Clinical impact of DNMT3A mutations in younger adult patients with acute myeloid leukemia: results of the AML Study Group (AMLSG)

In acute myeloid leukemia (AML), mutations have recently been identified in a number of genes encoding for proteins involved in epigenetic regulation, such as mutations in the ten-eleven-translocation 2 (TET2), isocitrate dehydrogenase 1 and 2 (IDH1/2), additional sex comb-like 1 (ASXL1), histone-lysine N-methyltransferase (EZH2), and DNA (cytosine-5)-methyltransferase 3 α (DNMT3A) genes.^1,2  This discovery suggests a novel class of mutations that complements the class I and class II mutations affecting genes primarily involved in proliferation and hematopoietic differentiation, respectively.^3,4 

Mutations of DNMT3A (DNMT3A^mut) were discovered by array and next-generation sequencing approaches.^5-7  In the study by Ley et al,⁷  sequencing of 281 AML patients revealed DNMT3A^mut in 22%. Mutations were associated with older age, intermediate-risk cytogenetics, mutations of NPM1 (NPM1^mut) and IDH1 (IDH1^mut), FLT3 internal tandem duplications (ITDs), and poor outcome.⁷  Subsequent studies revealed similar results with regard to mutation frequency and prognostic impact. However, in most studies, patient populations were small and selected.^8-16 

DNMT3A is located in chromosomal band 2p23 and, together with DNMT3B and DNMT1, is a member of a DNA methyltransferase (MTase) family.¹⁷  It has three conserved domains: the PWWP domain targeting the enzyme to nucleic acids, the cystein-rich PHD zinc-finger domain interacting with unmodified histone H3, and the highly conserved catalytic domain representing the MTase domain in the C-terminal region. It has high ubiquitous expression in embryonic tissues and undifferentiated embryonic stem cells.^17,18 

By catalyzing the conversion of cytosine to 5-methylcytosine, DNMT3A (and DNMT3B) add methyl groups to unmodified DNA; DNMT1 maintains existing DNA methylation after cell division.¹⁹  Deletion of Dnmt3a in mouse hematopoietic stem cells resulted in inhibition of differentiation, but Dnmt3a deficiency alone appears insufficient to generate AML in this model.²⁰ Dnmt3a-null hematopoietic stem cells show both increased and decreased methylation at distinct loci,²⁰  a finding which is in line with other studies: Ley et al could not identify a clear gene expression signature or global or focal alterations in DNA methylation patterns caused by the mutations.⁷  In contrast, in the study by Yan et al,⁵ DNMT3A^mut were associated with increased expression and hypomethylation of several HOXB genes.

Almost all DNMT3A^mut are heterozygous, and more than two thirds of the mutations cluster at codon R882 in exon 23 in the MTase domain.^7-12  R882 directly inhibits enzymatic activity of DNMT3A, and overexpression of R882 prompted proliferation of 32D cells.^5,6 

The increasing number of molecular markers in AML contributing to tremendous disease heterogeneity poses important clinical and practical challenges.¹  With the identification of new molecular markers, it is becoming more and more difficult to integrate this complex genetic information into a valid and informative prognostic marker model. Currently, only three molecular markers (NPM1^mut, CEBPA^mut, FLT3-ITD) have an impact on patient management and have been included in the European LeukemiaNet (ELN) classification.²¹ 

In this study, we analyzed the clinical phenotype and prognostic impact of DNMT3A^mut in the context of established molecular markers in 1770 younger adult patients treated in prospective treatment trials of the German-Austrian AML Study Group (AMLSG).

Analyses of diagnostic bone marrow (BM) and/or peripheral blood (PB) samples were performed in 1770 younger adult (age 18 to 60 years) patients with AML enrolled on the prospective AMLSG multicenter treatment trials AML HD98A (692 of 870 patients [80%]; NCT00146120) and AMLSG 07-04 (1078 of 1112 patients [97%]; NCT00151242; data supplement).^22,23 

All patients gave informed consent for both treatment and genetic analysis according to the Declaration of Helsinki. Approval was obtained from the ethical review boards of the participating AMLSG institutions. Patients were molecularly studied for the presence of the recurring gene fusions RUNX1-RUNX1T1, CBFB-MYH11, MLLT3-MLL, and PML-RARA and for mutations in FLT3 (ITDs and tyrosine kinase domain [TKD] mutations at codons D835 and I836), NPM1, IDH1/IDH2, and double mutated CEBPA (CEBPA^dm).^24-26 

The availability of a diagnostic BM and/or PB sample for DNMT3A mutation analysis was the only criterion for including patients in this study. The coding region of DNMT3A (exons 11 to 23) was amplified from DNA by polymerase chain reaction (PCR) using exon-intron flanking primer pairs and was followed by direct sequencing of purified PCR products according to standard protocols (Table 1; data supplement). In the first cohort of 95 patients, exons 2 to 10 were additionally analyzed; here, only one mutation in exon 4 was found. All DNMT3A sequence variations were validated by repeated PCR and sequencing analysis using both forward and reverse primer; in addition, all DNMT3A sequence variations were aligned to different single nucleotide polymorphism databases (http://www.ncbi.nlm.nih.gov/sites/snp; http://genome.ucsc.edu/cgi-bin/hgGateway; http://www.ensembl.org/Homo_sapiens/Info/Index) to detect known polymorphisms.

Statistical analyses for the clinical outcomes were performed according to previous reports.²⁷  The median follow-up for survival was calculated according to the method of Korn.²⁸  The definition of complete remission (CR), event-free survival (EFS), relapse-free survival (RFS), and overall survival (OS) as well as cytogenetic categorization into favorable-, intermediate-, and adverse-risk groups followed recommended criteria.^21,27  In addition, we analyzed RFS separately by censoring those patients who received an allogeneic hematopoietic stem cell transplantation (HSCT) in first CR (RFS_allo). Pairwise comparisons between patient characteristics (covariates) were performed by using the Mann-Whitney test for continuous variables and by using Fisher’s exact test for categorical variables. The Kaplan-Meier method was used to estimate the distribution of RFS and OS.²⁹  Estimation of confidence intervals (CIs) for the survival curves was based on Greenwood’s formula for the standard error estimation. A logistic regression model was used to analyze associations between baseline characteristics and the achievement of CR.³⁰  A Cox model was used to identify prognostic variables.³¹  In addition to the molecular marker DNMT3A, age, gender, hemoglobin level, logarithm of white blood cell (WBC) and platelet counts, type of AML (de novo, secondary AML, therapy-related AML), percentage of BM blasts, cytogenetic risk group according to ELN recommendations,²¹  and mutational status of NPM1, FLT3-ITD, FLT3-TKD, IDH (IDH1^R132, IDH2^R140, IDH2^R172), and CEBPA^dm were added as explanatory variables in the regression analyses, as indicated, without further selection. We estimated missing data for covariates by using 50 multiple imputations in chained equations that incorporated predictive mean matching.³²  ELN risk categories are defined as follows: favorable (RUNX1-RUNX1T1, CBFB-MYH11, NPM1^mut/FLT3-ITD^neg [cytogenetically normal AML; CN-AML], CEBPA^mut [CN-AML]), intermediate (intermediate I: NPM1^mut/FLT3-ITD^pos [CN-AML], wild-type NPM1 [NPM1^wt]/FLT3-ITD^neg [CN-AML], NPM1^wt/FLT3-ITD^neg [CN-AML]; intermediate II: MLLT3-MLL, RPN1-EVI1, DEK-NUP214, MLL rearranged, −5 or del(5q), −7, abnl(17p), or complex karyotype.²¹  All statistical analyses were performed with the statistical software environment R version 2.14.0, using the R packages rms version 3.3-1, survival version 2.36-8, and cmprsk version 2.2-2.³³ 

Overall, we found 384 DNMT3A mutations in 370 (20.9%) of 1770 patients (Figure 1 and Table 2; data supplement). There were 239 mutations (64.5%) clustered in the MTase domain at amino acid R882 in exon 23 (R882H, n = 165; R882C, n = 57; R882S, n = 9; R882P, n = 6; R882L, n = 1; R882G, n = 1). The other mutations were distributed as follows: exon 4 (n = 1; tested in 95 patients only), exon 11 (n = 4), exon 12 (n = 2), exon 13 (n = 14), exon 14 (n = 20), exon 15 (n = 11), exon 16 (n = 4), exon 17 (n = 10), exon 18 (n = 21), exon 19 (n = 29), exon 20 (n = 3), exon 21 (n = 2), exon 22 (n = 9), and exon 23 outside codon R882 (n = 16). Only two mutations were homozygous. Fourteen patients had two mutations. There were 348 single nucleotide substitutions leading to missense (n = 341 [88.8%]) and nonsense mutations (n = 7 [1.8%]) as well as 36 frameshift alterations due to insertions or deletions (9.4%).

DNMT3A^mut patients were significantly older (P < .0001) with an increasing mutation frequency until the age of 40 years, which reached a plateau in the range 40 to 60 years (Figure 1). DNMT3A^mut patients were more frequently female (P = .0006), had higher WBC (P < .0001) and platelet (P < .0001) counts as well as higher BM blast percentages (P < .0001) and lactate dehydrogenase serum levels (P = .0004). Treatment-related AML was less common in DNMT3A^mut patients (P = .032). No significant differences were found with regard to clinical characteristics between R882 and non-R882 mutations (Table 1).

DNMT3A^mut were associated with intermediate-risk cytogenetics, in particular CN-AML (P < .0001); mutations were very rare in core-binding factor AML (n = 4 [1.1%]) and occurred in 33 patients with adverse karyotype (9.5%). Interestingly, the association with CN-AML was significantly stronger for R882 mutations (Table 1). Of note, among AML with recurrent balanced rearrangements [eg, inv(16)/t(16;16), t(8;21), t(9;11), t(v;11q23), t(6;9), inv(3)/t(3;3)], no R882 mutation was found, but cases of non-R882 mutations [with the exception of t(9;11)].

Correlations with molecular markers revealed a significant positive association of DNMT3A^mut with NPM1^mut (P < .0001), FLT3-ITD (P < .0001), and IDH1/2^mut (IDH1^R132; P = .0002; IDH2^R140, P = .0004; IDH2^R172, P < 0.0001); DNMT3A^mut co-occurred less frequently with CEBPA^dm (P = .007; Table 1 and Figure 2; supplemental Table 3A-B). R882 mutations had a significantly higher incidence of concurrent NPM1^mut and FLT3-ITD, mainly due to higher incidences of NPM1^mut and FLT3-ITD in R882-mutated patients with abnormal karyotypes, whereas in CN-AML, no difference in these incidences was identified.

For correlation with clinical outcome, 1700 patients with non-acute promyelocytic leukemia AML (missing follow-up data, n = 14) were considered. In the entire cohort, DNMT3A^mut were associated with a higher CR rate (78.7% and 71.4% for DNMT3A^mut and DNMT3A^wt, respectively; P = .005). However, DNMT3A^mut had no impact on response in CN-AML and in the ELN molecularly defined CN-AML subgroups (Table 2).

In multivariable analysis of the entire cohort, DNMT3A^mut was a favorable factor for achievement of CR (odds ratio, 1.44; 95% CI, 1.04-2.00), mainly driven by R882 mutation. However, in CN-AML, DNMT3A^mut had no significant effect on response (odds ratio, 1.35; 95% CI, 0.84-2.17; supplemental Tables 4 and 5).

The median follow-up time for survival was 4.94 years (95% CI, 4.72-5.14 years); the estimated 5-year RFS and OS for the entire cohort were 41% (95% CI, 38%-44%) and 44% (95% CI, 42%-47%), respectively.

In univariable analysis, DNMT3A^mut had no impact on EFS (P = .57), RFS (P = .12), or OS (P = .91) in the entire cohort (n = 1700); similarly, there was no effect in CN-AML (n = 790; EFS, P = .88; RFS, P = .097; OS, P = .20) (Figure 3A-D). To explore a potential difference by localization of mutation, we performed a three-way comparison of DNMT3A^wt vs R882 DNMT3A^mut vs non-R882 DNMT3A^mut. There was no significant difference in outcome in the entire cohort (EFS, P = .84; RFS, P = .13; OS, P = .66) but a trend toward an inferior prognosis was seen in CN-AML patients exhibiting an R882 DNMT3A^mut for EFS (P = .095), RFS (P = .058), and OS (P = .070) (Figure 2A-D; data supplement).

Univariable exploratory subset analyses revealed a significant prognostic impact of DNMT3A^mut in ELN-unfavorable CN-AMLs (n = 435; EFS, P = .15; RFS, P = .002; OS, P = .011), whereas no impact was found in ELN-favorable CN-AMLs (n = 274; EFS, P = .91; RFS, P = .98; OS, P = .99) (Figure 3E-F). This negative impact was mainly driven by R882 DNMT3A^mut (ELN-unfavorable: EFS, P = .28; RFS, P = .004; OS, P = .013; ELN-favorable: EFS, P = .61; RFS, P = .79; OS, P = .82; Figure 2E-F; data supplement).

In multivariable analysis, DNMT3A^mut had no effect on RFS, RFS censored for allogeneic HSCT in first CR (RFS_allo), or OS both in the entire cohort (RFS: hazard ratio [HR], 1.15 [P = .18]; RFS_allo: HR, 1.22 [P = .11]; OS: HR, 0.93 [P = .43]) and in the subgroup of CN-AML (RFS: HR, 1.15 [P = .30]; RFS_allo: HR, 1.15 [P = .36]; OS: HR, 0.99 [P = .90]) (Table 3; supplemental Table 6). When including DNMT3A^mut mutation type (R882 vs non-R882), there was strong evidence that these two mutation types were associated with opposite effects: unfavorable for R882 DNMT3A^mut on RFS (entire cohort: HR, 1.29 [P = .026]; CN-AML: HR, 1.38 [P = .018]) and RFS_allo (entire cohort: HR, 1.39 [P = .017]; CN-AML: HR, 1.36 [P = .052]) and favorable for non-R882 DNMT3A^mut on OS (entire cohort: HR, 0.77 [P = .057]; CN-AML: HR, 0.73 [P = .083]; supplemental Tables 7 and 8).

To the best of our knowledge, this study of 1770 patients provides data on the clinical relevance of DNMT3A^mut in the largest cohort to date of younger adult AML patients who were homogeneously and intensively treated on prospective clinical trials. In accordance with previous studies, we found DNMT3A^mut in 20.9% of all AMLs and in 33.8% of CN-AMLs; 64.5% of mutations clustered at R882 in exon 23.

In our study, patients with DNMT3A^mut were older; more frequently female; and had higher WBC, platelet, and BM blast counts. Similar results were found by others with regard to age,^4,8,12,13  gender,⁹  WBC counts,^{4,5,7-9,12-14}  platelet counts,^8,12,13  and BM blast counts.^4,12,14  It is known, that NPM1^mut and FLT3-ITD are associated in general with higher WBC counts and NPM1^mut is associated with higher platelet counts. With regard to the DNMT3A^mut types in our cohort (R882-DNMT3A^mut vs non-R882-DNMT3A^mut), 72.1% and 38.2% of R882-DNMT3A^mut had a concurrent NPM1^mut or FLT3-ITD, whereas only 49.2% and 26.7% of non-R882-DNMT3A^mut were NPM1^mut or FLT3-ITD^pos. Thus, the effect of higher WBC counts in patients with DNMT3A^mut may be mainly due to cooperating mutations such as FLT3-ITD or NPM1^mut, whereas DNMT3A^mut seems to have an independent effect on platelet counts. With respect to age, we found a rising frequency of the mutation in patients up to the age of 40 years (Figure 1) reaching a plateau thereafter.

As in previous studies, DNMT3A^mut in our study were highly associated with intermediate-risk cytogenetics, in particular CN-AML^{7,8,10,12,13,15} ; DNMT3A^mut were infrequent in AML with adverse cytogenetics and rare in cytogenetically low-risk patients (Table 1).^7,8,13  A novel finding of our study was that R882 and non-R882 DNMT3A^mut were associated with different cytogenetic patterns (Table 1): (1) the association of R882 DNMT3A^mut with CN-AML was significantly stronger (83.9% vs 65.9% for non-R882); (2) no R882 mutations were found among AMLs with recurrent balanced rearrangements, but cases of non-R882 mutations [with the exception of t(9;11)]; and (3) non-R882 DNMT3A^mut were in trend more frequent in AML with adverse cytogenetics.

Molecularly, we found a highly significant association of DNMT3A^mut with NPM1^mut, FLT3-ITD, and IDH1/2^mut, whereas DNMT3A^mut were significantly less frequent in AML with CEBPA^dm (Table 1). These data are consistent with and extend the data from previous studies.^7-15  We identified different patterns of cooperating gene mutations between R882 and non-R882 DNMT3A^mut: R882 DNMT3A^mut had a significantly higher incidence of concurrent NPM1^mut and FLT3-ITD compared with non-R882 DNMT3A^mut (NPM1^mut, 72.1% vs 49.2% [P < .0001]; FLT3-ITD, 38.2% vs 26.7% [P = .04], respectively) These effects were restricted to patients with cytogenetic abnormalities, whereas in CN-AMLs, no significant differences were found. It is conceivable that these different patterns of cooperating events had an impact on clinical outcome analyses. Marcucci et al¹⁴  analyzed the genetic distribution of molecular markers with regard to age (<60 vs ≥60 years) and mutation type (R882 vs non-R882) and found similar results for the different groups.

In our study, DNMT3A^mut was a moderate but significant predictor for achievement of CR in the entire cohort, but not in CN-AML or the ELN-defined molecular CN-AMLs (Table 2). Similarly, a favorable prognostic effect of DNMT3A^mut on CR was described in trend (P = .058) by Ribeiro et al.¹³  In contrast, Thol et al⁸  found a lower CR rate for DNMT3A^mut in CN-AML, and in particular in a high-risk group (FLT3-ITD or NPM1^wt/FLT3-ITD^neg).⁸  Other studies did not reveal a significant influence of DNMT3A^mut on CR.^9-11,14,15  The higher CR rate in our study in the entire cohort is a result of different incidences of DNMT3A^mut with respect to the genetic subgroups. Because DNMT3A^mut are rare in the cytogenetic high-risk group, these patients accumulate in the DNMT3A^wt group with the direct consequence that the CR rate is substantially lower. However, after focusing on a distinct genetic group such as CN-AML, this effect is lost. Thus DNMT3A^mut are not directly linked in our study to higher CR rates.

In contrast with recent studies, our data showed virtually superimposing survival curves for DNMT3A^mut and DNMT3A^wt patients (Figure 3A-D). There was no effect on EFS, RFS, or OS in the entire cohort or in CN-AMLs in univariable and multivariable analyses. In addition, we also performed univariable and multivariable analyses for RFS with censoring of allogeneic HSCT in the first CR on the date of transplantation. Again, no significant difference was evident for the entire cohort (P = .079) or in CN-AMLs (P = .63) (Figure 3A-B; data supplement). The only prognostic effect was found in univariable subset analysis, with DNMT3A^mut being associated with an inferior prognosis in the ELN-unfavorable CN-AML subgroup (Figure 3E-F).

A direct comparison of data across studies is difficult, if not impossible, in many instances, because they differ with respect to which patient cohorts are included (all AMLs vs intermediate-risk AMLs vs CN-AMLs), subset analyses performed (intermediate-risk AMLs vs CN-AMLs; NPM1/FLT3 low- and high-risk groups vs ELN-defined molecular subgroups), age group (mostly age <60 vs >60 years), and type and intensity of therapy administered to the patients. Not surprisingly, published data provide conflicting results, with the majority of studies showing a negative prognostic impact in various but different subsets; conversely, some studies failed to show an impact. The following discussion focuses on those studies with intensively treated younger adult patients.

In one of the pivotal studies by Ley et al,⁷  the negative prognostic effect was highly significant in the subgroup of younger patients (n = 209 [age <60 years]), with an estimated long-term survival of about 20% for patients with DNMT3A^mut compared with almost 60% for patients with DNMT3A^wt (P < .001 by log-rank test).⁷  In the study by Thol et al⁸  (n = 489 [age <60 years]), DNMT3A^mut predicted shorter OS (not RFS) in the entire cohort and in CN-AMLs; among CN-AMLs, DNMT3A^mut had an unfavorable effect on RFS and OS in NPM1/FLT3-ITD high-risk (NPM1^wt/FLT3-ITD^neg or FLT3-ITD^pos) but not low-risk (NPM1^mut/FLT3-ITD^neg) patients.⁸  Ribeiro et al¹³  (n = 415 [age 15-60 years]) showed that in both univariable and multivariable analysis, patients with DNMT3A^mut had significantly inferior RFS and OS; the effect was particularly evident in a subset of cytogenetic intermediate-risk AMLs without FLT3-ITD and NPM1^mut; there was no effect in NPM1^mut/FLT3-ITD^neg AMLs. Renneville et al¹¹  (only CN-AMLs; n = 123 [age <60 years]) found (in multivariable analysis) a significant negative prognostic effect for DNMT3A^mut in AMLs with NPM1^mut, NPM1/FLT3/CEBPA low-risk, and in trend also for NPM1/FLT3/CEBPA high-risk patients. In contrast, Marcucci et al¹⁴  analyzed 415 patients with CN-AML (n = 181 [age <60 years]); when adjusted for other clinical and molecular markers, there was no significant effect of DNMT3A^mut on OS. In the subset of younger patients, only patients exhibiting non-R882 DNMT3A^mut had inferior outcome (see also in the paragraphs below).¹⁴  In the study by Patel et al (n = 398 [age <60 years]), DNMT3A^mut had no effect on survival in univariable and multivariable analysis.¹⁶ 

A direct comparison of treatment effects on DNMT3A^mut between the different studies is not possible because of the large spectrum of different treatment approaches ranging from intensive therapy including allogeneic HSCT to palliative schedules. To the best of our knowledge, we analyzed the largest cohort to date of homogenously, intensively treated younger AML patients, and our analyses revealed no impact on survival in the entire cohort or in CN-AML. In addition, we did not observe an effect of allogeneic HSCT on RFS in DNMT3A^mut patients. When including DNMT3A^mut mutation type (R882 vs non-R882), we saw an unfavorable effect for R882 DNMT3A^mut on RFS and RFS_allo in the entire cohort and in CN-AML.

Patel et al¹⁶  described in their study, that high-dose daunorubicin compared with standard-dose daunorubicin improved survival of patients with DNMT3A^mut, suggesting that dose-intensification may mitigate a potential negative prognostic effect in less intensively treated patients.

Another confounding variable that likely needs to be considered is the mutation localization. About two thirds of DNMT3A^mut are located in codon R882. It could be shown that R882H directly inhibits the enzymatic activity of DNMT3A⁶  and disturbs tetramerization, leading to disruption of methylation of multiple CpG sites.³⁴  At present, it is unknown whether different types of DNMT3A^mut have different biological effects. In the studies by Ley et al, Thol et al, Renneville et al, and Ribeiro et al,^7,8,11,13  no difference in outcome was found between R882 and non-R882 DNMT3A^mut. From our correlative clinical study, there is clear evidence that R882 and non-R882 DNMT3A^mut in younger adult patients differ with respect to their association to genetic subgroups (see above). Of note, when accounting for DNMT3A R882 vs non-R882 mutation type, there was strong evidence that the two mutational groups were associated with opposite effects: unfavorable for R882 DNMT3A^mut on RFS and favorable for non-R882 DNMT3A^mut on OS, each in both the entire cohort and in CN-AMLs (supplemental Table 7A-D). Our data are in contrast to Marcucci et al¹⁴  who reported on an age-related impact of R882 vs non-R882 DNMT3A^mut in 181 younger CN-AML patients; only non-R882 DNMT3A^mut were associated with worse clinical outcome, whereas R882 DNMT3A^mut had no prognostic impact; conversely, in older patients, only R882 DNMT3A^mut were independently associated with worse outcome.

Differences in results of the reported data may also be due to differences in biometrical analyses, such as selection bias and variances in model building. Notably, in the studies by Marcucci et al,¹⁴  Ribeiro et al,¹³  Ley et al,⁷  and Thol et al,⁸  a potential selection bias may be present because of the low percentage of patients selected for the analysis in relation to the whole study populations with 18%,⁷  13%,⁸  6% (only available for Cancer and Leukemia Group B samples),¹⁴  and 59%,¹³  respectively. In our study, 1770 (90%) of 1983 patients were analyzed by minimizing the risk of selection bias. Because patients with high initial WBC counts are more likely to have biobanked samples available for post hoc correlative analyses, it is not surprising that initial WBC counts in these studies were higher compared with those reported in our study. Consequently, inclusion of WBC (dichotomized or log-normalized) into multivariable models is statistically mandatory, as done in our study but not done consistently by the other investigators.^7,8,13,14 

For our prognostic marker study, we used multiple imputations in case of missing data. The amount of missing data is stated in Table 1, and the variable with the largest amount of missing data is CEBPA (269 [15%] of 1770). Comparing the multivariable models by using multiple imputations and full data set analyses for the OS and RFS models (for both the entire cohort and for CN-AML) showed that the estimate and the P value for DNMT3A are in the same range in both models (data not shown).

In addition, we addressed the question of whether, in terms of statistical power, we would have been able to detect an effect of DNMT3A^mut on outcome based on the range of previously reported HRs between 1.4¹⁴  and 1.9.⁷  Our post hoc power analysis showed that the sample size of 1700 patients with a proportion of 20% of DNMT3A^mut patients resulted in a power of nearly 100% to detect a prognostic impact of DNMT3A^mut for an HR of 1.4.³⁵  Therefore, we think that our analysis may give a less biased view of the prognostic impact of DNMT3A^mut in patients between the ages of 16 and 80 years compared with the other published studies.

In conclusion, in our study of 1700 younger adult AML patients, DNMT3A^mut had no impact on survival in the entire cohort or in CN-AMLs. A moderate negative prognostic effect, driven by R882 DNMT3A^mut, was found only in the ELN-unfavorable CN-AML subgroup. On the basis of the data from this study and data from other published reports, we consider it rather unlikely that DNMT3A^mut will be of major clinical relevance as prognostic markers in younger adult patients treated with intensive therapy. Since DNMT3A^mut appear to affect DNA methylation and epigenetic regulation of gene expression, it will be of clinical interest to evaluate DNMT3A^mut in the context of demethylation.

The authors are grateful to all members of the German-Austrian AMLSG for their participation in this study and for providing patient samples. A list of participating institutions and investigators appears in the data supplement.

Supported in part by grants 01GI9981 and 01KG0605 from the German Bundesministerium für Bildung und Forschung and by grant SFB 1074/B3 from the Deutsche Forschungsgemeinschaft. V.G. is a grant recipient of the Medical Faculty of Ulm University; L.B. is a Heisenberg Scholar of the Deutsche Forschungsgemeinschaft (BU 1339/3-1). AMLSG treatment trials were in part supported by Pfizer and Amgen.

Contribution: V.I.G., H.D., R.F.S., and K.D. conceived and designed the study; H.D., R.F.S., and K.D. provided financial support; K.D. provided administrative support; R.F.S., A.K., M.v.L.-T., W.B., H.G.D., S. Kremers, R.G., A.R., M.R., H.R.S., M.W., H.G.K., V.R., G.H., A.L.P., M.G., A.G., J.K., H.D., and K.D. provided study materials or patients; and all authors collected and interpreted data and provided data analysis.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Konstanze Döhner, Department of Internal Medicine III, University Hospital of Ulm, Albert-Einstein-Allee 23, 89081 Ulm, Germany; e-mail: konstanze.doehner@uniklinik-ulm.de.