Double mutant DNMT3A AML: a unique subtype experiencing increased DNA damage and poor prognosis

At least 1 mutation in a known driver gene can be identified by targeted next-generation sequencing (NGS) in >95% of patients with acute myeloid leukemia (AML), with DNMT3A being one of the most frequently mutated genes.¹ Large-scale genomic studies have identified mutations in DNMT3A as the most prevalent early event (∼20%) in adult AML. The high prevalence of DNMT3A mutations in myeloid neoplasms indicates that these mutations are important in driving leukemogenesis. DNMT3A mutations also represent the most frequent genetic aberration driving clonal hematopoiesis (CH) in healthy individuals.^2-5 In CH, DNMT3A-mutant hematopoietic stem and progenitor cells gain a competitive advantage over their normal counterpart. Their expansion is assumed to predispose them to acquiring additional cooperating leukemogenic mutations. The precise effect on disease pathogenesis of the different types of DNMT3A mutations remains to be uncovered.⁶ 

DNMT3A encodes an enzyme crucial for de novo DNA methylation. The enzyme contains 3 functional domains: a Pro-Trp-Trp-Pro domain, an Atrx-DNMT3-DNMT3L domain, and a catalytic C-terminal methyltransferase (MTase) domain.^7-10 The Pro-Trp-Trp-Pro domain targets heterochromatin by recognizing H3K36me2/3, whereas the Atrx-DNMT3-DNMT3L domain is required for the recognition and binding of histone tails lacking K4 methylation, thereby facilitating enzymatic interactions with various proteins.^2,11 The highly conserved MTase domain catalyzes 5-cytosine methylation and is thereby essential for establishing DNA methylation and the control of gene expression.^12,13 During human de novo DNA methylation, DNMT3A complexes with DNMT3L to form a tetramer composed of 2 DNMT3A proteins flanked by DNMT3L.^14,15 The tetramer binds DNA, and the catalytic interface catalyzes 5-cytosine methylation to establish methylation patterns.

Although different types of mutations, such as missense, frameshift, and splice site, occur within DNMT3A, the missense mutation at amino acid R882 is the most prevalent.^2,6,16 The effect on methylation activity varies among the different DNMT3A mutations and is therefore associated with epigenetic heterogeneity.^17-19 Several studies suggest that missense mutations in DNMT3A have a dominant-negative effect, especially the R882 hot spot mutation. However, no gross methylation differences are observed in DNMT3A mutant AML, with only focal hypomethylation detected.^19-24 Furthermore, in AML, DNMT3A mutations often co-occur with mutations in NPM1, FLT3, or IDH1/2, but little is known about the functional relationships between these comutations and different types of DNMT3A mutations.^1,25,26 In addition, the impact of mutant DNMT3A on treatment outcome in AML remains controversial.^16,27 For instance, the effect of monoallelic vs biallelic mutated DNMT3A on outcome has been sparsely investigated.^28,29 Although limited by sample size, patients with biallelic mutated DNMT3A AML may have an inferior prognosis.^16,28,30 

Spontaneous deamination of 5-methylcytosine (5mC) is a natural threat to our genomic stability,³¹ thereby replacing 5mC with thymine, resulting in a C-to-T transition mutation.³² The specific base substitution and sequence context of this mutational process is represented by a mutational signature, termed SBS1.³³ Increased DNA methylation damage is signified by an enrichment of SBS1 in the genomic mutation imprint.³³ TET enzymes catalyze a stepwise hydroxylation process in which 5mC is converted to 5-hydroxymethylcytosine (5hmC). 5hmC is subsequently hydroxylated to form 5-formylcytosine (5fC), followed by 5-carboxylcytosine (5caC).^34,35 During spontaneous deamination, the DNA glycosylases methyl-CpG binding domain 4 (MBD4) and thymine DNA glycosylase (TDG) replace, via base excision repair (BER), the 5mC derivatives with cytosine. In the hydroxylation pathway, only TDG is involved in this process (Figure 1A).³⁶ Of note, mutations in TET2, another frequent somatic gene mutation in AML, are often loss-of-function (LOF) and associated with DNA hypermethylation.^17,37-39 Furthermore, mutant isocitrate dehydrogenase 1 (IDH1) or IDH2 form the oncometabolite D-2-hydroxyglutarate, which inhibits TET2 activity, also leading to increased DNA methylation.⁴⁰ A rare germ line deficiency of MBD4, previously identified by our laboratory, predisposes to accelerated CH and early-onset AML.⁴¹ These patients share a common path to AML, marked by biallelic DNMT3A mutation followed by the acquisition of a mutation in either IDH1 or IDH2. This conserved path to AML is likely caused by impaired DNA methylation pathways, resulting in selective pressure on methylated CpG loci and subsequent mutations within cooperating cancer genes.

The excessive amount of DNA methylation damage in MBD4-deficient patients with AML indicates that TDG does not compensate for the complete loss of MBD4.⁴¹ DNMT3A and TDG are binding partners, affecting the enzymatic activity of both proteins, indicating that DNA methylation may be partially linked to BER (Figure 1A).⁴² Because driver mutations in leukemia are often caused by C-to-T mutations, we set out to examine whether mutant DNMT3A impairs TDG repair activity, thereby increasing the risk of C-to-T mutations and the acquisition of additional driver mutations.

Bone marrow or peripheral blood samples of 2545 patients with a confirmed diagnosis of high-risk myelodysplastic syndrome or AML were included. All patients were enrolled in the consecutive Dutch-Belgian Hemato-Oncology Cooperative group and Swiss Group for Clinical Cancer Research trials (HO04, HO29, HO42, HO42A, HO43, HO92, HO102, HO103, and HO132).^43-50 In these studies, patients with newly diagnosed AML were treated with intensive therapy according to trial protocol. Patients had given written informed consent in accordance with the Declaration of Helsinki.

Mutations at diagnosis were determined by NGS with the Illumina TruSight Myeloid Sequencing panel, according to the manufacturer’s instructions (Illumina, San Diego, CA). This sequencing panel covers 54 frequently mutated genes in myeloid malignancies.^51,CEBPA mutation detection was performed using a custom NGS amplicon panel previously designed by our laboratory.⁵² Detection of FLT3 internal tandem duplication was done as described previously.^53,54 Definitions of DNMT3A double mutant (DM) are as follows: either a single DNMT3A mutation with a variant allele frequency (VAF) >55% or the presence of 2 DNMT3A mutations with a combined VAF >50%. DNMT3A single mutant (SM) refers to a DNMT3A mutation with a VAF ranging from 10% to 55%; DNMT3A wild type (WT) refers to the absence of a DNMT3A mutation or a DNMT3A mutation with a VAF <10% (considered as CH).

Cytogenetic testing was performed at local reference centers using standard protocols. All data were centrally peer reviewed by clinical genetics laboratory specialists. All clonal and numerical chromosomal abnormalities were reported in accordance with the International System for Human Cytogenetic Nomenclature and the European LeukemiaNet (ELN) 2022 recommendations.

The pET28-MHL hexahistidine-tagged DNMT3A WT and mutants, MBD4 residues 430-580 (pET28, Addgene, Watertown, MA), and TDG residues 111 to 348 (pET28; kindly provided by Hashimoto et al)⁵⁵ were expressed in Escherichia coli BL21(DE3) Gold and BL21(DE3) pLysS cells (Life Technologies, Carlsbad, CA). Single colonies of BL21(DE3) containing pET28 expression vector with the desired insert were cultured for 16 hours at 37°C in Luria Broth medium, supplemented with 50 μg/mL kanamycin. The cultures were subsequently diluted 25 times in Luria Broth medium without a selection marker and grown at 37°C until the optical density 600 (OD600) reached 0.5 to 0.8. The cells were induced by adding 0.5-mM isopropyl b-D-1-thiogalactopyranoside and incubated for another 4 hours at 37°C. Subsequently, cell cultures were centrifuged at 38 000g for 30 minutes at 4°C, and pellets were suspended in lysis buffer (pH 7.4, containing 20-mM sodium phosphate, 500-mM NaCl, and 10-mM imidazole, supplemented with 1 mg/mL lysozyme, 200 μg/mL DNase, 1× Sigma fast protease inhibitor EDTA free; Sigma-Aldrich, Saint Louis, MO) to minimize protein degradation and incubated on ice for 30 minutes. The cells were sonicated on ice for 6 minutes: 10 seconds on, 20 seconds off, and amplitude 60% (Branson Digital Sonifier; Branson, Brookfield, CT). The resulting lysate was clarified by centrifugation at 38 000g for 30 minutes at 4°C. The hexahistidine-tagged proteins were isolated from the lysate using Nickel-Nitrilotriacetic acid (nickel-nitrilotriacetic acid (Ni-NTA) Superflow (QIAGEN, Hilden, Germany) and an Econo-chromatography column (2.5 × 10 cm; Bio-Rad, Hercules, CA). To get rid of nonspecific interacting proteins, the Ni-NTA resin was washed twice with cold washing buffer (pH 7.4, 20-mM sodium phosphate, 500-mM NaCl, and 10-mM imidazole). The hexahistidine-tagged proteins were eluted from the Ni-NTA group on the matrix with cold elution buffer (pH 7.4, containing 20-mM sodium phosphate, 500-mM NaCl, and 500-mM imidazole). Directly after the proteins were eluted in 500-mM imidazole–containing buffer, the suspensions were dialyzed in a Slide-A-Lyzer G2 dialysis cassette, gamma irradiated with pore size 10 kDa molecular weight cutoff (MWCO; Thermo Fisher Scientific, Waltham, MA), against 300× volume dialysis buffer (50-mM Tris-HCl, pH 7.6, and 150-mM NaCl), for 2 hours at 4°C. Dialysis buffer was exchanged for 2 additional times and dialyzed for 2 hours and 16 hours respectively, at 4°C, before use or storage in small aliquots. The protein concentrations were quantified using a Qubit 3.0 fluorometer (Life Technologies) and Qubit protein assay kit, according to the manufacturer’s protocol. Subsequently, proteins were verified by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), using a Bis-Tris XCell SureLock Mini-Cell system (Thermo Fisher Scientific). The samples were loaded onto a precasted NuPAGE Novex 4% to 12% bis-tris gel (thickness, 1.5 mm; Thermo Fisher) and were run in 1× MOPS (4-morpholinepropanesulfonic acid; Thermo Fisher Scientific) at 200 V for 90 minutes. The blots were incubated with appropriate antibodies: α-His H-15 (Santa Cruz Biotechnology, Dallas, TX), α-DNMT3A (Abcam, Cambridge, UK), α-TDG (Abcam), and α-MBD4 (Abcam) in blocking buffer (5% bovine serum albumin [BSA], 0.1% Tween, and 1× phosphate-buffered saline [PBS]). Finally, the proteins were visualized with the use of a Li-Cor Odyssey 3.0 (Westburg Life Sciences, Leusden, the Neterlands) and corresponding Li-Cor Odyssey software.

MBD4 and TDG glycosylase activity and bandshift assays were performed as described by Hashimoto et al.⁵⁵ See Figure 1B for a schematic visualization of the glycosylase activity assays, and see supplemental Methods for detailed description.

Whole-genome sequencing (WGS) was performed on DNA of 22 bone marrow aspirates from patients with de novo AML (6 DNMT3A WT, 7 DNMT3A SM, and 9 DNMT3A DM). Libraries were generated using the Illumina DNA polymerase chain reaction (PCR)-Free library Prep, tagmentation kit following manufacturer protocol (Illumina). Quality control and quantification of generated libraries were done using Qubit Fluorometric Quantitation 3.0 (Thermo Fisher Scientific), TaqMan quantative PCR (qPCR)) analysis and Agilent 2100 Bioanalyzer platforms (Agilent, Santa Clara, CA). The final WGS libraries were analyzed by 2× 151 cycles paired-end sequencing on a NovaSeq 6000 system (Illumina), with an average coverage of 35× for AML samples and 20× for matched controls. Genomic sequences were aligned against the hg19 reference genome using bwa-mem2 (https://github.com/bwa-mem2/bwa-mem2). The calling of variants, structural variations, and copy number changes was performed using the cancer genome project WGS (CGPWGS) software package (https://github.com/cancerit/dockstore-cgpwgs), using the hg19 reference database, settings, and filtering strategies previously described in detail.⁵⁶ 

AlphaFold version 2.2.0 was used to predict the interaction of WT and mutant forms of DNMT3A with TDG.⁵⁷ The top-ranked relaxed multimer model of each predicted structure was visualized in ChimeraX, and interacting residues were quantified using “interfaces areaCutOff0.”

For patient characteristics, statistical associations between categorical variables were assessed using the Fisher exact test and by the Mann-Whitney U test for continuous variables. Mutation counts derived from WGS were tallied with MutationalPatterns.⁵⁸ Enrichment of mutational signatures SBS1, SBS5, and SBS18 within the tallied mutation counts were determined with sigfit (https://github.com/kgori/sigfit) using default settings. The Kruskal-Wallis test was used to statistically compare SBS1-enrichment scores representing the amount of methylation-linked DNA damage that occurs in each AML. Overall survival (OS) analyses were performed using the Kaplan-Meier method, and survival differences were assessed using the log-rank test. A multivariable Cox proportional-hazards model was used to perform a univariate survival analysis of DNMT3A SM vs DNMT3A DM patients. The model was adjusted for the following covariates: age, white blood cell count at diagnosis, cycles of chemotherapy required to achieve complete remission (CR), mutational status of NPM1, FLT3 internal tandem duplication, myelodysplasia-related gene mutations (ASXL1, BCOR, EZH2, RUNX1, SF3B1, SRSF2, STAG2, U2AF1, and ZRSR2), TP53 at diagnosis, and Dutch-Belgian Hemato-Oncology Cooperative group trial. Allogeneic transplantation was included as a time-dependent variable. Values of P <.05 were considered statistically significant. Statistical analyses were executed with STATA Statistical Software, release 18.0 (College Station, TX) and with R 4.3.1 (Boston, MA).

WT DNMT3A and TDG are known to interact, thereby increasing the repair activity of the latter.⁴² Therefore, we hypothesized that mutant DNMT3A negatively affects the function of the TDG-DNMT3A complex. To assess the impact of mutant DNMT3A on TDG function, we first performed glycosylase activity assays with WT DNMT3A and TDG proteins. In these assays, a TDG construct, including both the catalytic domain and an N-terminal regulatory region (TDG^AA111-348), was incubated with a double-stranded fluorescein amidite (FAM)-labeled 32-bp oligonucleotide harboring a G-T mismatch. Functional TDG^AA111-348 identifies and excises the nucleotide mismatch in this assay. The subsequent hot alkaline treatment facilitates a double-stranded DNA break at the position of the excised base (Figure 1B, product). To verify that recombinant full-length WT DNMT3A enhanced TDG repair activity, 40-nM TDG^AA111-348 was incubated with increasing concentrations of WT DNMT3A. This led to elevated product formation in a dose-dependent matter, thereby confirming the enhanced effect of recombinant WT DNMT3A on TDG repair activity (Figure 1C-D; supplemental Figure 1A,C). Increasing the concentration of WT DNMT3A did not affect MBD4 glycosylase activity (Figure 1E-F; supplemental Figure 1B-C).

Next, we hypothesized that mutant DNMT3A could negatively affect the capacity of TDG to repair DNA damage. Four DNMT3A mutant forms commonly found in CH in healthy as well as in MBD4-deficient individuals (DNMT3A R635W, R688C, R882C, and A884W)⁴¹ were added to glycosylase assays in increasing concentrations to determine the effect on DNA damage repair. All full-length mutant DNMT3A proteins showed no TDG stimulation at low concentrations and even inhibited TDG function at higher concentrations compared to WT DNMT3A (Figure 2; supplemental Figure 2). However, differences in rates of inhibition were apparent between the different DNMT3A mutants. DNMT3A R635W closely mirrored the pattern observed with WT DNMT3A but, in contrast to WT DNMT3A, showed decreased TDG activity at high concentrations (Figure 2A-B; supplemental Figure 2). In ∼80% to 85% of CH and AML cases, DNMT3A mutations are heterozygous. To mimic this setting, mutant forms of recombinant DNMT3A were cotitrated with WT DNMT3A in the glycosylase assays. All DNMT3A mutants clearly suppressed the capacity of TDG to repair DNA damage compared with WT DNMT3A alone. Furthermore, at high concentrations, DNMT3A mutants imposed complete inhibition of TDG, irrespective of the concentration of WT DNMT3A present (Figure 3). Band shift assays confirmed decreased binding affinity of the TDG to DNA in the presence of mutant DNMT3A compared with the WT counterpart (supplemental Figure 3). We were interested in whether these DNMT3A mutations are positioned at the interaction interfaces of DNMT3A and TDG. AlphaFold 2.0 was used to predict the protein interactions between TDG and both WT and mutant forms of DNMT3A. The predictions indicated that all mutations, except for DNMT3A R635W, were likely to disturb the interaction surface between TDG and DNMT3A (supplemental Figure 4). Of note, the R635W mutation was the only mutation leading to less inhibition of TDG repair capacity in our glycosylase assays compared with the other assessed mutations (Figure 2A-B).

We investigated how DNMT3A SM vs DM status affects the mutational AML landscape and whether it correlates with survival. Within our cohort, 1872 individuals were classified as DNMT3A WT, 585 patients were classified as DNMT3A SM, and 88 patients were identified as DNMT3A DM (system of classification; Figure 4A). Among the 88 DNMT3A DM patients, 38 patients harbored a homozygous DNMT3A mutation (supplemental Table 1). One homozygous DNMT3A DM patient had a chromosome 2p deletion that encompasses the DNMT3A gene. All other DM patients had a combination of 2 different DNMT3A mutations with relatively high VAFs (supplemental Table 2). DNMT3A DM patients were significantly older than DNMT3A SM patients (median age, 59 and 55 years, respectively; P = .006), achieved less frequently CR after the first cycle of chemotherapy (69.6% vs 48.9%; P ≤ .001), and were less likely to receive consolidation therapy or allogeneic transplantation in CR (supplemental Table 3). DNMT3A DM patients acquired significantly higher frequencies of mutations in ASXL1 (3.6% vs 9.1%; P = .042), BCOR (5.0% vs 19.3%; P < .001), BCORL1 (1.9% vs 5.7%; P = .046), IDH2 (17.6% vs 34.1%; P = .001), IKZF1 (0.5% vs 4.5%; P = .007), and STAG2 (3.1% vs 9.1%; P = .013) than DNMT3A SM patients, whereas the frequencies of FLT3-TKD (13.2% vs 4.5%; P = .021), and mutations in NPM1 (64.6% vs 26.1%; P < .001) and WT1 (8.0% vs 1.1%; P = .014) were more prevalent in DNMT3A SM (Figure 4B; supplemental Table 4). Of note, a 10-fold enrichment of the IDH2 R172 hot spot mutation was detected in DNMT3A DM patients compared with DNMT3A SM patients (supplemental Figure 5). In both DNMT3A DM and SM mutants, DNMT3A mutations were distributed throughout the entire gene, with R882 mutations being the most prevalent mutation in DNMT3A SM patients, accounting for 328 of 585 mutations (Figure 4D; supplemental Table 1). Surprisingly, the DNMT3A R882 hot spot mutation was only detected in 5 of 142 mutations in the DNMT3A DM cohort (Figure 4D; supplemental Table 1). Within the DNMT3A DM subset with compound heterozygous DNMT3A mutations (n = 50), the presence of 2 nonsynonymous single nucleotide variants was the most frequent. Notably, within the subset of DNMT3A DM patients with 2 LOF mutations, we did not observe the combination of 2 nonsense, 2 splicing, or 2 frameshift mutations. Yet, combinations of 2 different LOF mutations were commonly observed (Figure 4E). In univariate survival analysis, DNMT3A DM patients had significantly poorer OS than DNMT3A SM patients (Figure 4C, independently validated using BEAT-AML data; supplemental Figure 6). After adjusting for covariates known to have a prognostic impact, DNMT3A DM status was considered an independent prognostic marker and associated with poor OS compared with the DNMT3A SM status (hazard ratio, 1.40; 95% confidence interval, 1.05-1.86; P = .021; supplemental Table 5A-B).

Based on the in vitro glycosylase assays, we hypothesized that patients with a complete loss of WT DNMT3A might exhibit greater accumulation of DNA damage. To investigate this hypothesis, we selected 16 patients from our biobank who had either SM (n = 7) or DM (n = 9) DNMT3A mutations that were most likely to result in the complete loss of DNMT3A, along with 6 corresponding controls, and performed WGS on their diagnostic bone marrow aspirates. An overview of the selected patients with AML is listed in Table 1. All selected DNMT3A SM and DM cases carried IDH1/2 mutation. This selection was made to rule out possible biases introduced by IDH mutations, which are known to be associated with DNA hypermethylation. The DNMT3A WT group contained both IDH mutant and IDH WT patients. For an overview of all comutations, see supplemental Table 6A-C. AlphaFold 2.0 was used to model the interaction between DNMT3A and TDG to determine the possible impact of DNMT3A mutations positioned at the interaction interface. In general, DNMT3A DM patients had a relative higher SBS1 enrichment than DNMT3A SM and WT patients (Figure 5A). In line with our hypothesis, AMLs with 2 DNMT3A mutations located at the TDG-DNMT3A interaction interface tended to have a SBS1 enrichment compared with patients with 1 or no DNMT3A mutation at this interface (Figure 5B). One DNMT3A DM patient, with a low SBS1 enrichment (DM-2), had a homozygous L637P mutation, which is predicted not to be involved in DNMT3A-TDG interaction, thereby potentially explaining the lower SBS1 enrichment observed. We observed no differences in gene expression between DNMT3A SM and DM patients (data not shown); therefore, we believe that DNMT3A DM patients form a unique subtype based on their level of methylation damage and subsequent genetic landscape, rather than a distinct gene expression profile.

Spontaneous deamination of 5mC is a naturally occurring mutational process. If this DNA damage is not corrected on time, it will lead to C-to-T transition mutations. To prevent the acquisition of these mutations, life has evolved a DNA repair system involving the 2 DNA glycosylase enzymes, MBD4 and TDG, which safeguard against DNA damage through BER. DNMT3A, after repair, methylates the 5-cytosine position, restoring the original DNA methylation pattern. Li et al showed that, in the process of DNA (de)methylation, DNMT3A and TDG interact, thereby regulating the activity of both proteins and linking DNA repair to methylation.⁴² 

DNMT3A mutations predicted to impair the function of the TDG-DNMT3A complex, that is, DNMT3A R688C, R882C, and A884W, confer a remarkable decrease in TDG repair activity in our glycosylase assays. This inhibitory effect persists when mutant DNMT3A was cotitrated with WT DNMT3A, mimicking a heterozygous situation (the most common situation in CH and AML). In contrast, a single DNMT3A mutant (DNMT3A R635W), predicted to not be present at the interaction interface, only exhibits decreased TDG activity at high concentrations. Interestingly, we observed greater DNA damage in primary AML samples carrying 2 TDG-DNMT3A interaction–altering variants than those with 1 or no interaction altering mutation. Notably, the DNMT3A DM patient homozygous for the DNMT3A L637P mutation showed relatively low DNA damage. This missense mutation is proximal to the DNMT3A R635W mutation, which shows only moderate inhibition of TDG in our glycosylase assay. Collectively, these results indicate that mutated DNMT3A retains its ability to bind and enhance TDG function, provided that the mutations are positioned outside the interacting interface. It is postulated that WT DNMT3A facilitates TDG binding to G-T mismatches and with subsequent processing; however, mutant DNMT3A may induce complex conformational changes that interfere with the ability of TDG to bind DNA, thereby diminishing its repair function.⁴² Supporting this notion, our band shift assays confirm a decrease in TDG binding affinity to DNA in the presence of mutant DNMT3A.

Additionally, AlphaFold 2.0 predicted changes in residues of both mutant DNMT3A and TDG, resulting in novel interactions. Most forms of mutant DNMT3A are predicted to interact with residues located within the uracil-DNA glycosylase domain of TDG, whereas WT DNMT3A is predicted not to interact with this domain (supplemental Figure 7). Given the significance of this domain for TDG repair capacity, such changes in interaction may further affect TDG repair activity. However, to validate these interactions and potential structural changes, protein crystallography experiments would be necessary to visualize the TDG-DNMT3A protein complex upon mutation of DNMT3A. In addition, the exact mechanisms through which mutant DNMT3A impairs TDG function remain unknown and should be investigated in further detail.

The results of the glycosylase assays, representing a heterozygous mutant context, indicate that titration of increasing amounts of recombinant mutant DNMT3A at steady levels of WT DNMT3A inhibit TDG repair capacity. In fact, as the concentration of mutant DNMT3A equals or surpasses that of WT DNMT3A, TDG function appears to decline. In this situation, the activity level of TDG becomes even less compared with baseline without the addition of any WT DNMT3A (Figure 3). These results suggest that both WT and mutant DNMT3A compete to interact with TDG. When mutant DNMT3A concentration exceeds that of WT DNMT3A, a likely greater proportion of TDG interaction sites will be occupied by mutant DNMT3A, leading to a reduction in glycosylase activity.

DNMT3A DM AML with biallelic inactivating mutations, likely lacking functional DNMT3A, were subjected to WGS. The trend of increased DNA damage in these patients with AML, although the cohort size is limited by biobank availability, suggests that the acquired DNMT3A mutations impair TDG function. Moreover, these patients still had functional MBD4, which remains capable of recognizing and repairing G-T mismatches. Because both our initial experiments and available literature did not suggest interactions between MBD4 and DNMT3A, we did not include the results on the impact of mutant DNMT3A on MBD4 repair function.

To assess the clinical impact of DNMT3A mutational status, we compared patients carrying single or double DNMT3A mutations. We found that both DNMT3A SM and DNMT3A DM mutations were distributed across the entire gene. However, mutations occurring at R882 (hot spot mutation), which accounts for almost 60% of all DNMT3A mutations in AML, were almost exclusively seen in DNMT3A SM patients.⁵⁹ Previous studies have shown that R882 mutations inhibit WT DNMT3A to form active tetramers, thereby altering the process of CpG methylation, which might be another contributing factor of DNMT3A mutation in oncogenesis.^20,23,60 Furthermore, patients with DNMT3A DM AML were of significantly older age, and in the context of an intensive chemotherapy setting, they were less likely to attain an early CR (ie, after the first cycle of remission induction chemotherapy); and their OS was reduced compared with that of patients with DNMT3A SM AML. These findings extend the limited research that has been conducted on this AML subgroup previously and establish DNMT3A DM as a distinct subset of AML.²⁹ In a multivariable analysis investigating OS, after adjusting for several clinically relevant covariates, DNMT3A DM status remained a significant independent risk factor for poorer OS. Taking into account the older age and mutational landscape of patients with DNMT3A DM AML, it would be of great interest to explore the responses to hypomethylating agents in combination with venetoclax or ivosidenib (in IDH1 mutant AML), considering that these treatments show higher response rates and longer OS in unfit patients with AML.^61,62 Altogether, our study indicates that DNMT3A DM AML, a unique subgroup comprising ∼4% of the complete AML population, experiences increased DNA damage, and DNMT3A mutational status carries significant prognostic value.

The authors thank all the patients, clinicians, research nurses, data managers, and laboratory scientists who provided samples, conducted cytogenetic analyses, and performed analyses at the participating centers of the Dutch-Belgian Cooperative Trial Group for Hematology-Oncology and Swiss Group for Clinical Cancer Research, including centers in the Netherlands, Belgium, Switzerland, Norway, Sweden, Lithuania, Germany, and Denmark. The authors also thank Eric Bindels for performing next-generation sequencing and Remco Hoogenboezem and Dorien Pastoors for their help with data analysis.

M.A.S. and S.M. are supported by a Koningin Wilhelmina Fonds voor de Nederlands Kankerbestrijding Young Investigator Grant (12797, Bas Mulder Award; Dutch Cancer Foundation).

Contribution: P.J.M.V. and M.A.S. designed the research; P.J.M.V. and B.L. provided patient information and materials; E.L.B., A.Z., M.R., and F.G.K. performed experiments; E.L.B., S.M., M.A.S., and J.K. analyzed data; T.G. and J.V. provided statistical advice; E.L.B. and S.M. prepared the figures; E.L.B., S.M., P.J.M.V., and M.A.S. wrote the manuscript; and all authors approved the final version of the manuscript.

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

Correspondence: Mathijs A. Sanders, Department of Hematology, Erasmus Medical Center Cancer Institute, Ee-1330e, Dr. Molewaterplein 80, 3015 GD Rotterdam, the Netherlands; email: m.sanders@erasmusmc.nl.