TCA-controlled histone succinylation identifies IDH3B as a prognostic and therapeutic target in AML

Abstract

Background

Mitochondrial metabolism-driven epigenetic modifications have emerged as crucial regulators for acute myeloid leukemia (AML) progression, linking metabolic activity in leukemic stem cells to epigenetically controlled transcriptional programs that drive oncogenic gene expression.

Results

Here, by integrating proteomic and transcriptomic data, we identified six genes whose expression were able to predict outcome in AML. Among these, IDH3B was highly expressed in leukemic stem cells and associated with poor prognosis. Functional studies revealed that IDH3B deletion in KMT2A-rearranged AML increased global protein succinylation, reduced acetylation, and sensitized cells to the menin–KMT2A inhibitor, both in vitro and in vivo. Mechanistically, loss of IDH3B, by increasing histone succinylation and reducing H3K79 methylation at the MYC promoter, amplified Revumenib-induced transcriptional repression of MYC.

Conclusions

These findings establish IDH3B as a key metabolic–epigenetic regulator in AML and highlight it as a potential synergistic target to enhance menin inhibition therapy.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13148-026-02197-8.

Introduction

Acute myeloid leukemia (AML) is a heterogeneous and aggressive hematologic malignancy that continues to carry a poor prognosis, despite decades of therapeutic development [1]. A growing body of evidence highlights mitochondrial metabolism as a critical vulnerability in AML, particularly within the chemoresistant leukemic stem cells (LSCs) compartment [2, 3]. These cells exhibit a strong reliance on oxidative phosphorylation. This metabolic state is predominantly fueled by mitochondrial fatty acid oxidation and glutamine metabolism, which feed the tricarboxylic acid (TCA) cycle and sustain anabolic and energetic demands [4, 5]. Accordingly, targeting metabolic nodes within this mitochondrial axis has emerged as a promising therapeutic strategy in AML.

Beyond its classical roles in energy production and biosynthesis, the TCA cycle functions as a central metabolic–epigenetic hub. It coordinates cell fate through the production of intermediate metabolites that act as cofactors or substrates for chromatin-modifying enzymes [6, 7]. These include α-ketoglutarate (a cofactor for dioxygenases such as TETs and KDMs), acetyl-CoA (for histone acetylation), and succinyl-CoA, which has recently been recognized as a donor for lysine succinylation, a post-translational modification with transcriptional impact [8–10].

We previously demonstrated that SUCLG1 deficiency in AML patient samples leads to elevated global and histone succinylation, associated with reduced leukemic aggressiveness [11]. SUCLG1, a subunit of succinyl-CoA synthetase, has also been shown to modulate mitochondrial function by coupling succinyl-CoA availability to the succinylation of mitochondrial RNA polymerase POLRMT, thereby integrating mitochondrial metabolism with gene regulation [12]. Furthermore, histone hypersuccinylation resulting from SUCLG1 loss was shown to impair BRD4-dependent oncogenic transcription, particularly MYC target genes, through competition with acetylation and disruption of BRD4–chromatin binding [11].

A parallel epigenetic vulnerability in AML lies in the Menin–KMT2A axis. Menin, encoded by MEN1, is an essential epigenetic scaffold in KMT2A-rearranged leukemias. Menin could bind the N-terminal region of KMT2A fusion proteins and recruits critical chromatin-modifying enzymes such as DOT1L, which catalyzes H3K79 methylation, and indirectly facilitates BRD4-mediated transcription of leukemogenic genes [13, 14]. Accordingly, therapeutic inhibition of the Menin–KMT2A interaction has shown promising efficacy in KMT2A-rearranged leukemias [15, 16]. This study shows that loss of isocitrate dehydrogenase 3 (NAD⁺) beta (IDH3B) slows TCA cycle activity, leading to elevated protein and histone succinylation and rendering AML cells highly sensitive to the Menin–KMT2A inhibitor Revumenib. These findings highlight the therapeutic potential of combining metabolic and epigenetic interventions.

Methods

Machine learning

Least Absolute Shrinkage and Selection Operator (LASSO) regression was applied with cross-validation to identify prognostic variables among candidate genes and the clinical feature. A prognostic model was developed using Partial Least Squares Regression Generalized Linear Model (PLS-RGLM) with the plsRglm R package. The final formula was:

Model discrimination was assessed using Receiver Operating Characteristic (ROC) analysis, and calibration was evaluated with the Hosmer–Lemeshow test. Patients were then stratified into high- and low-risk groups based on risk scores. Kaplan–Meier survival analysis was performed, and the model-based classification was further integrated with the ELN2017 risk stratification system to assess its potential refinement. Specifically, patients classified as ELN adverse and predicted as high-risk by our model were designated as the refined adverse group. Patients classified as ELN favorable and predicted as low-risk were designated as the refined favorable group. All other cases were categorized as intermediate. This refined stratification scheme was then used for subsequent survival analysis.

scRNA-seq data analysis

For scRNA-seq analysis from GSE116256 [17], the raw expression matrix was first imported. Quality control was performed based on the number of detected genes per cell (nFeature_RNA), total transcript count per cell (nCount_RNA), and the proportion of mitochondrial transcripts (percent_MT). Cells were retained if they met the following criteria: 100 ≤ nFeature_RNA ≤ 4000, 100 ≤ nCount_RNA ≤ 20,000, and percent_MT < 10%, resulting in 20,266 cells passing QC. Dimensionality reduction was performed using principal component analysis (PCA) with the RunPCA function in Seurat. A k-nearest neighbors (KNN) graph was constructed using FindNeighbors, followed by unsupervised clustering with the Louvain algorithm via FindClusters. Clustering resolution was optimized using clustree to visualize cluster structures across resolutions. UMAP (RunUMAP) was applied for two-dimensional visualization of cell clusters. Cell clusters were annotated based on canonical marker genes obtained from published literature [17].

To identify the rare putative LSC populations, clusters were analyzed for CD34 expression enrichment, followed by subclustering analysis. LSC identity was evaluated according to the expression pattern of classical stemness markers (CD34 and CD38) together with previously reported LSC gene signatures, including the LSC6 and LSC17 scores, calculated using the AddModuleScore function in Seurat.

Primary samples collection

Bone marrow samples were obtained from patients with AML and from healthy donors at the First Affiliated Hospital of USTC. Mononuclear cells were isolated by Ficoll density gradient centrifugation and stored as dry pellets at − 80 °C until further analysis.

Cell culture

MOLM-13, and THP-1 cells were cultured in RPMI-1640 medium (HyCyte) supplemented with 10% FBS (Sigma) and 1% (v/v) penicillin–streptomycin (Solabio). MV4;11 cells were cultured in IMDM medium (HyCyte) supplemented with 10% FBS (Sigma) and 1% (v/v) penicillin–streptomycin. All cells were maintained at 37 °C in a humidified incubator with 5% CO₂. All cell lines were authenticated by short tandem repeat (STR) analysis.

Lentiviral production and infection

LentiGuide-Puro plasmids (Addgene, 52,963) carrying single-guide RNAs (sgRNAs) targeting IDH3B were co-packaged with Lenti-Cas9-Blast (Addgene, 83,480) using PsPAX2 ((Addgene, 12,260) and pMD2.G (Addgene, 12259) plasmids. The sgRNA target sequences were: non-targeting control: GTTCAGGATCACGTTACCGC, sgIDH3B#1: CGGCATTGAGCGGAGTCCGC, sgIDH3B#2: CTCCCCCTTATACTCCATCG. Lentiviral supernatants were collected at 48 h post-transfection, concentrated using Amicon Ultra filters (Millipore), and incubated with target cells. Transduced cells were selected with 3 μg/mL puromycin (Beyotime) and 10 μg/mL blasticidin (Beyotime), and editing efficiency was assessed by immunoblotting prior to subsequent experiments.

Immunoblotting

Equal number of cells were lysed in 8M urea and sonicated. Cell lysates were then precleared with centrifugation and denatured in SDS loading buffer. Immunoblotting was performed according to standard protocols. Antibodies used in this study were as follows: IDH3B (Abcam, ab247089), H3K79me2 (Abcam, ab3594), H3K79succ (PTM biolabs, PTM-412), H4K5ac (PTM biolabs, PTM-119), HRP-conjugated goat anti-mouse IgG (Proteintech, SA00001-1), HRP-conjugated goat anti-rabbit IgG (Proteintech, SA00001-2).

Immunofluorescence

MOLM-13 cells (5 × 10⁴) were deposited onto glass slides by cytocentrifugation at 500 × g for 5 min. The attached cells were fixed in 4% paraformaldehyde and permeabilized using 0.3% Triton X-100 for 15 min at room temperature. After blocking with 10% BSA for 1 h, samples were incubated with primary antibodies overnight at 4 °C, and then exposed to the corresponding secondary antibodies for 1 h at room temperature. The antibodies used were anti-TOM20 (Proteintech, 11,802–1–AP, 1:200), and anti-pan-Ksucc (PTM Biolab, PTM-419, 1:500), goat anti-mouse Alexa Fluor™ 488 (Invitrogen, A11029, 1:1000), and goat anti-rabbit Alexa Fluor™ 55 (Invitrogen, A21428, 1:1000). counterstained with DAPI (BD Biosciences, 564,907) and mounted. Fluorescent images were captured with confocal microscope (Zeiss, LSM800) and processed with ImageJ.

In vitro drug treatment assays

MOLM-13, MV4;11 or THP-1 cells expressing sgNT, sgIDH3B#1, or sgIDH3B#2 were treated with indicated concentrations of Revumenib (GlpBio) or EPZ004777 for the indicated time. For cell proliferation assays, cells were incubated for the indicated consecutive days before being analyzed using the CCK-8 assay (Dojindo) according to the manufacturer’s instructions. For apoptosis analysis, cells were harvested at indicated time, washed with binding buffer, and stained with Annexin V-APC and 7-AAD (BD Biosciences) prior to analysis by flow cytometry (BD FACS Aria III).

In vivo xenograft transplantation

6–8-week-old female NCG (NOD/ShiLtJGpt-Prkdc^em26Cd52Il2rg^em26Cd22/Gpt) mice were purchased from GemPharmatech and maintained in the animal facility of the First Affiliated Hospital of USTC. To establish cell line-derived xenograft models, 5 × 10⁵ MOLM-13 cells expressing sgNT, sgIDH3B#1, or sgIDH3B#2 were intravenously injected into each mouse. Five days post-injection, Revumenib was orally administered at 50 mg/kg twice daily (BID) for 5 consecutive days, followed by 2 days of rest, for a total of 3 weeks. Mice were sacrificed upon reaching a moribund state, and spleen, liver, and bone marrow were collected and subjected to flow cytometry using anti-human CD45-FITC (eBioscience, 11–0459-42) and anti-mouse CD45-eF506 (BioLegend, 103,138) antibodies or hematoxylin and eosin (H&E) staining to assess leukemia infiltration.

Metabolomics

5 × 10⁶ MOLM-13 cells were suspended in 1 mL of ice-cold 80% (v/v) methanol and incubated at − 80 °C for 4 h to extract metabolites. Following centrifugation at 14,000 × g for 20 min at 4 °C, the supernatants were collected and evaporated to dryness using a vacuum SpeedVac (Beijing JM Technology) at 16 °C to obtain metabolite pellets.

Targeted metabolomic experiment was analyzed by TSQ Quantiva (Thermo, CA). C18 based reverse phase chromatography was utilized with 10mM tributylamine, 15mM acetate in water and 100% methanol as mobile phase A and B respectively. This analysis focused on TCA cycle, glycolysis pathway, pentose phosphate pathway, amino acids and purine metabolism. In this experiment, we used a 25-min gradient from 5 to 90% mobile B. Positive–negative ion switching mode was performed for data acquisition. The resolution for Q1 and Q3 are both 0.7FWHM. The source voltage was 3500v for positive and 2500v for negative ion mode. The source parameters are as follows: spray voltage: 3000v; capillary temperature: 320 °C; heater temperature: 300 °C; sheath gas flow rate: 35; auxiliary gas flow rate: 10. Metabolite identification was based on Tracefinder search with home-built database containing about 300 compounds.

For succinyl-CoA quantification, the LC–MS/MS system was a 6500plus QTrap mass spectrometer (AB SCIEX, USA) coupled with ACQUITY UPLC H-Class system (Waters, USA). An ACQUITY UPLC HSS T3 column (2.1 × 100mm, 1.8μm, Waters) was employed with mobile phase A: water with 5 mM ammonium bicarbonate, and mobile phase B: methanol. Linear gradient is: 0 min, 0% B; 1.5min, 0% B; 6 min, 95% B; 7.4 min, 95%B; 7.5 min, 0%B; and 10 min,0%B. Flow rate was 0.3 mL/min. Column chamber and sample tray were held at 40 °C and 10 °C, respectively. Data were acquired in multiple reaction monitor (MRM) mode for Succinyl-CoA with transitions of 868.1 / 361.1. The ion transitions were optimized using chemical standards. The nebulizer gas (Gas1), heater gas (Gas2), and curtain gas were set at 50, 50, and 35 psi, respectively. The ion spray voltage was 5000 v in positive mode. The optimal probe temperature was determined to be 500 °C, and the column oven temperature was set to 35 °C. The SCIEX OS 1.6 software was applied for metabolite identification and peak integration.

RT–qPCR and RNA-seq

sgCtrl, sgIDH3B#1, and sgIDH3B#2 MOLM-13 cells were treated with 1μM Revumenib for 48 h. Total RNA was extracted with TRIzol reagent and reverse-transcribed into cDNA using SuperScript III reverse transcriptase (Invitrogen, 18,080–051) following the manufacturer’s instructions. Quantitative PCR was performed with SYBR Green Master Mix (YEASEN) on an ABI QuantStudio 5 system. Primers used in this study were:

MYC-F: TCTTCCCCTACCCTCTCAACGA; MYC-R: GCCAGGAGCCTGCCTCTTTT.

HOXA9-F: CCCGGTGCGCTCTCCTTC; HOXA9-R: GTCTCCGCCGCTCTCATTCTC.

MEIS1-F: AAGACACGGGACTCACCATC; MEIS1-R: TGCCCATTCCACTCATAGG.

For RNA-seq, total RNA extracted from sgCtrl, sgIDH3B#1, and sgIDH3B#2 MOLM-13 cells (n = 3 per group) was subjected to sequencing on the NovaSeq 6000 platform using the PE150 mode. Gene set enrichment analysis (GSEA) was performed using the GSEA software (Broad Institute) with the Molecular Signatures Database (MSigDB).

ChIP, qPCR and sequencing

A total of 5 × 10⁷ control, sgIDH3B#1, and sgIDH3B#2 MOLM-13 cells treated with 1 μM Revumenib for 48 h were harvested, cross-linked with 2% formaldehyde for 10 min, and quenched with 0.125 M glycine. Cells were lysed in lysis buffer (50 mM Tris–HCl pH 8.0, 10 mM EDTA, 1% SDS, 1% Triton X-100, 1 × protease inhibitor cocktail, 10 mM NAM, 1 μg/mL TSA) and sonicated. Immunoprecipitation was performed in RIPA buffer (10 mM Tris–HCl pH 7.6, 1 mM EDTA, 0.1% SDS, 1% Triton X-100, 1 × protease inhibitor cocktail, 10 mM NAM, 1 μg/mL TSA) with Protein G magnetic beads (Thermo) and the following antibodies: anti-BRD4 (Bethyl Lab, A301-985), anti-H3K79me2 (Abcam, ab3594), anti-H3K79succ (PTM Biolabs, PTM-412), anti-H3K27ac (Abcam, ab4729), or anti-DOT1L (CST, 77,087), incubated overnight at 4 °C. Beads were washed twice with RIPA buffer, twice with RIPA containing 300 mM NaCl, and twice with RIPA containing 500 mM NaCl. Immunoprecipitates were eluted in elution buffer (10 mM Tris–HCl pH 8.5, 1 mM EDTA, 1% SDS) and reverse cross-linked at 65 °C overnight. Samples were then treated with RNase A and Proteinase K and purified using a PCR purification kit (QIAGEN).

ChIP–qPCR was performed using SYBR Green reagents on an ABI QuantStudio 5 system. The primers used were as follows:

MYC-F: GCTATACACGCACCCCTTT; MYC-R: CTCATCCTTGGTCCCTCAC.

For ChIP–seq analysis, libraries were sequenced on the Illumina NovaSeq 6000 platform using paired-end 150 bp reads (PE150). Raw FASTQ files were quality-filtered with fastp, and the resulting clean reads were mapped to the human reference genome (hg38) using BWA (v0.7.12) [18]. Peak enrichment regions were identified with MACS2 (v2.2.7.1) [19] using a q-value cutoff of 0.05.

Statistics

Statistical analyses were performed using GraphPad Prism (v10) and R software (v4.5.1). Data are presented as mean ± SEM in bar plots and median with interquartile range in boxplots or violin plots. Survival analyses were conducted using Kaplan–Meier methods, and differences between groups were assessed with the log-rank test. Comparisons between two groups were performed using Student’s t-test. For comparisons among more than two groups, one-way ANOVA followed by Dunnett’s post hoc test was applied to compare each group with the control. For experiments involving multiple parameters, two-way ANOVA with Dunnett’s or Turkey multiple comparisons test was used. p value < 0.05 was considered statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Results

Mitochondrial-related gene profiling in acute myeloid leukemia

We analyzed proteomic data from primary bone marrow mononuclear cells of patients with AML (n = 8) and healthy donors (n = 3), and identified 243 downregulated and 198 upregulated proteins in AML groups compared with healthy individuals (|log2FC|> 0.5, p < 0.05) (Fig. 1A and B). We then compared these differentially expressed proteins (DEPs) with mitochondrial metabolism-related genes (MRGs) [20] and identified 39 overlapped genes (Fig. 1C). Among them, 25 genes showed consistent expression changes in the large public transcriptomic dataset GSE13159 [21, 22], in agreement with the proteomic findings (Fig. 1D and E). Specifically, genes that were upregulated at the protein level in primary AML samples were also significantly upregulated in the GSE13159 transcriptomic dataset (|log2FC|> 0.5, p < 0.05), whereas genes that were downregulated at the protein level were likewise significantly downregulated in the transcriptomic dataset. Functional enrichment analysis (GO and KEGG) analysis revealed that these genes were enriched in oxidoreductase complexes, granule membrane, fatty acid metabolism, etc. (Fig. 1F and G).

Fig. 1.

Identification of mitochondrial-associated genes in AML. A. A heatmap displaying differentially expressed proteins (DEPs) in cells from patients with AML (n = 8) and healthy controls (n = 3). B. A volcano plot shows significantly upregulated (n = 198) or downregulated (n = 243) proteins in primary cells from patients with AML compared with healthy donors. The absolute log2FC value > 0.5 and p < 0.05 was used to determine the significance. C. The Venn diagram shows the overlap between differentially expressed proteins (DEPs) in primary AML samples and mitochondrial metabolism–related genes (MRGs) retrieved from MSigDB databases [20]. A total of 39 overlapping genes were identified. D. Relative expression of a subset of overlapping genes identified in (C) that show consistent expression changes (|log2FC|> 0.5, p < 0.05) in the public RNA-seq dataset (GSE13159). Statistical significance for each gene was determined by Student’s t test. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. E. A bar plot showing protein level of the subset of overlapping genes from (D) in AML primary samples. F. Gene ontology (GO) analysis of the subset of overlapping genes from (D) in terms of molecular function (MF), cellular component (CC), biological process (BP).G. Top ten significantly enriched pathways of the genes from (D) using Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis. Adjusted p value was reported

A six-gene signature predicts outcome in AML

To identify the association of these genes with prognosis, we applied a LASSO regression model to the 25 candidate genes together with age in beat AML dataset [23], which serves as the training cohort. The LASSO regression model showed that the binomial deviance was minimized at the optimal parameter (λ = 0.03053483), suggesting the robustness of the selected model at this point (Supplementary Fig. S1A). We then identified six genes, CYP4F3, NME1, DGKA, AOC3, IDH3B, ALDH1A1, and age that are associated with prognosis in AML at the optimal λ value (Supplementary Fig. S1B). Next, we established a prognosis model with PLS-RGLM with these six genes, and established the risk prediction model, with each gene has a coefficient (Fig. 2A). Receiver operating characteristic (ROC) curve analysis based on the maximum Youden index was applied to stratify patients into high- and low-risk groups. The six-gene based algorithm was able to distinguish prognosis, with an area under the curve (AUC) value of 0.781 in beat AML training cohort, and 0.663 in the independent TCGA cohort as validation cohort (Fig. 2B). In addition, the Hosmer–Lemeshow test indicated a satisfactory goodness-of-fit of our model for predicting disease risk (p = 0.908; Supplementary Fig. S1C), and a nomogram was further constructed to estimate the contribution of each gene to prognosis (Supplementary Fig. S1D). Based on the model and the cut-off value of 0.532, patients with AML were stratified into high- and low-risk groups (Supplementary Fig. S1E–F).

Fig. 2.

A six-gene based algorithm predicts prognosis in AML. A. Horizontal bar chart showing the coefficient values of six genes and age that predict AML outcome in the PLS-RGLM model. B. ROC curve shows the sensitivity and 1-specificity value of PLS-RGLM model in predicting risks of outcomes in the training dataset (Beat AML, left) and validation cohort (TCGA, right). C. Kaplan–Meier curves showing the overall survival (OS: defined as time from diagnosis to death from any cause) of high- and low-risk patients as determined by the six-gene–based algorithm in the TCGA cohort (n = 76 high-risk, n = 66 low-risk, left), Beat AML cohort (n = 104 high-risk, n = 64 low-risk, middle), and GSE106291 cohort (n = 155 high-risk, n = 95 low-risk, right). D. Kaplan–Meier curves showing the overall survival of high and low group of patients in FLT3-ITD non-mutated (n = 85 high-risk, n = 49 low-risk, left), CEPBA biallelic non-mutated (n = 104 high-risk, n = 59 low-risk, middle), or TP53 nonmutated (n = 26 high-risk, n = 26 low-risk, right) cases. E. Survival plots of the favorable (n = 49), intermediate (n = 48) and adverse (n = 58) group of patients determined by the ELN 2017 risk classification system. F. Survival plots of the favorable (n = 32), intermediate (n = 81) and adverse (n = 42) group of patients determined by the refined ELN 2017 risk classification with six genes (left). This refined algorithm could stratify intermediate with adverse group of patients (p = 0.00054, right). P value was calculated by log-rank test for C–F

We next checked the efficiency of six-gene based algorithm in predicting prognosis in public AML cohorts. Notably, our model could successfully distinguish prognosis in six large AML datasets, TCGA-LAML, Beat AML, GSE106291, GSE37642-GPL96, GSE37642-GP570, and GSE 12417 (Fig. 2C, Supplementary Fig. S2A). In addition, we compared our risk prediction model with the known cytogenetic characteristics, including FLT3-ITD, TP53 mutation, and CEBPA biallelic status in Beat AML. Consistently, FLT3-ITD, TP53 mutation were associated with poor outcome in this dataset, whereas CEBPA biallelic mutation is associated with favorable prognosis. However, the p value did not reach statistical significance, likely due to the limited number of mutated cases in this dataset (n = 5) (Supplementary Fig. S2B). Our model could further stratify FLT3-ITD, TP53 and CEBPA-nonmutated cases into high- and low- risk of groups, those with high value were associated with shorter overall survival (OS) (Fig. 2D). However, in FLT3-ITD and TP53 mutated situation, our model could not distinguish prognosis (Supplementary Fig. S2C).

The ELN 2017 system, which integrates multiple cytogenetic characteristics, stratifies patients and guides clinical treatment [24]. In the Beat AML cohort, patients were classified into favorable, intermediate, and adverse groups according to the original ELN 2017 criteria (Fig. 2E). However, no significant difference in survival was observed between the adverse- and intermediate-risk groups (Supplementary Fig. S2D). We therefore investigated whether our six-gene based model could refine the ELN 2017 classification system. Specifically, ELN adverse-risk cases and high-risk patients defined by our model were designated as a revised adverse group, whereas the ELN favorable- combined with low-risk groups in our model were assigned into a revised favorable group, with all remaining cases designated as the intermediate group. This refined risk stratification revealed a significantly lower survival rate in the adverse group compared to the intermediate group (Fig. 2F). These findings suggest that the model-based scoring system may also serve as a complementary tool to the ELN 2017 risk stratification system.

IDH3B is highly expressed in leukemic stem cells

Leukemic stem cells (LSCs), which are at the apex of leukemia hierarchy, are the key cells for leukemia initiation and disease relapse [25, 26], we therefore asked whether our genes could be specifically expressed in LSCs. We analyzed the sc-RNA-seq data from GSE116256 and identified 9 cell clusters using canonical markers: hematopoietic stem cells (HSC) or HSC-like, granulocyte-monocyte progenitors (GMP) or GMP-like, proliferating cells (PC) or PC-like, dendritic cells (DC) or DC-like, monocytes or monocyte-like cells, T cells, erythrocytes, plasma cells, and B cells in healthy or AML bone marrows, respectively (Fig. 3A). To identify the LSC population, we examined the expression of CD34, which was predominantly detected in the HSC-like and PC-like clusters in AML samples (Fig. 3B). Based on this observation, we selected the HSPC-like and PC-like cell populations for further subclustering, ultimately resulting in the identification of 9 distinct subclusters (Fig. 3C).

Fig. 3.

IDH3B is highly expressed in leukemic stem cells. A. Umap visualization of single-cell transcriptome from healthy bone marrow (left) and patients with AML (right) from GSE116256. Similar cell types were clustered and annotated as -like cells in AML according to the healthy cells. B. Expression distribution of CD34 in AML and healthy samples. The color intensity indicates CD34 expression levels, ranging from light yellow (low expression) to dark purple (high expression). C. Subclustering of HSPC and PC cell populations. A total of nine clusters (labeled 0–8) were identified, with colors representing distinct clustering results. D. UMAP visualization of CD34 and CD38 expression across cell clusters. The spatial distribution and expression levels of CD34 (left) and CD38 (right) are shown, with color gradients from light yellow to dark purple indicating low to high gene expression. E. Dot plot showing IDH3B expression across different cell populations. The color scale indicates the average expression level, while dot size represents the proportion of cells expressing the gene within each population. IDH3B shows the highest average expression level in the LSC population among all cell types. F. Kaplan–Meier curve showing overall survival of patients with high or low IDH3B expression in the TCGA cohort (n = 85 high-level, n = 57 low-level). G. Gene Ontology (GO) biological process (BP) analysis of IDH3B. The top five positively and negatively enriched pathways are shown. Adjusted p-values are reported

We first examined the classical LSC markers, CD34 and CD38, and found that cluster 2 exhibited high CD34 expression while showing minimal CD38 expression (Fig. 3D). We then applied the AddModuleScore function to calculate LSC 6 scores [27] and LSC17 score [28] across all clusters, which revealed that cluster 2 had the highest score (Supplementary Fig. S3A–B). Taken together, these findings led us to annotate cluster 2 as the LSC population. We analyzed our prognostic genes in the LSC cells and found that IDH3B was upregulated in LSC compared to other cellular clusters (Fig. 3E), indicating a potential biological role of IDH3B in AML pathogenesis. Indeed, we confirmed that high IDH3B expression was associated with poor prognosis in the TCGA AML cohort (Fig. 3F). GO biological process annotation of IDH3B indicated enrichment in mitochondrial translation and related mitochondrial gene expression programs, but negative association with chemical stimulus response and angiogenesis-related processes (Fig. 3G). Consistently, GSEA based on genes correlated with IDH3B expression revealed enrichment of stem cell related programs and oxidative phosphorylation pathways (Supplementary Fig. S3C).

Loss of IDH3B increased global succinylation

The prognostic value of IDH3B and its overexpression in LSC cells prompted us to investigate its potential role in acute myeloid leukemia. We first confirmed that IDH3B is highly expressed in AML primary samples compared to healthy donors with immunoblotting analysis (Fig. 4A). Remarkably, transcriptomic analysis of IDH3B from GSE13159 public datasets revealed that IDH3B is most highly expressed in KMT2A-rearranged (KMT2A-r) cases compared to other cytogenetic subtypes, while showing no significant variation across different KMT2A-r cases (Fig. 4B, Supplementary Fig. S4A), suggesting that IDH3B may play a particularly prominent role in this genetic context. We therefore explored the functional role of IDH3B in KMT2A-r cells. Using CRISPR/Cas9-mediated gene deletion (Fig. 4C), we observed that loss of IDH3B markedly increased global succinylation but led to reduced acetylation in MOLM-13 cells (Fig. 4D). Since IDH3B is a key IDH3 regulator, we reasoned that the alteration in succinylation/acetylation in IDH3B knockout (KO) cells could be attributed to TCA alteration, which is crucial metabolic intermediates controlling lysine modifications [29, 30].

Fig. 4.

Loss of IDH3B increased global succinylation. A. Western blot analysis of AML (n = 7) and healthy samples (n = 3) showing IDH3B expression. β-Actin was used as a loading control. See also supplementary Table S1 for the characteristics of AML samples. B. Relative expression of IDH3B across different AML karyotype subtypes, including t(15;17), inv(16), t(8;21), t(11q23), and healthy cells in public datasets (GSE13159). C. IDH3B was knocked down in MOLM-13 cells using two sgRNAs (sgIDH3B#1 and sgIDH3B#2). Western blot shows the knockdown efficiency. D. Western blot showing the global levels of succinylation (Pan-Ksucc) and acetylation (Pan-Kac) in control and IDH3B knockdown MOLM-13 cells. E. Mass spectrometry analysis of TCA cycle metabolites in control and two IDH3B knockdown MOLM-13 cells, as shown in the figure. Quantification was normalized to the protein concentration of each sample (n = 4 biological replicates). F and G. Relative quantification of NADH/NAD⁺ (F), succinyl-CoA (G, left), and acetyl-CoA (G, right) in IDH3B knockdown and control MOLM-13 cells. H. Immunofluorescence analysis of global Ksucc in mitochondrial and nuclear compartments in control and IDH3B KO MOLM-13 cells. The right panel shows quantification of Ksucc fractions in mitochondrial and nuclear compartments. Scale bar, 20 μm. I. Western blot analysis of global succinylation levels after glucose (Glu) starvation in MOLM-13 control and IDH3B knockdown cells. For A, C–D, and I, representative blots from at least three independent experiments are shown. For panels E–G, data were presented as mean ± SEM from four biological replicates. For panel H, data were obtained from 12 cells from three independent experiments. P values were determined by t test for B, post hoc two-way ANOVA for panel E and H, and post hoc one-way ANOVA for panels F and G. ns, not significant. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001

We first assessed mitochondrial function with TMRM dye, and observed decreased mitochondrial membrane potential in IDH3B KO cells (Supplementary Fig S4B). Metabolomic analysis with control and IDH3B KO cells, further revealed a significant increase in isocitrate and a decrease in NADH/NAD + ratio. These data suggest impaired TCA cycle, particularly at the isocitrate dehydrogenation reaction caused by IDH3B deletion (Fig. 4E-F). Interestingly, no reduction in other metabolites of TCA cycle, including succinyl-CoA, was noted (Fig. 4G, left). In contrast, we detected a significant decrease in acetyl-CoA, which corresponded to reduced level of protein acetylation (Fig. 4G, right).

Since TCA cycle intermediates are continuously diverted into multiple biological pathways [31, 32], disruption of TCA flux via IDH3B deletion may lead to metabolic rerouting. Indeed, glycolytic and related metabolite profiling revealed increased fructose 6-phosphate and decreased fructose 1,6-bisphosphate levels, suggesting impaired phosphofructokinase-1 activity that might result from increased cytosolic citrate accumulation [33] (Supplementary Fig. S4C). Based on these observations, we hypothesized that, beyond absolute succinyl-CoA abundance, its intracellular flux and redistribution across compartments may contribute to regulation of protein succinylation. To test this hyphothesis, we performed immunofluorescence analysis and observed decreased mitochondrial and increased nuclear Ksucc signals in IDH3B KO cells (Fig. 4H). These data suggest a central role of TCA cycle integrity in regulating protein succinylation in AML cells. Consistently, the withdrawal of glucose, the major source fueling the TCA cycle, diminished protein hypersuccinylation in IDH3B KO cells (Fig. 4I).

IDH3B deletion sensitized KMT2A-rearranged AML cells to revumenib treatment

We next investigated the role of IDH3B-restricted succinylation in the pathogenesis of KMT2A-r leukemia. Indeed, we did not observe significant changes in cell proliferation or apoptosis in IDH3B KO MOLM-13 cells (Supplemental Fig. S5A–B). However, treatment with Revumenib, a specific menin–KMT2A inhibitor with promising clinical activity in KMT2A-rearranged AML, led to a more pronounced reduction in cell viability in IDH3B KO compared with control MOLM-13 cells (Fig. 5A, Supplemental Fig. S5C). The same effect was observed in MV4;11, a KMT2A-AFF1 translocated cell line (Fig. 5A, Supplemental Fig. S5D). In addition, Revumenib significantly increased the apoptotic rate in IDH3B-depleted MOLM-13 and MV4;11 cells (Fig. 5B, Supplemental Fig. S5E). Notably, although insensitive to menin inhibition, higher doses of Revumenib caused more cell viability inhibition and apoptosis in THP-1 cells (Supplemental Fig. S5F-I).

Fig. 5.

IDH3B deletion sensitized KMT2A-rearranged AML cells to Revumenib treatment. A. Cell viability assays of MOLM-13 (left) and MV4;11 (right) cells expressing control, sgIDH3B#1, or sgIDH3B#2 treated with 1 μM (for MOLM-13) or 0.5 μM (for MV4;11) Revumenib for the indicated days. Data are presented as mean ± SEM from five biological replicates. Statistical significance was determined by Dunnett’s post hoc test following two-way ANOVA. ****p < 0.0001. See also Fig. S5D–H for the knockout efficiency of IDH3B in MV4; 11 and the Revumenib treatment in THP-1 cells. B. Cell apoptosis assay determining the apoptotic rate of MOLM-13 cells expressing control, sgIDH3B#1, or sgIDH3B#2 after treatment with 1 μM Revumenib (REV) for 6 days, showing representative flow cytometry plots (left) and quantification of apoptotic cells (right). Data are presented as mean ± SEM from three biological replicates. Statistical significance was determined by Dunnett’s post hoc test following two-way ANOVA. ns, not significant. ****p < 0.0001. C. Schematic illustration of mouse xenograft transplantation. A total of 5 × 10⁵ MOLM-13 cells expressing control or sgIDH3B were injected into NCG mice via the tail vein. Revumenib or vehicle was administered orally from 5 days post-transplantation for a total of 3 weeks. Mice were sacrificed at the end of the experiment, and their organs were analyzed for human leukemia cell infiltration. D. Kaplan–Meier curve of control or sgIDH3B mice treated with vehicle or Revumenib (REV). n = 8 mice per group. P value was determined by the log-rank test. ns, not significant. *p < 0.05, **p < 0.01. E. Percentage of human leukemia cells in mouse organs, including bone marrow (left), liver (middle), and spleen (right), in control or sgIDH3B mice treated with vehicle or Revumenib (REV). n = 5 mice per group. Data are presented as mean ± SEM. Statistical significance was determined by Tukey’s post hoc test following one-way ANOVA. ns, not significant. *p < 0.05, **p < 0.01. F. Representative images of H&E staining of bone marrow (left), liver (middle), and spleen (right) in control or sgIDH3B mice treated with vehicle or Revumenib (REV). Scale bar, 50 µm

To further assess the role of IDH3B in vivo, we transplanted mice with control or IDH3B KO MOLM-13 cells and treated them with Revumenib for three weeks (Fig. 5C). Consistent with previous findings [15, 34], treatment with Revumenib significantly prolonged survival in control mice. We found that this survival benefit was further enhanced in the IDH3B KO group (Fig. 5D). At the moribund stage, mice were sacrificed for evaluation of leukemic infiltration in various organs. Flow cytometric and histopathological (H&E) analyses revealed that IDH3B KO mice retained fewer human leukemic cells compared with controls following Revumenib treatment (Fig. 5E-F). Collectively, these data suggest that loss of IDH3B sensitizes KMT2A-r AML cells to Revumenib treatment, highlighting a potential functional interaction between IDH3B-restricted succinylation and menin–KMT2A signaling in leukemogenesis.

IDH3B deletion-mediated histone hypersuccinylation repressed MYC expression

In KMT2A-r leukemic cells, the Menin–KMT2A interaction drives the recruitment of DOT1L, which demethylating histone H3K79, is essential for maintaining the expression of leukemogenic genes [35]. Previous studies reported that DOT1L-controlled H3K79me2 could facilitate H4 acetylation and cooperate with BRD4 to preserve leukemogenic gene expression in KMT2A-r context [13]. We therefore hypothesized that, in IDH3B KO model, increased H3 succinylation, which could prevent BRD4 binding to chromatin [11], acts synergistically with Menin or DOT1L inhibition to repress leukemogenic gene expression. Indeed, we observed an increase in H3K79 succinylation in IDH3B KO cells, however, the global levels of H3K79me2 and H3K27ac showed no significant changes (Fig. 6A). qPCR analysis revealed that while Revumenib significantly repressed MYC, HOXA9 and MEIS1 expression in MOLM-13 wildtype cells, a more pronounced decrease was observed in IDH3B-depleted cells, particularly in MYC expression, which is a well-established BRD4 target gene (Fig. 6B and Supplemental Fig. S6A). These data support a functional collaboration between IDH3B deletion-mediated hypersuccinylation and Menin inhibition in suppressing leukemogenic gene transcription.

Fig. 6.

IDH3B deletion-mediated histone hypersuccinylation repressed MYC expression. A. Immunoblot analysis of H3K79succ, H3K79me2, and H3K27ac in sgCtrl and two sgIDH3B MOLM-13 cells. H3 was used as a loading control. Representative blots are shown from at least three independent experiments. Immunoblotts were performed separately with equal loading of protein for each experiment. B. Relative expression of MYC in sgCtrl and two sgIDH3B MOLM-13 cell lines treated with vehicle or 1 μM Revumenib for 48 h. C. IGV track view showing the genomic coverage of H3K79me2 and H3K79succ at the MYC gene locus in sgCtrl and two sgIDH3B MOLM-13 cell lines treated with 1 μM Revumenib for 48 h. Scale bar, 1kb. D. H3K79me2, H3K79succ, DOT1L and BRD4 ChIP-qPCR analysis of MYC gene locus in two sgCtrl and two sgIDH3B MOLM-13 cell lines treated with 1 μM Revumenib for 48 h. E. Cell apoptosis assay determining the apoptotic rate of MOLM-13 cells expressing control, sgIDH3B#1, or sgIDH3B#2 after treatment with vehicle or 3 μM EPZ004777 (EPZ) for 13 days, showing representative flow cytometry plots (left) and quantification of apoptotic cells (right). For B, D and E, data were presented as mean ± SEM from three biological replicates. Statistical significance was determined by Dunnett’s post hoc test following two-way ANOVA or one-way ANOVA, respectively. ns, not significant. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

In order to visualize the genomic distribution of H3K79succ and H3K79me2 in the regulation of MYC, we performed ChIP-seq analysis in IDH3B knockout cells treated with Revumenib. IGV Track views showed that H3K79me2 occupancy at the MYC gene locus was decreased in KO cells compared to control cells, while H3K79succ occupancy was increased (Fig. 6C). These observations were validated by ChIP-qPCR (Fig. 6D). Furthermore, we also detected decreased BRD4 binding and reduced DOT1L occupancy at the MYC gene locus by ChIP-qPCR (Fig. 6D). Notably, H3K27ac at the MYC locus was not decreased in IDH3B KO cells (Supplemental Fig. S6B), suggesting that the reduced BRD4 occupancy was more likely associated with increased Ksucc rather than changes in H3K27ac. To validate the role of Ksucc in preventing BRD4 binding, MOLM-13 cells were treated with diethyl succinate, which increased Ksucc levels globally, followed by BRD4 ChIP analysis. Consistently, BRD4 occupancy at the MYC locus was significantly reduced following diethyl succinate treatment (Supplemental Fig. S6C–D). Interestingly, similar to the effect of Revumenib, inhibition of DOT1L using EPZ004777 also led to a stronger induction of apoptosis in IDH3B KO cells compared with controls (Fig. 6E). Finally, to investigate the link between IDH3B deletion-mediated histone hypersuccinylation and the LSCs signature, we performed RNA-seq analysis and found that genes associated with LSCs were downregulated in IDH3B KO cells (Supplemental Fig. S6E).

Discussion

We conducted an integrative multi-omic analysis of primary cells from patients with AML or healthy donors and public datasets and identified 25 genes that are associated with mitochondrial metabolism and epigenetic regulation in AML. However, given the relatively small size of the proteomic cohort (n = 8 AML vs. n = 3 controls), these findings should be interpreted with caution. We then applied LASSO regression for feature selection to reduce dimensionality and minimize overfitting, while PLS-RGLM was subsequently used to construct the final prognostic model to account for potential biological and statistical correlations among selected genes. Through these analyses, we identified six genes whose expression was associated with prognostic stratification across several AML cohorts. However, the predictive performance of the model in the validation cohort (TCGA, AUC = 0.66) was moderate. These may reflect the biological heterogeneity of AML as well as differences in patient composition, clinical characteristics and treatment strategies among cohorts. Notably, the six-gene based algorithm failed to stratify patients with FLT3-ITD or TP53-mutations. These AML subgroups are characterized by highly aggressive biological behavior and complex molecular alterations, suggesting that their clinical behavior is influenced by multiple factors rather than mitochondrial metabolism–related features alone. Our six-gene model improves the discriminatory power of the ELN 2017 risk stratification. However, its robustness remains to be validated in independent AML cohorts due to the lack of publicly available datasets with sufficiently comprehensive molecular and clinical annotations. Further optimization using larger multicenter cohorts and integration with additional genomic or clinical variables are warranted to improve predictive performance in future studies.

IDH3, composed of subunits IDH3A, IDH3B, and IDH3G, encodes the NAD⁺-dependent mitochondrial isocitrate dehydrogenase complex responsible for the oxidative decarboxylation of isocitrate to α-ketoglutarate within the TCA cycle. IDH3B is not directly catalytic but modulates IDH3 activity allosterically by stabilizing the heterotetrameric complex [36]. In contrast to the frequently mutated IDH1 and IDH2 isoforms in AML, mutations in IDH3 subunit genes are rare in cancer [37–40]. Nonetheless, IDH3B has recently been proposed as a biomarker in various solid tumors [41–43], although its role in AML and chromatin regulation remains largely unexplored.

In this work, we noted that the IDH3B is, among the mitochondrial genes related to succinylation, the most highly upregulated in leukemic stem cells, and that IDH3B expression is significantly dysregulated in this AML contest. The role of IDH3B suggests that targeting this subunit may selectively impair leukemic metabolism while sparing non-transformed cells. This is consistent with observations in IDH3B knockout mice, which exhibited metabolic disruptions only in highly active tissues (e.g., testis), but not in other organs such as the retina, indicating limited systemic toxicity [44].

Loss of IDH3B leads to an accumulation of isocitrate without variation of downstream metabolites in according with results reported in IDH3B KO mice [44]. However, the significant decrease of NADH/NAD + in TCA cycle suggest a defect in IDH3 reaction guiding a significant slowdown in TCA cycle progression. The downstream metabolites were largely unchanged despite impaired TCA activity, suggesting that steady-state measurements may not fully capture dynamic metabolic routing. Indeed, TCA lag caused by IDH3B deletion recapitulates the protein and histone hypersuccinylation reported upon disrupting the TCA cycle by SUCLG1 knockout [11]. Moreover, immunofluorescence revealed increased extramitochondrial succinylation in IDH3B KO cells, supporting a model in which TCA cycle dysfunction affects protein succinylation via altered metabolic flux and compartmental redistribution. In fact, the withdrawal of glucose, a major nutrient that fuels mitochondrial TCA cycle, abolished the hypersuccinylation in IDH3B KO AML cell lines. However, since glucose deprivation might induce broader alterations beyond IDH3B-associated pathways, the observed phenotypes may partially involve secondary metabolic effects. We anticipate that future compartment-resolved metabolite tracing will be essential to validate the role of succinyl-CoA flux in regulating lysine succinylation.

Histone acetylation, particularly at H3K9, H3K27 and H4K5 residues, supports oncogenic transcription by recruiting epigenetic readers such as BRD4, thereby promoting leukemic progression [45–47]. In contrast, histone succinylation antagonizes this pathway, by weakening the interaction between the BRD4 bromodomain and chromatin, which disrupts BRD4-mediated leukemogenic transcriptional activity and restores homeostatic BRD4-dependent gene regulatory circuits [11]. This type of competitive antagonism between histone acylations has been previously proposed as a key mechanism in metabolic-epigenetic crosstalk, especially in relation to lysine site competition and acyl-CoA availability [11].

Furthermore, we here found this metabolic-epigenetic shift to succinylation sensitized AML cells to Menin–KMT2A inhibitor. During Revumenib treatment, in vivo and in vitro expansion of KO cell lines were arrested, together with increased apoptosis compared to control cells. Prior preclinical and clinical studies demonstrated anti-leukemia activity of Revumenib at relatively low plasma concentrations (1 ~ 3 μM) in vivo [15], supporting the concentration selected for MOLM-13 and MV4;11 experiments. In contrast, THP-1 cells showed measurable responses only at higher concentrations (50 μM), indicating potential differences in drug sensitivity between cell models and that the response in THP-1 cells may have limited biological and translational relevance.

In KMT2A-rearranged leukemic cells, the Menin–KMT2A interaction drives the recruitment of transcriptional cofactors, primarily the histone methyltransferase DOT1L, is essential for maintaining the expression of leukemogenic genes [35]. In the IDH3B KO cell line, residual succinylation could to occupy sites normally associated with histone methylation on oncogene MYC. This leads to a synergistic effect characterized by the mis-recruitment of DOT1L and BRD4 to the MYC gene locus. Mechanistically, DOT1L and BRD4 have been shown to functionally cooperate, with DOT1L-dependent dimethylation of histone H3 on lysine 79 (H3K79me2) on HOXA9 and MEIS1 genes promoting subsequent acetylation of histone H4 [13]. Notably, IDH3B loss does not affect global H3K79me2 levels, supporting a context-dependent regulation of MYC expression through the balance between H3K79 succinylation and DOT1L-mediated methylation at the MYC locus. In addition, histone acetylation at the MYC locus was not significantly altered in IDH3B KO cells, supporting a more prominent role of succinylation in disrupting BRD4 interaction with chromatin. IDH3B deficiency enhanced histone succinylation, which in turn impaired DOTL1 and BRD4 chromatin binding and potentiated the anti-leukemic effects of Revumenib both in vitro and in vivo.

Supporting its clinical relevance, in a large AML patient cohort from the TCGA dataset, high IDH3B expression is associated with poor prognosis and significantly reduced overall survival compared to patients with low IDH3B expression. While no pharmacological inhibitors of IDH3B are currently available, its non-catalytic role may make it amenable to selective disruption via protein–protein interaction inhibitors or targeted protein degradation, potentially avoiding the systemic toxicity associated with conventional metabolic enzyme inhibition. It is noteworthy that the conclusions of the current study are based on genetic approaches, which may not fully recapitulate the effects of pharmacological inhibition. In addition, the limited cell lines used in this study did not comprehensively reflect the metabolic and molecular heterogeneity of AML. Further development of selective IDH3B inhibitors and validation in broader AML models and primary samples will be necessary before the therapeutic intervention.

Conclusions

In conclusion, these findings uncover a previously unrecognized mitochondria-to-epigenome regulatory axis in AML, position IDH3B as a modulator of chromatin state and disease aggressiveness, and support the rationale for combinatorial targeting of IDH3B and menin–KMT2A as a therapeutic strategy in KMT2A-rearranged AML.

Author contributions

M.H. and K.Y performed the experiments; H.D. analyzed the data; N.Z., J.M. and A.H. collected the primary samples; Y.W. provided the scientific counseling. D.I. and M.G. designed the project and wrote the manuscript. X.Z., D.I., and M.G. supervised the project.

Funding

This research is supported in part by research grants from National Natural Science Foundation of China (82200197, 12201601, U23A20453, 82270223 and 82170209), Anhui Provincial Department of Education Scientific Research Project (2023AH010079), Anhui Provincial Natural Science Foundation (2308085J09), USTC Research Funds of the Double First-Class Initiative (YD9110002047), International Cooperation Projects in Anhui Province (2023h11020005). D.I. acknowledges support from PRIN 2022 PNRR (P20225KJ5L), MUR Italy, NextGenerationEU. Y.W. receives support from the NHC Key Laboratory of Thrombosis and Hemostasis, the First Affiliated Hospital of Soochow University (KJS2420), and the Gusu Talent Program (GSWS2023083).

Data availability

Public RNA-seq data was from TCGA (n = 142), Beat AML (n = 168), GSE106291 (n = 250), GSE37642-GPL96 (n = 417), GSE37642-GP570 (n = 136), GSE 12417 (n = 162). Mitochondrial-related genes were retrieved from MSigDB databases published in previous study. Raw RNA-seq data of control and IDH3B knockout MOLM-13 cell line have been deposited to GEO (GSE330782). Proteomic on bone marrow mononuclear cells from healthy donors (n = 3) and patients with AML (n = 8) was performed as previously described and deposited to iProX (PXD069471, PXD054053). Raw ChIP-seq data have been deposited to SRA (PRJNA1365942). All analytic code and generated materials are available from the corresponding author upon request.

Declarations

Ethics approval and consent to participate

All animal and human study protocols were approved by the Ethics Committee of the First Affiliated Hospital of USTC, with approval numbers 2023-N(A)-55 and 2024KY-184. Written informed consent was obtained from all human participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.