DNMT3A mutations in acute myeloid leukemia: stability during disease evolution and clinical implications

DNMT3A encodes the enzyme DNA methyltransferase (DNMT) 3A, which catalyzes the addition of methyl groups to the cytosine residue of CpG dinucleotides in DNA.^1,2  DNMT3A contains 3 main structure domains: an ATRX, DNMT3, and DNMT3L–type zinc finger domain, a proline-tryptophan-tryptophan-proline domain, and the methyltransferase (MTase) domain.¹  The proline-tryptophan-tryptophan-proline domain is responsible for targeting the enzyme to nucleic acid, whereas the zinc finger domain mediates protein-protein interactions with the transcription factors Myc and RP58, the heterochromatin protein HP1, histone deacetylases, and the histone methyltransferase Suv39h1.²  Recently, mutations in DNMT3A were identified in patients with AML, myelodysplastic syndromes, and myeloproliferative neoplasms.^3–7  The incidences of this mutation in AML varied: 4.1% in a Japanese study,⁸  9% (among all AML, including M4/M5 and other subtypes) in a Chinese study,⁹  and approximately 20% in 2 Western studies.^3,4  Whether there is a geographic difference in the incidence of DNMT3A mutations needs to be determined. Furthermore, sequential analyses to evaluate the stability of DNMT3A mutations during the clinical course were limited to a small number of patients. In the present study, we investigated the DNMT3A mutation in 506 patients with de novo AML and analyzed its interactions with 16 other gene alterations. Sequential analysis of the DNMT3A mutation during the clinical course was also performed on 138 patients to investigate the stability and pathogenic role of this mutation in AML. Further, to better stratify AML patients into different risk groups, a scoring system integrating DNMT3A mutations with 8 other prognostic factors, including age, WBC count, cytogenetics, NPM1/FLT3 internal tandem duplication (NPM1/FLT3-ITD), CEBPA, AML1/RUNX1, WT1, and IDH2 mutations, into survival analysis was proposed.

This study was approved by the institutional review board of the National Taiwan University Hospital (NTUH), and written informed consent was obtained from all participants in accordance with the Declaration of Helsinki. From March 1995 to December 2008, a total of 506 adult patients who were newly diagnosed as having de novo AML at NTUH and had enough cryopreserved cells for analysis were enrolled consecutively. Patients with antecedent hematologic diseases or therapy-related AML were excluded. Diagnosis and classification of AML were made according to the French-American-British (FAB) Cooperative Group Criteria.

In total, 363 (71.7%) patients received standard induction chemotherapy (idarubicin 12 mg/m²/d on days 1-3 and cytarabine 100 mg/m²/d on days 1-7) and then consolidation chemotherapy with 2-4 courses of high-dose cytarabine (2000 mg/m² every 12 hours on days 1-4 for a total of 8 doses), with or without an anthracycline (idarubicin or Novantrone), after achieving complete remission (CR).^10,11  The patients with acute promyelocytic leukemia (M3 subtype) received concurrent all-trans retinoic acid and chemotherapy. The remaining 143 patients received palliative therapy with supportive care and/or low-dose chemotherapy because of underlying comorbidities or based on patient decision. Forty-five patients received allogeneic hematopoietic stem cell transplantation (HSCT) in first CR.

BM cells were harvested directly or after 1-3 days of unstimulated culture, as described previously.¹²  Metaphase chromosomes were banded with the trypsin-Giemsa technique and karyotyped according to the International System for Human Cytogenetic Nomenclature.

A panel of mAbs to myeloid-associated antigens, including CD13, CD33, CD11b, CD15, CD14, and CD41a, as well as lymphoid-associated antigens, including CD2, CD5, CD7, CD19, CD10, and CD20, and lineage-nonspecific antigens HLA-DR, CD34, and CD56, were used to characterize the phenotypes of the leukemia cells, as described previously.¹³ 

Mutation analysis of DNMT3A exons 2-23 was performed by PCR and direct sequencing as described previously.⁴  Abnormal sequencing results were confirmed by at least 2 repeated analyses. Sequential analysis of the DNMT3A mutation during the clinical course was performed in 316 samples from 138 patients. Mutation analyses of 16 other relevant molecular marker genes, including class I mutations, such as FLT3-ITD and FLT3-TKD,¹³ N-RAS,¹⁴ K-RAS,¹⁴ JAK2,¹⁴ KIT,¹⁵  and PTPN11¹⁶  mutations, and class II mutations, such as MLL-PTD,¹⁷ CEBPA,¹⁸  and AML1/RUNX1¹¹  mutations, as well as NPM1,¹⁹ WT1,¹⁰ ASXL1,²⁰ IDH1,²¹ IDH2²²  (including R140 and R172 mutations), and TET2²³  mutations, were performed as described previously. To detect DNMT3A mutation at diagnosis, we used DNA amplified in vitro from BM cells with the Illustra GenomiPhi V2 DNA-amplification kit as described by the manufacturer (GE Healthcare). All mutations detected were verified in the original nonamplified samples. All nucleotide alterations causing premature truncation of the DNMT3A proteins (nonsense or frame-shift mutations) were regarded as true mutations. Missense mutations were regarded as true only if they were documented in other studies or could be verified by sequencing of normal somatic tissues or matched remission BM samples.

For patients with double mutations, Taq polymerase-amplified (TA) cloning was performed to determine whether the 2 mutations were in the same or different alleles, as described previously.¹⁵  Briefly, the cDNA was amplified to cover both mutations and the PCR products were then cloned into the TA-cloning vector pGEM-T Easy (Promega) and more than 10 clones were selected for sequencing.

The discrete variables of patients with and without gene mutation were compared using the χ² tests, but if the expected values of contingency tables were smaller than 5, the Fisher exact test was used. If the continuous data were not normally distributed, Mann-Whitney U tests were used to compare continuous variables and medians of distributions. To evaluate the impact of the DNMT3 mutation on clinical outcome, only the patients who received conventional standard chemotherapy were included in the analysis.^10,11  Overall survival (OS) was measured from the date of first diagnosis to the date of last follow-up or death from any cause, whereas relapse was defined as a reappearance of at least 5% leukemic blasts in a BM aspirate or new extramedullary leukemia in patients with a previously documented CR.²⁴  Relapse-free survival (RFS) was measured from the date of attaining the leukemia-free state until the date of AML relapse or death from any cause, whichever occurred first. Cox regression survival estimation was used to plot survival curves and to test the differences between groups. Multivariate Cox proportional hazard regression analysis was used to investigate independent prognostic factors for OS and RFS. The proportional hazards assumption (constant hazards assumption) was examined using time-dependent covariate Cox regression before conducting multivariate Cox proportional hazard regression. The variables including age, WBC counts, karyotype, NPM1/FLT3-ITD, WT1, CEBPA, AML1/RUNX1, TET2, ASXL1, IDH2, and DNMT3A mutations were used as covariates. Those patients who received HSCT were censored at the time of transplantation in survival analysis to ameliorate the influence of the treatment.^10,11 P < .05 was considered statistically significant. All statistical analyses were performed with SPSS Version 18 software and Statsdirect (2.7.8b, 2011).

Excluding the 8 single-nucleotide polymorphisms (P9P, S267S, G291G, A398A, P385P, L422L, V435V, and V563V) that were detected in 316 patients but did not alter the amino acid residues, and the 7 missense mutations (C586W, P896L, G543C, Y735C, A644T, G699D, and G707D) that were found in 6 patients but had uncertain biologic significance (because they were not reported previously and could not be verified because of lack of matched BM samples at CR), DNMT3A mutations at 30 different positions were identified in 70 patients (Table 1 and Figure 1). Twelve were missense mutations, 8 were nonsense mutations, 9 were frame-shift mutations, and 1 was an in-frame mutation. The most common mutation was R882H (n = 26), followed by R882C (n = 15), R882S (n = 3), R736H (n = 3), and R320X (n = 2). All other mutations were detected in only 1 patient each. Mutations at exon 23 occurred in 47 patients, including the 44 patients with R882 mutations. Four patients had double heterozygous mutations (patients 43, 64, 65, and 68); in 1 of them (patient 64), the 2 mutations were confirmed to be biallelic by DNA PCR and TA cloning, and for the other 3, the nature of the double mutations was not verified by this method because the 2 mutations were located in different exons too far apart to be amplified by a single-DNA PCR reaction. The remaining 66 patients showed only one mutation; all were heterozygous.

In total, 500 de novo AML patients, including 70 (14%) DNMT3A-mutated and 430 DNMT3A-wild patients were enrolled into the study. The 6 patients with missense mutations of unknown significance were censored and were not included in the following analyses. A comparison of clinical characteristics of patients with and without distinct DNMT3A mutations is given in Table 2. DNMT3A-mutated patients were older (median, 61 vs 49 years, P < .0001) and had higher WBC, blast, and platelet counts than DNMT3A-wild patients (P = .0018, .0012, and .0001, respectively). Patients with the FAB M5 subtype of AML had the highest incidence (50%, P < .0001) of DNMT3A mutation, followed by those with the FAB M4 subtype (22.6%, P = .0026). DNMT3A mutations were positively associated with the expression of CD13 (P = .022) and CD14 (P = .0015), but inversely associated with the expression of CD34 (P = .0039) on leukemic cells (supplemental Table 1, available on the Blood Web site; see the Supplemental Materials link at the top of the online article). There was no difference in the expression of other antigens between the patients with and without the DNMT3A mutation.

Chromosome data were available for 482 patients at diagnosis, including 66 DNMT3A-mutated and 416 DNMT3A-wild patients (Table 3). DNMT3A mutations occurred more frequently in patients with intermediate-risk cytogenetics (19.5%) than in those with favorable karyotype or unfavorable cytogenetics (2.4%, P = .0069). There was a significant difference in the incidence of the DNMT3A mutation among patients with normal karyotype (22.9%), simple abnormalities with 1 or 2 changes (6.2%), and complex cytogenetics with 3 or more abnormalities (3.9%, P < .0001). None of the patients with t(8;21), t(15;17) inv(16), or 11q23 translocation had a DNMT3A mutation. There was no association of the DNMT3A mutation with other chromosomal abnormalities, including +8, +11, +13, +21, −5/del(5q), and −7/del(7q).

To investigate the interaction of gene mutations in the pathogenesis of adult AML, a complete mutational screening of 16 other genes was performed in all 500 patients (Table 4). Among the 70 patients with DNMT3A mutations, 68 (97.1%) showed additional molecular abnormalities at diagnosis (supplemental Table 2). Fifteen had 1 additional change, 37 had 2, 13 had 3, and 3 had 4. The most common associated molecular event was the NPM1 mutation (n = 38), followed by FLT3-ITD (n = 30), IDH2 (n = 16), and FLT3-TKD (n = 9) mutations. Patients with DNMT3A mutations had a significantly higher incidence of the NPM1 mutation and FLT3-ITD, IDH2, and PTPN11 mutations than those with DNMT3A-wild type (54.3% vs 15.3%, P < .0001; 42.9% vs 19.3%, P < .0001; 22.9% vs 9.1%, P = .0016; and 10% vs 3.5%; P = .007, respectively). Conversely, CEBPA mutation was rarely seen in patients with DNMT3A mutations (4.3% vs 14.7%, P = .0134). There was no difference in the incidence of other molecular mutations between patients with and without the DNMT3A mutation. Among the 68 patients with concurrent other genetic alterations, 51 (75%) had at least 1 concomitant class I mutation; 16 (23.5%) had class II mutations; and 38 (54.3%) had NPM1 mutations, which behave more like class II mutations.¹³  In total, 40 patients (58.8%) had concurrent class I and class II or NPM1 mutations at diagnosis. Twenty-one patients had concomitant FLT3-ITD and NPM1 mutations (supplemental Table 2).

Of the 363 AML patients undergoing conventional intensive induction chemotherapy, 284 (78.5%) patients achieved a CR. The probability of achieving a CR was similar between patients with and without DNMT3A mutations (74.4% vs 79%, P = .5531). However, the patients with DNMT3A mutations had a trend of higher relapse rate than those without (65.6% vs 48.8%, P = .0911). With a median follow-up of 55 months (range, 1.0-160), patients with the DNMT3A mutation had significantly poorer OS and RFS than those without the DNMT3A mutation (median, 14.5 months vs 38 months, P = .013, and median, 7.5 months vs 15 months, P = .012, respectively, Figure 2A-C). The same was true among patients with non-M3 AML (P = .04 and P = .036, respectively). In the subgroup of 130 younger patients (< 60 years) with normal karyotype AML (CN-AML), the differences between patients with and without the DNMT3A mutation in OS (median, 15.5 months vs not reached, P = .018, Figure 2B) and RFS (median, 6 months vs 21 months, P = .004, Figure 2D) were still significant. We also observed that the prognostic impact of the DNMT3A mutation could only be demonstrated in the patients with a poor prognostic genotype (NPM1-mutated (NPM1⁺)/FLT3-ITD⁺, NPM1-wild (NPM1⁻)/FLT3-ITD⁺ or FLT3/ITD⁻), but not in those with favorable genotype (NPM1⁺/FLT3-ITD⁻) among total AML patients (P < .001 and P = .823, respectively) or in CN-AML patients (P < .001 and P = .970, respectively). There was no significant difference in survival between patients with mutations of R882 and those with other mutations (P = .612).

In multivariate analysis (Table 5), the independent poor risk factors for OS were older age (> 50 years), high WBC count (> 50 000/μL), unfavorable karyotype, DNMT3A mutation, AML1/RUNX1 mutation, and WT1 mutation. Conversely, CEBPA^{double-mutation} and NPM1⁺/FLT3⁻ITD⁻ were independent favorable prognostic factors. There was a trend of better OS in patients with the IDH2 mutation (hazard ratio [HR], 0.573; 95% confidence interval [95% CI], 0.296-1.110, P = .099). The independent poor risk factors for RFS included high WBC count (> 50 000/μL), unfavorable karyotype, DNMT3A mutation, and WT1 mutation. NPM1⁺/FLT3-ITD⁻ was an independent favorable factor for RFS. In 130 CN-AML patients younger than 60 years, the DNMT3A mutation was still an independent poor prognosis for OS and RFS (HR, 2.303; 95% CI, 1.088-4.876, P = .029 and HR, 3.496; 95% CI, 1.773-6.896, P < .001, respectively, supplemental Table 3).

To better stratify the AML patients into different risk groups, a scoring system incorporating 9 prognostic markers—age, WBC count, cytogenetics at diagnosis, NPM1/FLT3-ITD, and mutations of CEBPA, DNMT3A, AML1/RUNX1, IDH2, and WT1 mutation—into the survival analysis was formulated based on the results of our Cox proportional hazards model. A score of −1 was assigned for each parameter associated with a favorable outcome (ie, CEBPA^{double-mutation}, IDH2 mutation, and NPM1⁺/FLT3-ITD⁻), whereas a score of +1 for each factor associated with an adverse outcome (ie, DNMT3A, WT1, and AML1/RUNX1 mutations, older age, and higher WBC counts at diagnosis). The karyotypes were stratified into 3 groups (+2, unfavorable; +1, intermediate; and 0, favorable). The algebraic summation of these scores for each patient was the final score. This score system divided the AML patients into 5 groups with different clinical outcomes (P < .001 for both OS and RFS, Figure 3).

DNMT3A mutations were studied sequentially in 316 samples from 138 patients, including 35 patients with distinct DNMT3A mutations and 103 patients without mutations at diagnosis (Table 6). Among the 34 patients with DNMT3A mutations who had ever obtained a CR and had available samples for study, 29 lost the original mutation at remission status, but 5 (patients 5, 8, 28, 32, and 33) retained it (Table 6); all 5 patients relapsed within a median of 3.5 months and died of disease progression, suggesting the presence of leukemic cells. In the 13 patients who had available samples for serial study at relapse, 12 patients regained the original mutations, but 1 (patient 9) lost the mutation at relapse. Because direct sequencing might not be sensitive enough to detect low levels of DNMT3A mutation signal, we sequenced TA clones of the PCR product from patient 9 and 1 mutant clone of 17 was detected. Among the 103 patients who had no DNMT3A mutation at diagnosis, none acquired the DNMT3A mutation at relapse, whereas karyotypic evolution was noted at relapse in 39% of these patients (data not shown).

In the present study, we found that the DNMT3A mutation was associated with distinct clinical and biologic features and was a poor prognostic factor in AML patients independent of age, WBC counts, karyotype, and other genetic markers.

DNMT3A mutations at 30 different positions, most commonly in the MTase domain, were demonstrated (Figure 1). All of the nonsense, frame-shift, and in-frame mutations generated truncated peptide with complete or partial deletion of the MTase domain and were thought to abolish the catalytic activity of this enzyme. The missense R882 mutations, the most common DNMT3A mutations, resulted in impaired enzyme activity,^4,9  but the influence of other missense mutations on the enzyme remains unclear. These missense mutations all involved amino acid residues well conserved through evolution. We censored 6 patients with missense variants of unknown significance and did not include them in the analyses because there were no available remission BM samples or normal tissues to verify that their DNMT3A variants were true somatic mutations. In contrast to the report of Thol et al,³  who only found mutations between exons 15 and 23, 10 mutations in our patients were located outside of this region (Figure 1). Nine of these mutations were frame-shift or nonsense mutations and were expected to impair enzyme activity. Similar to our finding, Ley et al also detected mutations outside of exons 15 to 23.⁴  In the study by Thol et al, all 23 exons of DNMT3 were initially sequenced in 40 patients.³  Because only mutations between exons 15 and 23 were found in these patients, they subsequently sequenced exons 15 to 23, but not other exons, in other patients. Because all but one mutation outside of exons 15 to 23 in our study were detected in only one patient each (an incidence of 1 in 500 for each mutation), the absence of mutation in this area in 40 patients might not mean that it would not happen in other patients.

In this study, DNMT3A mutations were found in 14%, 15.2%, 19.5%, and 22.9%, respectively, in whole cohort, non-M3 AML, intermediate-risk cytogenetics, and CN-AML groups, lower than the reports of Ley et al (22.1% for total patients, 33.7% for those with intermediate-risk cytogenetics, and 36.7% for CN-AML patients)⁴  and Thol et al (17.8% in non-M3 and 27.2% in CN-AML patients).³  In a study of Chinese AML patients by Yan et al, DNMT3A mutations were detected in 20.5% and 13.6%, respectively, of patients with the FAB M5 and M4 subtypes of AML, but none of the patients with FAB M1 or M2 subtypes had the mutation, leading to an overall incidence of 9% for the DNMT3A mutation in the entire group of AML patients.⁹  Yamashita et al also reported a low incidence (4.1%) of DNMT3A mutations in Japanese AML patients.⁸  The reason for the variability in the incidence of DNMT3A mutations in different studies is unknown, but may be because of the differences in ethnic background, patient populations recruited, and methods used. Whether DNMT3A mutations occur less frequently in Asian than in Western AML patients needs to be determined by further studies.

In our comprehensive analysis of the 17 gene mutations in 500 patients, we found that the DNMT3A mutation was the third commonest recurrent genetic alteration, followed by FLT3-ITD and NPM1 mutations, in AML patients. In addition to its close association with NPM1 mutations and FLT3-ITD, which has been shown previously,^3,4  we demonstrated herein that DNMT3A mutations were also positively associated with PTPN11 and IDH2 mutations and negatively associated with the CEBPA mutation. More intriguingly, we found that the DNMT3A mutation rarely occurred alone; all but 2 patients with DNMT3A mutations showed concurrent mutations of other genes, more frequently class I (51 of 68, 75%), but also class II mutations (16 of 68, 23.5%) and NPM1 mutations (38 of 68, 54.3%, supplemental Table 2), which behave more like class II mutations.¹³  In short, the development of AML may require concerted interaction among different genetic alterations.

The stability of DNMT3A mutations in the evolution of AML remains unclear. In a serial study of 5 patients with DNMT3A mutations at diagnosis, Thol et al found that the mutations disappeared at CR and reappeared at relapse in one patient tested.³  To the best of our knowledge, the present study recruited the largest number of AML patients for sequential analysis of DNMT3A mutations during the clinical course. In contrast to the instability of FLT3-ITD during disease evolution, we found that the DNMT3A mutation seemed to be stable, analogous to NPM1 and IDH1/2 mutations.^13,21,22  At relapse, all DNMT3A-mutated patients who had available samples for serial study regained the same mutations, including the one in whom the mutation could be detected by a sensitive gene-cloning technique, but not by direct sequencing. Conversely, all 103 patients without DNMT3A mutation at diagnosis remained DNMT3A-wild at relapse. These results suggested that although DNMT3A mutations are important for the development of AML, they may play little role in disease progression. Given the stability of the DNMT3A mutation during AML evolution, it may be a potential biomarker for monitoring minimal residual disease.

We found that AML patients with DNMT3A mutations had distinct clinical and laboratory characteristics and a poor prognosis. Recently, many gene mutations have been detected in AML and some found to be independent prognostic factors. In the present study, to better stratify AML patients into different risk groups, a survival scoring system incorporating the DNMT3A mutation and 8 other prognostic factors, including age, WBC count, cytogenetics, and NPM1/FLT3-ITD, CEBPA, AML1/RUNX1, WT1, and IDH2 mutations, into the survival analysis was formulated. This scoring system was found to be more powerful than any single marker at separating patients into different prognostic groups. However, further study with an independent cohort will be needed to validate the proposed scoring system.

In summary, this study demonstrated that DNMT3A mutations could be detected in a substantial number of patients with de novo AML and were closely associated with older age and FAB M4/M5 subtypes. DNMT3A mutations occurred more frequently in patients with intermediate-risk cytogenetics and normal karyotype. They were mutually exclusive with CEBPA mutation, but were closely associated with FLT3/ITD, NPM1, PTPN11, and IDH2 mutations. Furthermore, the DNMT3A mutation was an independent poor risk factor for OS and RFS among total cohort and CN-AML patients. Sequential study during the clinical course showed that the DNMT3A mutation was stable during AML evolution. We conclude that the incorporation of the DNMT3A mutation with 8 other prognostic factors into survival analyses can better stratify AML patients into different risk groups.

This work was partially sponsored by grants from the National Science Council (NSC 97-2314-B002-015-MY3, 99-2314-B-002-143, 100-2325-B-002-032, and 100-2628-B-002-003-MY3) and the Department of Health (DOH100-TD-C-111-001), Taiwan, Republic of China, and the Department of Medical Research (NTUH.99P14 and 100P07), National Taiwan University Hospital, Taipei, Taiwan.

Contribution: H.-A.H. collected the literature, managed and interpreted the data, performed the statistical analysis, and wrote the manuscript; Y.-Y.K. collected the literature, managed and interpreted the data, and wrote the manuscript; C.-Y.L. performed and interpreted the statistical analysis; L.-I.L. performed and interpreted the mutation analysis; C.-Y.C., W.-C.C., M.Y., S.-Y.H., J.-L.T., B.-S.K., S.-C.H., S.-J.W., W.T., and Y.-C.C. contributed patient samples and clinical data; M.C.L., M.-H.T., C.-F.H., Y.-C.C., C.-Y. L., F.-Y.L., and M.-C.L. performed the gene mutation and chromosomal studies; and H.-F.T. planned, designed, and coordinated the study and wrote the manuscript.

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

Correspondence: Hwei-Fang Tien, MD, PhD, Department of Internal Medicine, National Taiwan University Hospital, 7 Chung-Shan Street, Taipei, Taiwan; e-mail: hftien@ntu.edu.tw.