA methylation that were directly correlated with local changes in gene expression. These data might suggest that DNMT3A
mutations do not directly affect the cytosine methyltransferase properties of DNMT3A
. However, a recent study has suggested that DNMT3A
may alter the methylation of nonpromoter-associated CpG regions, affecting gene expression indirectly.36
Many further experiments will be required to define the precise mechanisms by which these mutations act.
The discovery of highly recurrent mutations in DNMT3A
may provide a new tool for the classification of intermediate-risk AML. If these data are reproduced in other series, clinical trials designed to assess the effects of early intensification of treatment in patients with DNMT3A
mutations may be warranted. In our small series, allogeneic transplantation provided a significant benefit for patients with DNMT3A
mutations (Fig. 18 in the Supplementary Appendix
). The careful analysis of many additional AML samples with unbiased genomic methods may reveal changes in methylation or expression that help to define the mechanisms of action of DNMT3A