Genomic and transcriptomic determinants of clinical outcomes in patients with AML and DNMT3A mutations

Abstract

Acute myeloid leukemia (AML) and DNMT3A mutations (DNMT3A^mut) are considered to carry intermediate risk under the 2022 European LeukemiaNet (ELN-2022) classification in the absence of other co-mutations or cytogenetic abnormalities. However, this group is highly heterogeneous. In this study, the genomic and transcriptomic features influencing outcomes in DNMT3A-mutated AML were examined in a cohort of 884 patients with AML receiving standard chemotherapy. Stratification by NPM1 and FLT3-ITD status revealed worse survival among patients with NPM1 mutations and wild-type FLT3-ITD (NPM1^mut/FLT3-ITD^wt) than patients in the ELN-2022 favorable risk group. The other three subgroups (NPM1^mut/FLT3-ITD^mut, NPM1^wt/FLT3-ITD^mut, and NPM1^wt/FLT3-ITD^wt) exhibited worse prognoses than patients in the ELN-2022 intermediate risk group. Additionally, the presence of TET2^mut in patients with AML and DNMT3A^mut/NPM1^mut/FLT3-ITD^wt led to reclassification from favorable risk to intermediate risk in the ELN-2022. RNA-sequencing analysis revealed a distinct transcriptomic profile in patients with TET2^mut, highlighting the enrichment of leukemic stem cell signatures and dendritic cell migration, with MMP14, CD200, and CT45A5 identified as key differentially expressed genes. In conclusion, co-mutation patterns strongly affected AML outcomes in patients with DNMT3A^mut. Patients with TET2^mut constituted a unique subgroup within the ELN-2022 favorable DNMT3A^mut/NPM1^mut/FLT3-ITD^wt group, characterized by distinct transcriptomic features and an unfavorable prognosis.

Introduction

DNMT3A encodes the DNA methyltransferase 3 alpha enzyme, which adds methyl groups to cytosine residues in CpG dinucleotides [1]. This process is essential for gene expression regulation and the maintenance of genomic stability. Mutations in DNMT3A (DNMT3A^mut) have been identified in approximately 14–22% of patients with acute myeloid leukemia (AML) [1,2,3,4,5], with a higher prevalence of 29–34% observed in those with normal karyotype AML [6, 7]. Numerous studies have demonstrated a correlation between DNMT3A^mut and poorer survival outcomes [1, 2, 8,9,10,11]. However, the prognostic implications of DNMT3A^mut remain uncertain, because these mutations frequently exist alongside other genetic alterations, hindering identification of their independent effects on clinical outcomes. Notably, DNMT3A^mut is not listed as a major prognostic factor in the 2022 European Leukemia Net (ELN-2022) classification [12].

Most DNMT3A^mut appear alongside other molecular abnormalities, such as FLT3-ITD, NPM1, and IDH1, underscoring the clinical heterogeneity among individuals with these mutations. NPM1 mutations (NPM1^mut) have exhibited a strong association with favorable outcomes, whereas FLT3-ITD mutations (FLT3-ITD^mut) have been linked to poorer outcomes [13, 14]. However, the prognostic implications of other concurrent mutations in patients with AML and DNMT3A^mut require further clarification. Moreover, the transcriptomic features that affect survival in patients with DNMT3A^mut remain poorly understood, indicating a need for additional research. Dysregulated pathways that function independently of specific mutational backgrounds should be elucidated to uncover new opportunities for therapeutic intervention. In this study, we performed a detailed subgroup analysis of clinical heterogeneity in patients with AML and DNMT3A^mut based on comprehensive genomic and transcriptomic data. Our findings were also thoroughly validated in several external cohorts.

Methods and materials

Sample

A total of 884 newly diagnosed patients with de novo non-M3 AML who received standard chemotherapy at National Taiwan University Hospital (NTUH) were enrolled in this study. AML diagnosis and classification were performed according to the International Consensus Classification [15] and the 5^th World Health Organization Classification [16]. Detailed patient characteristics and treatment regimens are provided in the supplementary data [17].

Ethics approval and consent to participate

All methods were performed in accordance with the relevant guidelines and regulations. Live vertebrates were not used in this project. The study was approved by the NTUH institutional review board, and written informed consent was obtained from all participants in accordance with the Declaration of Helsinki (approval number 201709072RINC and 202109078RINB).

Cytogenetics

Chromosomal analyses were performed according to a procedure described in a previous study [18]. Karyotypes were classified on the basis of risk groups defined by the Medical Research Council [19].

Mutation analysis

We performed targeted next generation sequencing (NGS) with a TruSight myeloid sequencing panel (Illumina, San Diego, CA, USA) to analyze mutations in 15 full exon genes and 39 oncogenic hotspot genes. A HiSeq platform (Illumina, San Diego, CA, USA) with a median reading depth of 12,000× was used for sequencing. Due to the suboptimal sequencing sensitivity of this method, CEBPA mutations were confirmed through Sanger sequencing. FLT3-ITD was analyzed through polymerase chain reaction (PCR) followed by fluorescence capillary electrophoresis. KMT2A-PTD was analyzed through PCR followed by Sanger sequencing [20].

RNA sequencing analysis

Bone marrow (BM) mononuclear cells were obtained from 264 patients with AML, including 80 with DNMT3A^mut and 184 with wild-type DNMT3A (DNMT3A^wt) and normal karyotypes. The BM samples were then analyzed through RNA sequencing (RNA-seq) per the method of a previous study [21]. RNA was processed into sequencing libraries with the TruSeq Stranded mRNA Library Prep Kit (Illumina, San Diego, CA, USA) and sequenced on an Illumina NovaSeq 6000 sequencing system (150 bp paired-end). Reads were trimmed with Cutadapt (v2.3), aligned to GRCh38.p12 using STAR (v2.6.1a) [22], and quantified with Gencode annotations (v28) [23]. Read counts were normalized across all samples according to the trimmed mean of M-values with the calcNormFactors function of the edgeR package in R software [24]. Gene expression was calculated in terms of log₂(CPM + 1) (counts per million reads) for further analysis. We used the limma package in R to identify differentially expressed genes (DEGs) and determined log2 fold changes (logFC) and false discovery rate (FDR)-adjusted P values using the Benjamini–Hochberg method.

We conducted analyses of gene ontology to investigate the enrichments of expression profiles associated with DNMT3A^mut in gene sets curated in the Molecular Signatures Database (MSigDB) [25]. The statistical significance of the degree of enrichment was assessed through a random permutation test with 1000 repetitions [26].

External datasets

We downloaded AML RNA-seq data from the Cancer Genome Atlas (TCGA) generated with the Illumina HiSeq 2000 platform (n = 173) and Beat-AML RNA-seq data for validation [27, 28]. The prognostic relevance of concurrent mutations was further validated on the basis of patient data with molecular annotations from the German–Austrian Acute Myeloid Leukemia Study Group (AMLSG, n = 1540) and the UK National Cancer Research Institute (UK-NCRI) trials (n = 2113) [4, 29].

Statistical analysis

Chi-square tests were performed for comparisons of discrete variables. If the expected contingency table values were below 5, Fisher’s exact test was performed. Mann–Whitney U tests were performed to compare continuous variables and the medians of distributions. Overall survival (OS) was defined as the time from initial diagnosis to the date of last follow-up or death from any cause. Event-free survival (EFS) was defined as the time from diagnosis to the first instance of the following: treatment failure, relapse after first complete remission (CR1), or death from any cause. Disease-free survival (DFS) was defined as the time from diagnosis to either relapse or death from any cause [30]. A P value < 0.05 indicated statistical significance. The effect of allogeneic hematopoietic stem cell transplantation (allo-HSCT) at CR1 on outcomes in patients with DNMT3A^mut was assessed according to the Mantel–Byar approach, and results were visualized in Simon–Makuch plots [31]. Statistical analyses were performed with SPSS 23 (SPSS, Chicago, IL, USA) and R software 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Patient characteristics and clinical outcomes

DNMT3A^mut was identified in 16.2% of the 884 patients in our cohort (143) (Supplementary Table 1). The median age of patients with DNMT3A^mut was 53 years (range: 17–82 years), significantly higher than that of patients with DNMT3A^wt. Patients with DNMT3A^mut also had significantly higher white blood cell counts, platelet counts, and hemoglobin levels at diagnosis. Cytogenetic analysis revealed a significantly higher proportion of patients with DNMT3A^mut had intermediate risk profiles than those with DNMT3A^wt (93.0% vs. 72.1%, respectively, P < 0.001). Patients with DNMT3A^mut had a higher incidence of NPM1^mut but a lower incidence of AML with core-binding factor positivity, CEBPA mutations, and KMT2A rearrangements (Table 1).

Table 1 Comparison of clinical and laboratory features between AML patients with and without DNMT3A mutations.

CR1 rates were comparable between patients with DNMT3A^mut and DNMT3A^wt (74.1 vs. 75.8%, respectively, P = 0.662). Among patients who did not receive allo-HSCT at CR1, relapse rates remained comparable between both groups (40.6% vs. 32.0%, P = 0.102). Patients with DNMT3A^mut exhibited similar OS and EFS as those in the ELN-2022 intermediate risk group (Fig. 1A). Patients in this cohort with DNMT3A^wt had significantly better OS and EFS than those with DNMT3A^mut after a median follow-up time of 7.3 years (median OS 30.1 vs. 21.3 months, respectively, P = 0.005; median EFS 12.6 vs. 7.4 months, P = 0.009) (Fig. 1B).

Fig. 1: Survival outcomes in patients with DNMT3A^mut.

A Comparison of Kaplan–Meier survival curves for OS and EFS among patients with AML stratified according to ELN-2022 risk groups (favorable, intermediate, adverse) and DNMT3A mutation status. OS and EFS for the DNMT3A^mut group closely aligned with those of the ELN-2022 intermediate risk group. B DNMT3A^mut were associated with significantly worse OS and EFS than DNMT3A^wt.

Frequency and prognostic relevance of concurrent mutations in DNMT3A ^mut

At least one concurrent mutation was identified in most patients with DNMT3A^mut (97.2%) (Fig. 2A). The most frequently mutated genes were NPM1 (52.8%), FLT3-ITD (30.6%), IDH2 (26.4%), FLT3-TKD (15.3%), IDH1 (10.4%), and PTPN11 (9.7%) (Fig. 2B, Supplementary Table 2). Given the well-established prognostic significance of NPM1^mut and FLT3-ITD^mut, we stratified patients with DNMT3A^mut into four subgroups by whether these two genetic alterations were present. To further strengthen risk stratification, we also assessed the prognostic effects of concurrent mutations within each subgroup. Patients with DNMT3A^mut and different NPM1 and FLT3-ITD mutation statuses differed significantly in OS (P = 0.004) and EFS (P = 0.009) (Fig. 2C). Outcomes were significantly better among DNMT3A^mut patients with NPM1^mut and wild-type FLT3-ITD (FLT3-ITD^wt) than in patients with NPM1^mut/FLT3-ITD^mut, NPM1^wt/FLT3-ITD^mut, or NPM1^wt/FLT3-ITD^wt, in which prognoses were similarly poor. Furthermore, outcomes were significantly worse in the NPM1^mut/FLT3-ITD^wt subgroup than in the ELN-2022 favorable risk group (median OS, 97.8 months vs. not reached, P = 0.047; median EFS, 18.0 vs. 138.1 months, P = 0.036). The other three subgroups all exhibited worse outcomes than the ELN-2022 intermediate risk group (Supplementary Fig. 1A–D). These results indicate that outcomes among all patients with DNMT3A^mut were inferior to those predicted under the current ELN-2022 risk categories.

Fig. 2: Prevalence and prognostic implications of concurrent mutations in patients with DNMT3A^mut.

A Oncoprint of recurrent genetic mutations in patients with DNMT3A^mut and AML. Colored bars represent mutations, categorized according to functional gene classes. B Prevalence of concurrent mutations in patients with DNMT3A^mut and DNMT3A^wt. Mutations observed in at least 1% of patients in one subgroup are displayed. Significant differences between DNMT3A^mut and DNMT3A^wt groups are marked as follows: *P < 0.05, **P < 0.01, ***P < 0.001. C OS and EFS of patients with DNMT3A^mut stratified according to NPM1 and FLT3-ITD status. Patients with NPM1^mut/FLT3-ITD^wt exhibited significantly better OS and EFS than those in other groups. D OS and EFS in the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt group, stratified according to concurrent genetic mutations: TET2, IDH2, signaling pathway mutations (PTPN11, FLT3-TKD, and NRAS), and total subgroup patients. The worst outcomes were observed in the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^mut subgroup. E OS and EFS comparison between the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^mut subgroup and ELN-2022 favorable risk group. Worse outcomes were observed in the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^mut subgroup. F OS and EFS comparison between the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^wt subgroup and ELN-2022 intermediate risk group. Better outcomes were observed in the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^wt subgroup.

The highest frequency of co-mutations was observed in the NPM1^mut/FLT3-ITD^wt group, including FLT3-TKD (25.5%), IDH2 (19.1%), PTPN11 (17.0%), NRAS (14.9%), and TET2 (14.9%). Stratification according to these co-mutations revealed that TET2 mutations (TET2^mut) were associated with the lowest survival rates (Fig. 2D). Notably, in our analysis of patients based on ELN-2022 risk stratification, those with TET2^mut exhibited significantly worse OS and a trend towards inferior EFS than patients in the ELN-2022 favorable risk group (median OS, 22.9 months vs. not reached, P = 0.008; median EFS, 11.8 vs. 138.2 months, P = 0.054) (Fig. 2E). Patients in our cohort with TET2^mut shared similar OS and EFS with the ELN-2022 intermediate risk group (Supplementary Fig. 2A). By contrast, among patients without TET2^mut, OS and EFS were higher relative to patients in the ELN-2022 intermediate risk group (median OS, 98.2 vs. 24.2 months, P = 0.013; median EFS, 18.7 vs. 10.7 months, P = 0.028) (Fig. 2F) and similar relative to patients in the ELN-2022 favorable risk group (Supplementary Fig. 2B). The DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^mut group exhibited a trend towards worse OS comparing with the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/wild-type TET2 (TET2^wt) group (median, 22.9 vs. 98.2 months, respectively, P = 0.097) (Supplementary Fig. 2C). In patients with DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/PTPN11^wt/IDH2^wt, TET2^mut were associated with significantly worse OS (TET2^mut, median, 1.3 months vs. not reached, P = 0.024) and a trend toward worse EFS (TET2^mut, median, 0.001 vs. 33.2 months, P = 0.090) (Supplementary Fig. 2D).

In the NPM1^mut/FLT3-ITD^mut group, frequent co-mutations were observed in myeloid dysplasia related genes (MDS-R, 16.6%), including ASXL1, BCOR, EZH2, RUNX1, SF3B1, SRSF2, STAG2, U2AF1, and ZRSR2, and in signaling pathways (13.3%). MDS-R genes were most frequently detected in the NPM1^wt/FLT3-ITD^mut group (9.3%), whereas the proportion of other co-mutations remained low. Notably, no significant differences in OS or EFS were observed within each group when stratified according to the type of co-mutations (Supplementary Fig. 2E, F). A high frequency of co-mutations was observed in the NPM1^wt/FLT3-ITD^wt group, including IDH2 (48.1%), MDS-R genes (44.4%), and NRAS (16.7%). Patients with MDS-R genes were classified within the ELN-2022 unfavorable group and had a 5-year survival rate of approximately 25–30% [23]. Even patients with DNMT3A^mut/NPM1^wt/FLT3-ITD^wt who lacked MDS-R gene mutations were associated with poor survival outcomes that were similar to those of patients with MDS-R gene mutations, indicating the negative prognostic effects of DNMT3A^mut (Supplementary Fig. 2G). Patients with IDH2 mutations were excluded for further stratification of the NPM1^wt/FLT3-ITD^wt/IDH2^wt group according to NRAS mutation status, which revealed that patients with NRAS mutations had better OS (median, not reached vs. 12.4 months, P = 0.045) (Supplementary Fig. 2H).

Effects of Allo-HSCT at CR1 in patients with DNMT3A ^mut

Outcomes among patients with DNMT3A^mut, irrespective of their NPM1 and FLT3-ITD mutation status, were inferior to those predicted under the current ELN-2022 risk categories. We also examined whether allo-HSCT at CR1 alleviated the poor survival of patients with DNMT3A^mut. Among these patients, 64 (44.8%) underwent allo-HSCT, with 29 (20.3%) receiving allo-HSCT at CR1. Among patients who received allo-HSCT at CR1, 12 had NPM1 mutations and 10 had available NPM1 PCR data before HSCT. NPM1 was detected in 5 (50%) of these patients before the transplant (Supplementary table 3). To accurately evaluate the effects of allo-HSCT at CR1, we performed a Mantel–Byar analysis and constructed Simon–Makuch plots. Significantly better OS (median, 209.1 vs. 23.1 months, P = 0.001) and DFS (median, not reached vs. 10.2 months, P < 0.001) was observed among patients who received allo-HSCT at CR1 (Supplementary Fig. 3A). Among patients in the ELN-2022 favorable group with DNMT3A^mut/NPM1^mut/FLT3-ITD^wt, allo-HSCT at CR1 was associated with significantly improved DFS (median, not reached vs. 93.9 months, P = 0.037) but a nonsignificant difference in OS (median, not reached vs. 241.8 months, P = 0.196) compared with post-remission chemotherapy (Supplementary Fig. 3B).

External validation of survival outcomes stratified according to co-mutations in patients with AML and DNMT3A ^mut

We integrated data from the AMLSG and UK-NCRI trial cohorts to analyze the prognostic effects of co-mutations (Supplementary Table 4). In the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/PTPN11^wt/IDH2^wt subgroup, TET2^mut were associated with significantly worse OS (TET2^mut, median, 2.9 years vs. not reached, P = 0.020) (Supplementary Fig. 4), consistent with findings from our cohort.

Characteristics and clinical outcomes of patients with DNMT3A ^mut in relapse

Among patients with DNMT3A^mut, 63 (44.1%) experienced a relapse, and 34 (54.0%) of all relapsed patients achieved a second complete remission (CR2) after salvage reinduction chemotherapy (Supplementary Table 5). FLT3-ITD and WT1 mutations were associated with a significantly lower likelihood of CR2 (FLT3-ITD, 38% vs. 72%, P = 0.045; WT1, 0% vs. 100%, P = 0.012). The median OS from the time of relapse was 10.5 months. Subsequent allo-HSCT was administered to 30 patients, including 12 at CR2. Patients who received allo-HSCT after relapse had significantly better OS than those without allo-HSCT treatment (median, 14.0 vs. 5.6 months, P = 0.038) (Supplementary Fig. 5).

Comparison of transcriptional signatures between patients with DNMT3A ^mut and DNMT3A ^wt

Gene expression profiles were compared between 80 patients with DNMT3A^mut and 184 patients with DNMT3A^wt and normal karyotypes. Hierarchical clustering was performed for the top DEGs to increase the clustering resolution (Fig. 3A). A total of 818 upregulated and 454 downregulated DEGs were identified between patients with DNMT3A^mut and DNMT3A^wt on the basis of a cutoff of |logFC|> 0.585 (absolute fold change >1.5 or <0.67) and P < 0.05. Several Hox family genes were among the top upregulated DEGs (Fig. 3B). Using the MsigDB Hallmark gene sets, we observed that the DNMT3A^mut transcriptome was enriched for TNFα signaling (normalized enrichment score ([NES] 3.89, FDR 3.85E−10), complement system (NES 3.34, FDR 3.85E−10), inflammatory response (NES 3.27, FDR 3.85E−10), and IL6/JAK/STAT3 signaling (NES 2.74, FDR 3.85E−10). The chemical and genetic perturbations (CGP) gene sets revealed the positive enrichment of several AML-related gene clusters, including the VERHAAK NPM1-mutated AML signature (NES 4.53, FDR 1.44E−9) and AML with aberrant cytoplasmic localization of NPM1 (NES 3.80, FDR 1.44E−9) (Fig. 3C–E, Supplementary Table 6). In a subgroup analysis of 63 patients with DNMT3A^mut and normal karyotypes, the majority of originally identified DEGs remained significant (1215 of 1,272, 95.5%) (Fig. 3F).

Fig. 3: Comparison of transcriptional signatures between patients with DNMT3A^mut and DNMT3A^wt.

A Unsupervised hierarchical clustering of patients with DNMT3A^mut and DNMT3A^wt patients, with the major upregulated and downregulated DEGs displayed. Colors represent normalized gene expression values. B Volcano plot comparison of all RNA-seq genes between patients with DNMT3A^mut and DNMT3A^wt. Positive (logFC > 0) and negative (logFC < 0) fold changes indicate upregulated and downregulated genes, respectively, in patients with DNMT3A^mut. C, D Bubble plots illustrating a comparison of enriched and depleted MSigDB CGP and Hallmark gene sets between patients with DNMT3A^mut and DNMT3A^wt. Bubble size represents gene counts, and colors indicate statistical significance. E Gene Set Enrichment Analysis plots displaying overrepresentation of immune gene set signatures within the DNMT3A^mut transcriptome. F Venn diagram of DEGs identified in initial and updated comparisons between patients with DNMT3A^mut and DNMT3A^wt. An initial analysis included RNA-seq data from 80 patients with DNMT3A^mut and 184 with DNMT3A^wt and normal karyotypes. In an updated analysis, a subset of 63 DNMT3A^mut patients with normal karyotypes were compared with the original 184 patients with DNMT3A^wt.

To validate our RNA-seq data, we compared the transcriptomes of 40 patients with DNMT3A^mut and 59 patients with DNMT3A^wt and normal karyotypes in the TCGA cohort. Hallmark gene sets for TNFα signaling, the complement system, inflammatory response, and shared AML-related gene clusters in the CGP gene sets were among the primary enriched pathways. Further validation within the Beat-AML cohort revealed comparable results. These findings confirm the alignment between our RNA-seq data and that previously reported (Supplementary Figs. 6 and 7) and underscore the unique immune microenvironment in these patients.

Distinct transcriptomic pathways in patients with DNMT3A ^mut/NPM1 ^mut/FLT3-ITD^wt/TET2 ^mut

We observed a correlation between TET2^mut and unfavorable outcomes in patients with DNMT3A^mut/NPM1^mut/FLT3-ITD^wt and further investigated whether these patients exhibited overlapping transcriptomic profiles. We compared the RNA-seq data between patients with TET2^mut (n = 4) and those with TET2^wt (n = 25) within the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup. The TET2^mut group was identified as a distinct cluster through hierarchical clustering (Fig. 4A). Additionally, the principal component analysis plot effectively distinguished patients with TET2^mut from those with TET2^wt. By contrast, patients with FLT3-ITD, NRAS, and PTPN11 mutations could not be clustered separately from those without these mutations (Supplementary Fig. 8A).

Fig. 4: Differential transcriptional signatures between patients with TET2^mut and TET2^wt within the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup.

A Unsupervised hierarchical clustering of TET2^mut and TET2^wt patients within the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup, displaying the top upregulated and downregulated DEGs. Colors represented normalized gene expression values. B Number of upregulated and downregulated DEGs identified through comparison of transcriptomes in patients with AML with and without TET2, FLT3-TKD, IDH2, NRAS or any other co-mutations. Bubble plots illustrating enriched and depleted MSigDB Hallmark (C) and CGP and Gene Ontology Biological Processes gene sets (D) between patients with TET2^mut and TET2^wt in the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup. Bubble size represents gene counts, and color indicates statistical significance. E Gene Set Enrichment Analysis plots indicating enrichment of LSC and immune gene set signatures within the TET2^mut transcriptome. F, G Volcano plot of DEGs with genes downregulated (logFC < 0) and upregulated (logFC > 0) in patients with DNMT3A^mut/NPM1^mut/FLT3-ITD^wt based on TET2 mutation status. Key DEGs are highlighted in a heat map, with colors representing normalized gene expression values.

A total of 3801 DEGs were identified between the TET2^mut and TET2^wt groups (Fig. 4B and Supplementary Table 7). The number of DEGs between the TET2^mut and TET2^wt groups was notably higher than that between other mutated subgroups. Leukemic stem cell (LSC) signatures were enriched in the TET2^mut group (NES 2.42, FDR 1.47E-4), and genes downregulated in LSC were negatively enriched. An analysis of the MsigDB Hallmark and Gene Ontology Biological Process gene sets revealed significant negative enrichment of IL6/JAK/STAT3 signaling (NES −1.60, FDR 8.42E−3), pathways associated with dendritic cell migration (NES − 2.20, FDR 8.80E-5), and lymphocyte chemotaxis (NES −1.95, FDR 2.08E−3) (Fig. 4C–E and Supplementary Fig. 8B). The top DEGs included MMP14 (logFC = 1.89), CD200 (logFC = 1.39), CT45A5 (logFC = –2.60), and CT45A8 (logFC = –1.71) (Fig. 4F–G). MMP14, derived from mesenchymal stem cells, promotes AML progression and chemoresistance [32]. CD200 is a novel LSC marker that can capture the entire LSC compartment from samples provided by patients with AML, including those with NPM1 mutations [33]. The CT45 gene family consists of 10 distinct but highly similar genes that are nearly identical at the protein level (CT45A1–CT45A10). CT45 peptides can induce CD8⁺ T cell activation through intracellular IFNγ staining [34].

To validate these results and strengthen the robustness of our data, we analyzed RNA-seq data from 40 patients with DNMT3A^mut in the TCGA AML cohort, 6 of whom had TET2^mut. Selected AML gene clusters, LSC signatures, and pathways related to dendritic cell migration were significantly enriched in patients with TET2^mut (Supplementary Fig. 9), in alignment with our findings. Additionally, MMP14 and CD200 were identified as DEGs in the TCGA cohort.

Discussion

Our findings underscore the complexity of AML in patients with DNMT3A^mut and the prognostic implications of its associated molecular landscape. Patients with DNMT3A^mut were stratified into four subgroups on the basis of their NPM1 and FLT3-ITD mutation status. All subgroups exhibited a poorer prognosis than that predicted under each designated ELN-2022 risk category. Additionally, within the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt group, TET2 mutations may have characterized a unique subgroup with distinct transcriptomic features and an unfavorable prognosis.

In previous studies, patients with DNMT3A^mut have exhibited inferior OS and EFS [1, 2, 35]. In our study, patients with DNMT3A^mut were stratified on the basis of their NPM1 and FLT3-ITD mutation status, and the results indicated the profound prognostic implications of these co-mutations. The DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup trended toward worse outcomes than the ELN-2022 favorable risk group. These findings suggest that despite the traditional classification of the NPM1^mut/FLT3-ITD^wt subgroup as favorable, patients in this group with DNMT3A^mut may fall under a distinct intermediate risk category. The remaining three subgroups, NPM1^mut/FLT3-ITD^mut, NPM1^wt/FLT3-ITD^mut, and NPM1^wt/FLT3-ITD^wt, all exhibited poorer prognoses than the ELN-2022 intermediate risk group, aligning more closely with the ELN-2022 unfavorable risk criteria. These results highlight the heterogeneity within the DNMT3A^mut population and the need for more precise risk stratification. Our findings further underscore the critical role of allo-HSCT at CR1 in improving survival outcomes for patients with AML and DNMT3A^mut. Among all patients with AML and DNMT3A^mut, significantly better OS and DFS was observed in those who received allo-HSCT at CR1 than in those who received post-remission chemotherapy. Within the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup, allo-HSCT at CR1 improved DFS, whereas OS remained comparable with other groups. Although we conducted Mantel–Byar analysis and constructed Simon–Makuch plots to address immortal time bias, further randomized studies are necessary to clarify the role of allo-HSCT in this patient population.

To better understand why worse outcomes were observed in the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup than in ELN-2022 favorable risk patients, we investigated the effects of co-mutations within this subgroup. Combined DNMT3A^mut and TET2^mut in patients with AML and NPM1^mut/FLT3-ITD^wt resulted in a risk reclassification from ELN-favorable to intermediate, whereas patients without TET2^mut had survival more similar to patients in the ELN favorable risk group. Our study presents the novel finding that TET2^mut may adversely affect prognosis in patients with DNMT3A^mut/NPM1^mut/FLT3-ITD^wt. We refined the subgroup to exclude PTPN11 and IDH2 mutations, which further underscored the deleterious effects of TET2^mut on survival and disease progression. This finding was sufficiently validated in external cohorts. Among all patients with AML, TET2^mut have been associated with adverse prognosis [36,37,38,39]. In a recent study, two epigenetic modifiers, TET2 and DNMT3A, synergistically induced leukemogenesis [40]. In a DNMT3A and TET2 double-knockout mouse model, dysregulated self-renewal of hematopoietic stem cells may have caused hematological abnormalities. Both DNMT3A and TET2 play key roles in repressing lineage-specific transcription factors in hematopoietic stem cells. The concurrent loss of both genes causes notable upregulation of downstream transcription factors, which eventually leads to malignant transformation [40]. In our study, patients with AML and TET2^mut formed a separated transcriptomic cluster within the DNMT3A^mut/NPM1^mut/FLT3-ITD^wt subgroup and exhibited the largest number of DEGs compared with patients with TET2^wt. Enrichment of LSC signatures may partly explain the poor outcomes observed in this group, and these results were validated in the TCGA cohort. Based on the distinct transcriptomic signatures, prognostic implications, reproducibility across cohorts, and functional evidence, DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^mut AML can be redefined as a distinct subgroup to improve ELN classification systems and risk stratification.

Our study has several limitations. First, we relied on bulk BM aspirates for RNA-seq analysis. Further research incorporating single-cell RNA-seq or mass cytometry can elucidate the primary contributor among cell components. Second, although we validated the poor outcomes in patients with DNMT3A^mut/NPM1^mut/FLT3-ITD^wt/TET2^mut AML with external cohort data, this finding should be interpreted with caution due to the small size of the TET2^mut subgroup. Further studies remain necessary to ensure the reproducibility of our results. Third, all patients in our study received intensive chemotherapy, and the interaction of specific mutation groups with venetoclax and hypomethylating agents was not examined [41, 42]. Furthermore, most patients with FLT3-ITD did not receive FLT3 inhibitor treatment before and after HSCT, potentially affecting their prognosis.

Our findings underscore the necessity of integrating genetic and transcriptomic analyses to comprehensively understand the heterogeneity of DNMT3A-mutated AML. Irrespective of NPM1 and FLT3-ITD mutation status, poorer prognoses were observed among patients with AML and DNMT3A^mut than anticipated under the assigned ELN-2022 risk category. Patients with TET2^mut comprised a unique subgroup within the ELN-2022 favorable DNMT3A^mut/NPM1^mut/FLT3-ITD^wt group, characterized by distinct transcriptomic features and an unfavorable prognosis. This study provides a robust and comprehensive dataset of patient demographic characteristics, clinical outcomes, NGS, and RNA-seq information, offering valuable resources for future AML research.