Targeting SMC3 deacetylation synergizes with XPO1 inhibition to reprogram the epigenetic landscape and suppress NPM1-mutated acute myeloid leukemia Authors: Jianfeng Fu, Huilin Xu, Xinying Zhao, Mengge Gao, Yonglu Tian, Yingjun Chang, Xiaofeng Zheng Published: 2026 ABSTRACT Acute myeloid leukemia (AML) is a prevalent hematologic malignancy. Mutations in Nucleophosmin (NPM1c) is the most frequent genetic alterations in AML. However, the epigenetic regulatory mechanisms of this AML subtype remain unclear. The cohesin complex, regulated by the acetyltransferase ESCO2, orchestrates chromatin organization and regulates gene expression. Here, we show that loss of ESCO2 promotes the progression of NPM1-mutated AML by destabilizing cohesin and the NuRD complex on chromatin, thereby enhancing the expression of genes associated with leukemia self-renewal capacity. Pharmacological HDAC8 inhibition restores SMC3 binding and induces differentiation and apoptosis in NPM1-mutated AML cells. Simultaneously targeting HDAC8 and nuclear exporter XPO1 reverses aberrant epigenetic landscape and represses self-renewal gene expression induced by ESCO2 deficiency. The combined treatment effectively eliminates NPM1-mutated AML cell lines and primary human AML cells in vitro and in vivo. This study reveals an ESCO2 deficiency-induced aberrant epigenetic landscape via SMC3 hypoacetylation and identifies a potential therapeutic strategy for NPM1-mutated AML. Introduction AML is characterized by the clonal expansion of hematopoietic stem and progenitor cells, leading to impaired hematopoiesis and bone marrow failure. NPM1 is one of the most frequently mutated genes, occurring in approximately 30% of adult AML patients1. Mutant form of NPM1 (NPM1c) loses its nucleolar localization and translocates to the cytoplasm through interaction with the nuclear exporter XPO12. Selective XPO1 inhibitor KPT-8602 (eltanexor) exhibits anti-leukemic potential3,4, but therapeutic efficacy with KPT-8602 alone is limited (NCT02649790) due to adverse events5. Chromatin-binding NPM1c directly upregulates transcription of leukemia-driving genes, including HOXA/B and MEIS6–8, inhibits caspase-6 and −8 activities9, and stabilizes the oncoprotein c-Myc10. However, mutations in NPM1 have not been detected in individuals with clonal hemopoiesis11–14. Furthermore, NPM1c alone can induce leukemia in vivo after a prolonged latency period15. Thus, the mechanism underlying NPM1c-driven leukemogenesis remains to be fully elucidated. Cohesin is a multi-subunit, ring-like structure protein complex that is crucial for sister chromatid cohesion and the organization of genome architecture16. Acetylation of the cohesin subunit SMC3 at K105/106, mediated by ESCO1/2, is essential for cohesion establishment17,18. This acetylation alters SMC3 ATPase active site, reinforcing cohesin’s DNA tethering while blocking Wapl-mediated ring dissociation19,20. While both ESCO1 and ESCO2 have acetyltransferase activity, they exhibit spatiotemporally distinct patterns in modulating SMC3 acetylation21,22. ESCO2 deficiency in humans leads to Roberts syndrome (RBS), which is characterized by craniofacial, limb, and neurological system developmental abnormalities23. Loss of cohesion in RBS patients cannot be rescued by overexpressing ESCO124, suggesting that the two acetyltransferases are not fully redundant. Additionally, HDAC8 acts as the deacetylase for SMC3, playing a pivotal role in cohesin-dependent functions25,26. Whole-genome sequencing has revealed that 8–13% of AML patients harbor somatic mutations in cohesin complex genes27–29. Loss-of-function mutations in cohesin genes lead to genome-wide chromatin remodeling, thereby contributing to the development of myeloproliferative neoplasms in vivo30–32. But the role of acetylation-mediated cohesin stability in AML progression and its clinical implications are needed to be investigated. In this work, we demonstrate that ESCO2 deficiency is associated with poor prognosis in AML patients, specifically with NPM1 mutation. Moreover, we find that ESCO2 depletion enhances the development of AML with NPM1c mutation. Mechanistically, ESCO2 stabilizes the binding of cohesin at loci associated with AML self-renewal genes, while cohesin suppresses the expression of these genes via its interaction with the NuRD complex. ESCO2 deficiency leads to cohesin destabilization and transcription activation of self-renewal genes. We further show that treatment with an HDAC8 inhibitor (PCI-34051) can counteract the tumor-promoting effects induced by ESCO2 deficiency. Finally, the combined inhibition of HDAC8 and XPO1 inhibitors exhibits potent anti-AML activity in both in vitro experiments and xenograft mouse models. Collectively, our findings highlight the biologic effects of ESCO2-mediated SMC3 acetylation in NPM1-mutated AML and provide a promising potential therapeutic strategy by simultaneously targeting HDAC8 and XPO1. Results ESCO2 deficiency impairs cohesin chromatin retention to promote the progression of NPM1-mutated AML Firstly, we assessed the expression level of ESCO2 in relation to survival outcomes and found that in TCGA cohort lower ESCO2 expression levels were particularly associated with poor overall survival in AML patients harboring NPM1 mutations (HR = 0.56, 95% CI 0.33–0.95 log-rank p = 0.029), but not in those with wild-type NPM1 (Fig. 1A). To validate this observation, we further analyzed an independent AML cohort from BeatAML2. Consistent with the findings from TCGA, NPM1-mutated patients with low ESCO2 expression in the BeatAML2 cohort also exhibited significantly poorer survival outcomes (Supplementary Fig. 1A). This clinical observation suggested a context-dependent interaction between ESCO2/cohesin function and NPM1c, promoting us to investigate the mechanistic basis of this connection.Fig. 1ESCO2 deficiency promotes AML progression in NPM1-mutated AML.A Kaplan–Meier overall survival analysis of all AML patients stratified by NPM1 mutation status based on their ESCO2 expression levels from TCGA data (expression range of the probe: 3–397; cutoff value: 78; n = 734, 139 and 318). B ESCO2 knockdown in OCI-AML3 cells by the CRISPR-Cas9 system was confirmed by western blotting. Representative western blot images from three independent experiments with similar results are shown. C Experimental scheme for investigating the effect of ESCO2 knockdown (sgESCO2) on OCI-AML3-induced leukemogenesis in lethally irradiated recipient mice. Created in BioRender. Fu, J. (2026) https://BioRender.com/7tf5snh. D Kaplan–Meier survival curves comparing OCI-AML3 control mice (n = 8) and sgESCO2 mice (n = 8). P-values calculated using the log-rank test. E Representative images (left) and weight quantification (right) of spleen harvested from transplanted mice. Data are presented as mean ± SD (n = 8). P-values determined using an unpaired, two-tailed Student t-test (P < 0.0001). F Colony numbers of control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P = 0.0004, P < 0.0001). G Cell proliferation assessed by CCK-8 assay (n = 3). H Cell cycle analysis of control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P < 0.0001). I SMC3 acetylation level in OCI-AML3 cells was confirmed by western blotting. Representative western blot images from three independent experiments with similar results are shown. J SMC3 CUT&Tag data profiles in control and sgESCO2#1 OCI-AML3 cells normalized by RPGC, n = 3 biologically independent experiments. K Overexpressing HA-SMC3 (WT, K2Q or K2R mutant) in ESCO2-depleted OCI-AML3 cells was confirmed by western blotting. Representative western blot images from three independent experiments with similar results are shown. L Colony formation assays assessing the clonogenic potential of OCI-AML3 control cells, sgESCO2#1 cells, and sgESCO2#1 cells stably overexpressing SMC3-WT or SMC3-K2Q (n = 3). M Kaplan–Meier survival analysis of AML patients with NPM1 mutations, based on their cohesin subunit gene expression levels from TCGA data (expression range of the probe: 889-3568; cutoff value: 2227.4, n = 139). Statistical analysis in panels F–H, L was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD. Source data are provided as a Source Data file. Survival analyses in panels A, M were conducted using the Kaplan–Meier Plotter database, with Cox proportional hazards regression for association testing and log-rank test for significance. ESCO2 deficiency promotes AML progression in NPM1-mutated AML. A Kaplan–Meier overall survival analysis of all AML patients stratified by NPM1 mutation status based on their ESCO2 expression levels from TCGA data (expression range of the probe: 3–397; cutoff value: 78; n = 734, 139 and 318). B ESCO2 knockdown in OCI-AML3 cells by the CRISPR-Cas9 system was confirmed by western blotting. Representative western blot images from three independent experiments with similar results are shown. C Experimental scheme for investigating the effect of ESCO2 knockdown (sgESCO2) on OCI-AML3-induced leukemogenesis in lethally irradiated recipient mice. Created in BioRender. Fu, J. (2026) https://BioRender.com/7tf5snh. D Kaplan–Meier survival curves comparing OCI-AML3 control mice (n = 8) and sgESCO2 mice (n = 8). P-values calculated using the log-rank test. E Representative images (left) and weight quantification (right) of spleen harvested from transplanted mice. Data are presented as mean ± SD (n = 8). P-values determined using an unpaired, two-tailed Student t-test (P < 0.0001). F Colony numbers of control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P = 0.0004, P < 0.0001). G Cell proliferation assessed by CCK-8 assay (n = 3). H Cell cycle analysis of control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P < 0.0001). I SMC3 acetylation level in OCI-AML3 cells was confirmed by western blotting. Representative western blot images from three independent experiments with similar results are shown. J SMC3 CUT&Tag data profiles in control and sgESCO2#1 OCI-AML3 cells normalized by RPGC, n = 3 biologically independent experiments. K Overexpressing HA-SMC3 (WT, K2Q or K2R mutant) in ESCO2-depleted OCI-AML3 cells was confirmed by western blotting. Representative western blot images from three independent experiments with similar results are shown. L Colony formation assays assessing the clonogenic potential of OCI-AML3 control cells, sgESCO2#1 cells, and sgESCO2#1 cells stably overexpressing SMC3-WT or SMC3-K2Q (n = 3). M Kaplan–Meier survival analysis of AML patients with NPM1 mutations, based on their cohesin subunit gene expression levels from TCGA data (expression range of the probe: 889-3568; cutoff value: 2227.4, n = 139). Statistical analysis in panels F–H, L was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD. Source data are provided as a Source Data file. Survival analyses in panels A, M were conducted using the Kaplan–Meier Plotter database, with Cox proportional hazards regression for association testing and log-rank test for significance. To explore the effect of ESCO2 deficiency on the progression of NPM1-mutated AML in vivo, we used CRISPR–Cas9 to knock down ESCO2 in OCI-AML3 and OCI-AML2 cells (Fig. 1B, Supplementary Fig. 1B). Subsequently, we injected sgControl or sgESCO2 OCI-AML3 cells which carrying the canonical NPM1c mutant, into NCG mice (NOD/ShiLtJGpt-Prkdcem26Cd52Il2rgem26Cd22) to generate xenograft models (Fig. 1C). This mouse strain is severely immunodeficient, lacking functional T, B, and NK cells, and serves as a leading model for studying AML proliferation and identifying leukemic stem and progenitor cells (LSPCs)33,34. We observed that mice bearing ESCO2-depleted OCI-AML3 cells exhibited reduced survival compared to controls (Fig. 1D), along with spleen enlargement (Fig. 1E) resulting from the accumulation of leukemic cells in the spleen. Moreover, ESCO2-depleted OCI-AML3 cells showed enhanced colony-forming capacity (Fig. 1F) and increased cell proliferation (Fig. 1G, H), which is consistent with the proliferation advantage driven by SMC3 haploinsufficiency in hematopoietic stem and progenitor cells30–32. These findings reveal a tumor-suppressive role of ESCO2 in NPM1-mutated AML, and ESCO2 deficiency substantially accelerates disease progression. To unravel the potential mechanism by which ESCO2 deficiency promotes NPM1-mutated AML progression, we initially evaluated the acetylation of SMC3 at lysine 105 and 106 (K105/106) and found that ESCO2 deficiency resulted in decreased SMC3 acetylation in OCI-AML3 cells (Fig. 1I). A similar reduction in SMC3 K105/106 acetylation was also observed in OCI-AML2 cells following ESCO2 depletion (Supplementary Fig. 1C). Moreover, we found that wild-type OCI-AML3 cells expressed lower levels of ESCO2 compared to OCI-AML2 cells, which was accompanied by correspondingly lower basal levels of SMC3 acetylation (Supplementary Fig. 1D). Given the ESCO2’s function in stabilizing the cohesin complex on chromatin through SMC3 acetylation17,18, we conducted CUT&Tag to determine whether ESCO2 depletion impacts the genome-wide binding pattern of SMC3. The result showed that ESCO2 loss significantly reduced genome-wide SMC3 binding on chromatin (Fig. 1J). To further validate that ESCO2 depletion promotes AML progression by reducing the acetylation levels of the cohesin subunit SMC3, we stably overexpressed wild-type SMC3, SMC3 K105/106Q mutant (SMC3-K2Q, which mimics the hyperacetylated state of SMC3), and SMC3 K105/106 R mutant (SMC3-K2R, which mimics the hypoacetylated state of SMC3) in ESCO2-depleted OCI-AML3 cells (Fig. 1K). Notably, the SMC3-K2Q mutant represents a charge mimic of acetylation rather than a direct acetylation modification (Supplementary Fig. 1E). Knockdown of ESCO2 promoted colony formation, and overexpression of the SMC3-K2Q mutant, but not wild-type SMC3, was able to reduce clonogenic capacity in ESCO2-depleted cells (Fig. 1L). Similar to the relation between ESCO2 expression levels and the survival of NPM1-mutated AML patients, we found that reduced expression of cohesin genes (SMC1A, SMC3, RAD21, STAG1 and STAG2) was associated with decreased overall survival in these patients (HR = 0.69, 95% CI 0.48–0.99, log-rank p = 0.041) (Fig. 1M). These findings implicate that cohesin dysfunction induced by ESCO2 deficiency plays an oncogenic role in the progression of NPM1-mutated AML. ESCO2 deficiency induces the expression of self-renewal genes in NPM1-mutated AML cells To explore the mechanism by which ESCO2 deficiency promotes NPM1-mutated AML, we performed RNA-seq on OCI-AML3 (with NPM1 mutation) and OCI-AML2 (without NPM1 mutation) cells. GO pathway analysis revealed that several oncogenic pathways were upregulated in ESCO2-deficient OCI-AML3 cells (Fig. 2A), which aligns with the accelerated cell proliferation observed in ESCO2-deficient AML cells (Fig. 1H, I). AML harbors a hierarchically organized population of leukemic stem and progenitor cells, which are characterized by their unique functional properties of self-renewal and tumor-initiating capacity. These cells recapitulate the heterogeneous leukemic landscape and serve as a critical reservoir for disease relapse and therapeutic resistance35–38. Based on previously reported transcriptomic analyses of 136 LSPC-specific genes involved in AML self-renewal maintenance, oncogenesis or both39–43, we identified a subset of 93 critical genes (such as HOXA/B genes) that were upregulated in ESCO2-depleted OCI-AML3 cells (Fig. 2B, C, Supplementary Fig. 2A). To assess the clinical significance of these self-renewal genes associated with ESCO2 deficiency, we evaluated the correlation with survival in cohort of NPM1-mutated AML patients. Our findings revealed that high expression levels of these 93 genes were linked to poor survival outcomes (Fig. 2D). However, we did not observe an increase in the expression of these self-renewal genes in ESCO2-depleted OCI-AML2 cells (Supplementary Fig. 2B). These results suggest that ESCO2 deficiency selectively activates leukemia self-renewal genes in NPM1-mutated AML, which aligns with the reduced survival rates in AML patients (Fig. 1B). Next, we studied the transcription profiles of control OCI-AML3 cells, ESCO2-depleted OCI-AML3 cells, and normal hematopoietic stem and progenitor cells (HSPCs), as well as their precursor/mature populations. We performed two-dimensional principal component analysis (PCA) and hierarchical clustering using the reported 500 most differentially expressed genes44. The result showed that OCI-AML3 was positioned between granulocyte-macrophage progenitor (GM-Pre) and granulocyte-macrophage mature (GM-mature) (Fig. 2E), which is consist with the transcription features of CD34− leukemia cells44. Moreover, the global transcription profiles of ESCO2-depleted OCI-AML3 cells were distinctly segregated from those of control cells and were found to be closer to the transcription profiles of the GM-Pre and GMP populations (Supplementary Fig. 3A, B). Single-cell transcriptomic profiling of hematopoietic lineages derived from normal donors (GSE4251945) demonstrated a dynamic expression pattern of ESCO2, characterized by a gradual upregulation across successive stages of cellular differentiation (Supplementary Fig. 3C). These findings revealed a reciprocal regulatory relationship between ESCO2 and hematopoietic stem cell differentiation, and ESCO2 depletion promotes self-renewal potential in NPM1-mutated AML.Fig. 2Loss of ESCO2 induces the expression of self-renewal genes in NPM1-mutated AML cells.A Bubble diagram showing the enrichment of Gene Ontology (GO) pathways by the transcripts that were upregulated following ESCO2 deficiency in OCI-AML3 cells. P-values were adjusted for multiple comparisons using the Benjamini–Hochberg method, with terms showing an adjusted P-value < 0.05 considered statistically significant. The results were visualized in bubble plots, where dot size corresponds to gene counts and color represents the adjusted P-value, n = 3 biologically independent experiments. B Heatmap illustrating the expression levels of the 93 ESCO2 deficiency-induced self-renewal genes in OCI-AML3 cells. Gene expression is normalized using a z-score. C Quantitative RT-PCR analysis of selected self-renewal maintenance genes in control and sgESCO2#1/2 OCI-AML3 cells (n = 3). D Kaplan–Meier survival analysis of NPM1-mutated patients based on self-renewal gene expression levels from TCGA data. P-value calculated using the log-rank test (n = 139). E Schematic diagram of the normal hemopoietic hierarchy (left, Created in BioRender. Fu, J. (2026) https://BioRender.com/hfgxgtb.) and Principal component analysis (PCA) plots (right) showing the positions of different normal hematopoietic cell populations, including HSC (n = 3), GMP (n = 3), CMP (n = 3), GM-Pre (n = 5), GM-Mat (n = 4), as well as control and sgESCO2#1/2 OCI-AML3 cells (n = 2), based on the expression levels of 500 selected genes. The percentage of variance represented by each principal component (PC) is shown. F Percentage of CD34−CD38high cell population in control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P < 0.0001). G Quantitative RT-PCR of CD38 expression in control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P < 0.0001, P = 0.0004). H Flow plot showing CD38 expression in control and sgESCO2#1/2 OCI-AML3 cells. Representative results from three independent experiments with similar results are shown. I True limiting dilution assay measuring the frequency of LICs in control and ESCO2-depleted cells. The frequencies of LICs were calculated by Poisson statistics using ELDA software. J Flow plot showing CD11b expression in control and sgESCO2#1/2 OCI-AML3 cells with or without ATRA treatment (100 nM, 72 h). Representative results from three independent experiments with similar results are shown. Statistical analysis in panels C, F, G was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD. Source data are provided as a Source Data file. Loss of ESCO2 induces the expression of self-renewal genes in NPM1-mutated AML cells. A Bubble diagram showing the enrichment of Gene Ontology (GO) pathways by the transcripts that were upregulated following ESCO2 deficiency in OCI-AML3 cells. P-values were adjusted for multiple comparisons using the Benjamini–Hochberg method, with terms showing an adjusted P-value < 0.05 considered statistically significant. The results were visualized in bubble plots, where dot size corresponds to gene counts and color represents the adjusted P-value, n = 3 biologically independent experiments. B Heatmap illustrating the expression levels of the 93 ESCO2 deficiency-induced self-renewal genes in OCI-AML3 cells. Gene expression is normalized using a z-score. C Quantitative RT-PCR analysis of selected self-renewal maintenance genes in control and sgESCO2#1/2 OCI-AML3 cells (n = 3). D Kaplan–Meier survival analysis of NPM1-mutated patients based on self-renewal gene expression levels from TCGA data. P-value calculated using the log-rank test (n = 139). E Schematic diagram of the normal hemopoietic hierarchy (left, Created in BioRender. Fu, J. (2026) https://BioRender.com/hfgxgtb.) and Principal component analysis (PCA) plots (right) showing the positions of different normal hematopoietic cell populations, including HSC (n = 3), GMP (n = 3), CMP (n = 3), GM-Pre (n = 5), GM-Mat (n = 4), as well as control and sgESCO2#1/2 OCI-AML3 cells (n = 2), based on the expression levels of 500 selected genes. The percentage of variance represented by each principal component (PC) is shown. F Percentage of CD34−CD38high cell population in control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P < 0.0001). G Quantitative RT-PCR of CD38 expression in control and sgESCO2#1/2 OCI-AML3 cells (n = 3) (P < 0.0001, P = 0.0004). H Flow plot showing CD38 expression in control and sgESCO2#1/2 OCI-AML3 cells. Representative results from three independent experiments with similar results are shown. I True limiting dilution assay measuring the frequency of LICs in control and ESCO2-depleted cells. The frequencies of LICs were calculated by Poisson statistics using ELDA software. J Flow plot showing CD11b expression in control and sgESCO2#1/2 OCI-AML3 cells with or without ATRA treatment (100 nM, 72 h). Representative results from three independent experiments with similar results are shown. Statistical analysis in panels C, F, G was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD. Source data are provided as a Source Data file. Given the low expression of the leukemia marker CD34 in NPM1-mutated AML cells44, and the fact that the fraction of CD34-CD38high NPM1-mutated cells possesses the capability of leukemia initiation in immunodeficient mice46,47, we quantified the proportion of CD34-CD38high cell subpopulation using FACS. The result revealed that this subpopulation increased following ESCO2 deficiency (Fig. 2F). Additionally, we found that ESCO2 depletion induced the transcription of the CD38 gene and increased the level of CD38 protein on the cell surface (Fig. 2G, H). To verify the enhanced self-renewal conferred by ESCO2 loss, we isolated the CD34−CD38high subpopulation from control and ESCO2-depleted OCI-AML3 cells and subjected them to a true limiting dilution assay. The frequency of leukemia-initiating cells (LICs) was significantly higher in the ESCO2-deficient subset (Fig. 2I, Supplementary Fig. 3D). CD11b is a ubiquitous marker present on the surfaces of mature macrophages and granulocytes. Prior work demonstrated that ATRA drives differentiation in NPM1-mutated cells48,49, we found that ATRA treatment increased CD11b in control cells. In contrast, ESCO2 depletion markedly blunted this response, conferring pronounced resistance to ATRA-induced differentiation (Fig. 2J). Overexpression of ESCO2 led to elevated CD11b levels in OCI-AML3 control cells, indicating that sustained ESCO2 expression may play a role in promoting myeloid differentiation (Supplementary Fig. 3E). Taken together, these findings suggest that ESCO2 deficiency promotes the progression of NPM1-mutated AML and increases the proportion of self-renewing cell population through the upregulation of self-renewal genes. Cohesin destabilization induced by ESCO2 depletion increases H3K27ac level at the loci of leukemia self-renewal genes We detected the binding of SMC3 at the loci of leukemia self-renewal genes and found that ESCO2 deficiency led to a decrease in SMC3 binding at these loci (Fig. 3A). The Venn diagram analysis also revealed that SMC3 exhibited a significant loss of chromatin binding at the majority of gene loci associated with ESCO2 depletion-induced self-renewal (Fig. 3B). Genome-wide analysis revealed a widespread reduction in cohesin-associated chromatin signals following ESCO2 loss. Correspondingly, a substantial proportion of self-renewal genes exhibited reduced cohesin binding upon ESCO2 depletion (Fig. 3B). Given the broad nature of this effect, overlap analysis alone is insufficient to infer specificity. Therefore, subsequent analyses focused on quantitative changes and locus-specific regulation at representative self-renewal gene loci. However, a small subset of loci demonstrated an unexpected increase in SMC3 binding (Supplementary Fig. 4A). These loci were distinct from those regulated by the ESCO2-SMC3-NuRD axis described above. Notably, genes associated with increased SMC3 occupancy displayed elevated H3K27ac levels and increased chromatin accessibility, accompanied by higher transcriptional activity (Supplementary Fig. 4B). The hyperacetylation-mimetic SMC3 mutation, rather than wild-type SMC3, is capable of reversing the abnormal expression of HOXA10 and CD38 genes induced by ESCO2 depletion (Fig. 3C). These data demonstrate that stable chromatin occupancy of the cohesin complex at loci of self-renewal-associated genes restricts their transcriptional activation.Fig. 3Loss of ESCO2 disrupts cohesin binding, increases H3K27ac levels, and enhances chromatin accessibility of self-renewal genes.A Average SMC3 occupancy at genomic regions corresponding to self-renewal genes induced by ESCO2 deficiency in control and sgESCO2#1 OCI-AML3 cells, n = 3 biologically independent experiments. B The Venn diagram illustrates the overlap between ESCO2 deficiency-activated self-renewal genes and regions lost in SMC3 CUT&Tag data, n = 3 biologically independent experiments. C Quantitative RT-PCR analysis of selected self-renewal genes in OCI-AML3 control cells, OCI-AML3 sgESCO2#1 cells, and OCI-AML3 sgESCO2#1 cells stably overexpressing SMC3-WT/K2Q. Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD (n = 3). D Tandem affinity purification was performed in OCI-AML3 cells, where SMC3 was immunopurified using an anti-SMC3 antibody and Protein G beads (with IgG as a negative control), followed by mass spectrometric analysis. A schematic diagram of the NuRD complex (left) and the identified NuRD subunits from the mass spectrometry analysis are listed in the table (right). E The interaction between Ac-SMC3 and HDAC1/HDAC2/MTA2 was confirmed by co-IP in OCI-AML3 and OCI-AML2 cells. Representative western blot images from three independent experiments with similar results are shown. F Average HDAC1 occupancy at self-renewal gene loci that are up-regulated by ESCO2 deficiency in control and sgESCO2#1 OCI-AML3 cells, n = 3 biologically independent experiments. G Average H3K27ac occupancy at self-renewal gene loci induced by ESCO2 deficiency, n = 3 biologically independent experiments. H ATAC-seq signals at self-renewal genes in control cells and OCI-AML3 sgESCO2#1 cells, n = 3 biologically independent experiments. I Genome browser view of SMC3, H3K27ac, HDAC1, and MTA2 CUT&Tag signals at HOXA loci in control and sgESCO2#1 OCI-AML3 cells, n = 3 biologically independent experiments. Source data are provided as a Source Data file. Loss of ESCO2 disrupts cohesin binding, increases H3K27ac levels, and enhances chromatin accessibility of self-renewal genes. A Average SMC3 occupancy at genomic regions corresponding to self-renewal genes induced by ESCO2 deficiency in control and sgESCO2#1 OCI-AML3 cells, n = 3 biologically independent experiments. B The Venn diagram illustrates the overlap between ESCO2 deficiency-activated self-renewal genes and regions lost in SMC3 CUT&Tag data, n = 3 biologically independent experiments. C Quantitative RT-PCR analysis of selected self-renewal genes in OCI-AML3 control cells, OCI-AML3 sgESCO2#1 cells, and OCI-AML3 sgESCO2#1 cells stably overexpressing SMC3-WT/K2Q. Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD (n = 3). D Tandem affinity purification was performed in OCI-AML3 cells, where SMC3 was immunopurified using an anti-SMC3 antibody and Protein G beads (with IgG as a negative control), followed by mass spectrometric analysis. A schematic diagram of the NuRD complex (left) and the identified NuRD subunits from the mass spectrometry analysis are listed in the table (right). E The interaction between Ac-SMC3 and HDAC1/HDAC2/MTA2 was confirmed by co-IP in OCI-AML3 and OCI-AML2 cells. Representative western blot images from three independent experiments with similar results are shown. F Average HDAC1 occupancy at self-renewal gene loci that are up-regulated by ESCO2 deficiency in control and sgESCO2#1 OCI-AML3 cells, n = 3 biologically independent experiments. G Average H3K27ac occupancy at self-renewal gene loci induced by ESCO2 deficiency, n = 3 biologically independent experiments. H ATAC-seq signals at self-renewal genes in control cells and OCI-AML3 sgESCO2#1 cells, n = 3 biologically independent experiments. I Genome browser view of SMC3, H3K27ac, HDAC1, and MTA2 CUT&Tag signals at HOXA loci in control and sgESCO2#1 OCI-AML3 cells, n = 3 biologically independent experiments. Source data are provided as a Source Data file. Considering the absence of reported transcription factor capabilities for the cohesin complex, we hypothesized that it modulates the expression of leukemia self-renewal genes through protein interaction networks. To test this hypothesis, we performed immunoprecipitation-mass spectrometry (IP-MS) and discovered multiple NuRD complex subunits as interacting proteins with SMC3 (Fig. 3D). The interaction between key NuRD complex subunits (HDAC1, HDAC2, MTA2) and SMC3 was further confirmed by co-IP assays in both OCI-AML3 and OCI-AML2 cells (Fig. 3E). Notably, ESCO2 depletion did not affect the cohesin-NuRD interaction (Supplementary Fig. 5A), and this interaction was not mediated by DNA (Supplementary Fig. 5B). CUT&Tag analysis revealed significantly reduced HDAC1 binding at self-renewal gene loci in ESCO2-depleted cells (Fig. 3F). The NuRD complex is characterized as a conserved transcription corepressor that has ATP-dependent chromatin remodeling and histone deacetylation activity50,51. NuRD complex deacetylates histone H3K27 and represses gene expression in B lymphocytic leukemia and erythroid progenitor cells52,53. Given the deacetylation activity of the NuRD complex, we further conducted H3K27ac CUT&Tag in both control and ESCO2-deficient cells. The results showed that H3K27ac histone modification at leukemia self-renewal gene loci was increased in ESCO2-deficient cells (Fig. 3G). Additionally, ATAC-seq analyses revealed that ESCO2 deficiency enhanced the chromatin accessibility of these self-renewal genes (Fig. 3H). Importantly, the CUT&Tag peaks for SMC3, HDAC1, and MTA2 overlapped at self-renewal gene loci, including the HOXA/B loci, and were reduced upon ESCO2 depletion. In contrast, H3K27ac enrichment at these loci was increased following ESCO2 loss (Fig. 3I, Supplementary Fig. 5C–G). Notably, MTA2 binding was reduced at the HOXB and additional self-renewal gene loci, whereas the reduction at the HOXA loci was more modest (Fig. 3I). To determine whether SMC3 acetylation is required for NuRD complex chromatin recruitment, we performed CUT&Tag to profile HDAC1 occupancy at self-renewal–associated gene loci. In ESCO2-deficient OCI-AML3 cells reconstituted with either wild-type SMC3 or the acetylation-mimetic SMC3-K2Q mutant, overexpression of SMC3-K2Q mutant markedly enhanced HDAC1 binding at these loci compared to wild-type SMC3 (Supplementary Fig. 5H). Collectively, these results suggest that the cohesin complex dampens the transcription of leukemia self-renewal genes in collaboration with the NuRD complex. ESCO2 depletion reduces the chromatin binding of both complexes, thereby promoting the expression of self-renewal genes. HDAC8 inhibition attenuates leukemia self-renewal through promoting cohesin chromatin retention HDAC8 is responsible for the deacetylation of SMC3 K105/106, as well as the release of cohesin from chromatin25,54,55. PCI-34051 is a selective HDAC8 inhibitor that blocks SMC3 deacetylation and exhibits cytotoxicity against leukemia56. We first confirmed that HDAC8 inhibition increased SMC3 K105/106 acetylation in both control and ESCO2-deficient cells (Fig. 4A). Consistent with this observation, CUT&Tag analysis revealed that HDAC8 inhibition enhanced SMC3 occupancy at self-renewal gene loci in ESCO2-depleted cells (Fig. 4B, C), suggesting that HDAC8 inhibition can restore SMC3 acetylation under hypoacetylated conditions. We treated both control and ESCO2-depleted OCI-AML3 cells with PCI-34051 and performed RNA-seq to analyze the resulting transcriptome changes. The result revealed that PCI-34051 treatment decreased the expression levels of leukemia self-renewal genes that were upregulated due to ESCO2 deficiency (Fig. 4D). Furthermore, several oncogenic biological processes, such as cell proliferation, DNA replication, and rRNA processing, which were enhanced following ESCO2 depletion, were suppressed by PCI-34051 treatment (Fig. 4E). We did not observe any changes in global histone acetylation upon PCI-34051 treatment (Supplementary Fig. 6A), thereby excluding the possibility that HDAC8 inhibition regulates gene expression through direct modulation of histone acetylation. Consistently, siRNA-mediated knockdown of HDAC8 in both control and ESCO2-depleted cells resulted in increased SMC3 K105/106 acetylation (Supplementary Fig. 6B). HDAC8 depletion also suppressed colony-forming capacity (Supplementary Fig. 6C) and promoted AML cell apoptosis (Supplementary Fig. 6D). Importantly, the combination of HDAC8 knockdown and PCI-34051 treatment did not produce additional effects on SMC3 acetylation, colony formation, or apoptosis (Supplementary Fig. 6B–D). These findings indicate that the cellular effects of PCI-34051 in OCI-AML3 cells are specifically mediated through HDAC8.Fig. 4A selective HDAC8 inhibitor attenuates self-renewal driven by ESCO2 deficiency in NPM1-mutated AML Cells.A Western blot analysis confirmed the SMC3 acetylation levels in OCI-AML3 cells treated with or without specific HDAC8 inhibitor PCI-34051 (50 μM, 48 h). Representative western blot images from three independent experiments with similar results are shown. B Average SMC3 occupancy at genomic regions associated with self-renewal genes in OCI-AML3 sgESCO2#1 cells treated with or without PCI-34051 (50 μM, 48 h), n = 3 biologically independent experiments. C Genome browser view of SMC3 at HOXA/B loci in OCI-AML3 sgESCO2#1 cells treated with or without HDAC8 inhibitor PCI-34051, n = 3 biologically independent experiments. D Heatmap showing the expression of 94 ESCO2 deficiency-induced self-renewal genes in the indicated treatment. Gene expression levels are normalized by z-score, n = 3 biologically independent experiments. E Bubble diagram showing the enrichment of Gene Ontology (GO) pathways in transcripts that were upregulated by ESCO2 deficiency and downregulated by PCI-34051 treatment. Enrichment significance was evaluated using the hypergeometric test with Benjamini–Hochberg correction, n = 3 biologically independent experiments. F Cell proliferation measured using CCK-8 assays (n = 3). G–I Colony numbers, percentage of CD34-CD38high cell population, and apoptotic cell rate of control and sgESCO2#1/2 OCI-AML3 cells with or without PCI-34051 treatment (n = 3). J Representative Ray-Giemsa staining images of OCI-AML3 sgESCO2#1 cells with or without PCI-34051 treatment. Scale bars, 10 μm. Data are presented as mean ± SD (n = 20). K PCA plots showing the distribution of normal HSC (n = 3), GMP (n = 3), CMP (n = 3), GM-Pre (n = 5), GM-Mat (n = 4), and control and sgESCO2#1/2 OCI-AML3 cells, with or without PCI-34051 treatment, using the expression levels of 500 selected genes. The percentage of variance represented by each PC is indicated. L, M Flow cytometry plot showing CD11b expression in control and sgESCO2#1/2 OCI-AML3 cells without or with PCI-34051 treatment at the indicated time points. Representative results from three independent experiments with similar results are shown. Statistical analysis in panels F–J was performed using an unpaired, two-tailed Student t test. Data are presented as mean ± SD. Source data are provided as a Source Data file. A selective HDAC8 inhibitor attenuates self-renewal driven by ESCO2 deficiency in NPM1-mutated AML Cells. A Western blot analysis confirmed the SMC3 acetylation levels in OCI-AML3 cells treated with or without specific HDAC8 inhibitor PCI-34051 (50 μM, 48 h). Representative western blot images from three independent experiments with similar results are shown. B Average SMC3 occupancy at genomic regions associated with self-renewal genes in OCI-AML3 sgESCO2#1 cells treated with or without PCI-34051 (50 μM, 48 h), n = 3 biologically independent experiments. C Genome browser view of SMC3 at HOXA/B loci in OCI-AML3 sgESCO2#1 cells treated with or without HDAC8 inhibitor PCI-34051, n = 3 biologically independent experiments. D Heatmap showing the expression of 94 ESCO2 deficiency-induced self-renewal genes in the indicated treatment. Gene expression levels are normalized by z-score, n = 3 biologically independent experiments. E Bubble diagram showing the enrichment of Gene Ontology (GO) pathways in transcripts that were upregulated by ESCO2 deficiency and downregulated by PCI-34051 treatment. Enrichment significance was evaluated using the hypergeometric test with Benjamini–Hochberg correction, n = 3 biologically independent experiments. F Cell proliferation measured using CCK-8 assays (n = 3). G–I Colony numbers, percentage of CD34-CD38high cell population, and apoptotic cell rate of control and sgESCO2#1/2 OCI-AML3 cells with or without PCI-34051 treatment (n = 3). J Representative Ray-Giemsa staining images of OCI-AML3 sgESCO2#1 cells with or without PCI-34051 treatment. Scale bars, 10 μm. Data are presented as mean ± SD (n = 20). K PCA plots showing the distribution of normal HSC (n = 3), GMP (n = 3), CMP (n = 3), GM-Pre (n = 5), GM-Mat (n = 4), and control and sgESCO2#1/2 OCI-AML3 cells, with or without PCI-34051 treatment, using the expression levels of 500 selected genes. The percentage of variance represented by each PC is indicated. L, M Flow cytometry plot showing CD11b expression in control and sgESCO2#1/2 OCI-AML3 cells without or with PCI-34051 treatment at the indicated time points. Representative results from three independent experiments with similar results are shown. Statistical analysis in panels F–J was performed using an unpaired, two-tailed Student t test. Data are presented as mean ± SD. Source data are provided as a Source Data file. Consistently, the results of CCK-8 and colony formation assays confirmed the inhibitory effect of PCI-34051 on tumor cell proliferation and clonogenic potential (Fig. 4F, G). Additionally, HDAC8 inhibition reduced the proportion of the CD34-CD38high cell subpopulation and induced cell apoptosis (Fig. 4H, I). Giemsa staining showed that PCI-34051 treatment reduced the nuclear-to-cytoplasmic ratio in AML cells compared to controls, accompanied by a marked decrease in nucleolar size and number (Fig. 4J). PCA analysis showed that PCI-34051 intervention significantly stimulated differentiation towards granulocyte-macrophage mature (GM-Mat) cell phenotypes (Fig. 4K). We also examined the impact of PCI-34051 on CD11b expression. The result indicated that PCI-34051 treatment increased CD11b levels (Fig. 4L), suggesting that HDAC8 inhibition induced leukemia cell differentiation. Time course analysis revealed that PCI-34051 treatment induced progressive differentiation increase in ESCO2-deficient cells in a time-dependent manner, whereas control cells maintained a persistent differentiation blockade (Fig. 4M, Supplementary Fig. 6E). Notably, ESCO2-depleted cells were hypersensitive to PCI-34051 (Figs. 4F–I, 4K–M), indicating that the differentiation and apoptosis induced by HDAC8 inhibition specifically attributed to the promotion of SMC3 acetylation in these cells. To further assess the role of SMC3 K105/106 acetylation in regulating AML self-renewal capacity and responsiveness to PCI-34051, we stably overexpressed the non-acetylatable SMC3-K2R mutant in both control and ESCO2-deficient cells. In control cells, expression of the SMC3-K2R mutant was sufficient to increase HOXA9/10 expression and enhance colony-forming capacity (Supplementary Fig. 6F, G). Notably, in contrast to the sensitivity of ESCO2-deficient cells to PCI-34051, control cells expressing SMC3-K2R exhibited resistance to HDAC8 inhibition, as PCI-34051 treatment failed to reduce HOXA9/10 expression or colony-forming ability (Supplementary Fig. 6F, G). This phenotype was consistent with that observed in ESCO2-deficient cells stably expressing SMC3-K2R, in which PCI-34051 likewise had no detectable effect (Supplementary Fig. 6H–J). Collectively, these results indicate that restoring SMC3 K105/106 acetylation through HDAC8 inhibition promotes differentiation and apoptosis in ESCO2-deficient NPM1-mutated AML. The combined inhibition of HDAC8 and XPO1 synergistically triggers epigenetic reprogramming and induces differentiation To investigate the synergistic interactions between ESCO2 depletion and NPM1c mutant in AML progression and to develop targeted therapeutic strategies, we evaluated the effect of ESCO2 deficiency and NPM1c mutant on the proportion of leukemia-initiating populations and cell proliferation rate. We established a doxycycline (Dox)-inducible expression system in OCI-AML2 cells (which are intrinsically NPM1c-negative) to enable controlled ectopic expression of both wild-type NPM1 and NPM1c mutant proteins (Fig. 5A, Supplementary Fig. 7A, B). To determine whether NPM1c mutant and ESCO2 loss cooperatively regulate chromatin acetylation, we analyzed H3K27ac levels using the doxycycline-inducible NPM1c expression system. Consistent with previous studies showing that NPM1c independently enhances H3K27ac levels by inhibiting HDAC1 activity7, induction of NPM1c expression led to increased H3K27ac enrichment at self-renewal gene loci. Similarly, ESCO2 depletion alone resulted in elevated H3K27ac levels at these regions. Notably, simultaneous NPM1c expression and ESCO2 loss further increased H3K27ac enrichment compared with either condition alone, indicating a cooperative effect on H3K27ac regulation (Fig. 5B). Consistent with the elevated H3K27ac levels, co-occurrence of NPM1c expression and ESCO2 loss was associated with a further increase in HOXA9/10 (Fig. 5C). In line with these observations, combined NPM1c mutant and ESCO2 depletion synergistically increased the proportion of leukemia-initiating cells and CD38 levels (Fig. 5D, E, Supplementary Fig. 7C). Although individual ESCO2 loss or NPM1c mutant alone does not significantly accelerate cell viability, their co-occurrence triggers a synergistic effect that promotes AML cell proliferation (Fig. 5F). In contrast, induction of wild-type NPM1 (NPM1wt) expression did not enhance self-renewal capacity, nor did it cooperate with ESCO2 loss to increase the proportion of leukemia-initiating cells (Supplementary Fig. 7D) or upregulate HOXA gene expression (Supplementary Fig. 7E). Additionally, NPM1c expression did not significantly affect genome-wide SMC3 binding or SMC3 occupancy at self-renewal gene loci (Supplementary Fig. 7F). These findings indicate that NPM1c is indispensable for the enhanced self-renewal capacity observed in ESCO2-deficient cells.Fig. 5The combination of HDAC8 and XPO1 inhibitors synergistically triggers epigenetic reprogramming and induces differentiation.A The doxycycline-induced expression of both wild-type Flag-NPM1 and NPM1c mutant was verified in both OCI-AML2 control and ESCO2-depleted cells using western blot analysis. Representative western blot images from three independent experiments with similar results are shown. B Average H3K27ac occupancy at self-renewal gene loci in the doxycycline-induced expression system following the indicated treatments, n = 3 biologically independent experiments. C Quantitative RT-PCR analysis of HOXA9/10 expression in the doxycycline-induced expression system following the indicated treatments (n = 3). D The bar chart illustrates the percentage of CD34-CD38high cell population in OCI-AML2 sgControl or sgESCO2 cells, either untreated or treated with Dox to induce the expression of NPM1c (n = 3). E Quantitative FACS analysis of CD38 expression in control and sgESCO2 OCI-AML2 cells, either untreated or treated with Dox to induce the expression of NPM1c (n = 3). F The effects of ESCO2 deficiency and NPM1c mutation on cell proliferation were detected using CCK-8 assays (n = 3). G Average H3K27ac occupancy at self-renewal gene loci in sgESCO2#1 OCI-AML3 cells following the indicated treatments (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h), n = 3 biologically independent experiments. H Quantitative RT-PCR analysis of HOXA9/10 expression in sgESCO2#1 OCI-AML3 cells following the indicated treatments (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h). Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD (n = 3). I Bubble diagram showing the enrichment of GO pathways in transcripts that were upregulated and downregulated by combinational treatment versus KPT-8602 mono-treatment (50 nM, 48 h) in sgESCO2#1 OCI-AML3 cells. Statistical significance of GO term enrichment was evaluated using the hypergeometric test, and P-values were adjusted for multiple comparisons using the Benjamini–Hochberg method, n = 3 biologically independent experiments. J Flow plot for CD11b in control and sgESCO2#1/2 OCI-AML3 cells following the indicated treatments (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h). Representative results from three independent experiments with similar results are shown. Statistical analysis in panels C–F was performed using two-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD (n = 3). Source data are provided as a Source Data file. The combination of HDAC8 and XPO1 inhibitors synergistically triggers epigenetic reprogramming and induces differentiation. A The doxycycline-induced expression of both wild-type Flag-NPM1 and NPM1c mutant was verified in both OCI-AML2 control and ESCO2-depleted cells using western blot analysis. Representative western blot images from three independent experiments with similar results are shown. B Average H3K27ac occupancy at self-renewal gene loci in the doxycycline-induced expression system following the indicated treatments, n = 3 biologically independent experiments. C Quantitative RT-PCR analysis of HOXA9/10 expression in the doxycycline-induced expression system following the indicated treatments (n = 3). D The bar chart illustrates the percentage of CD34-CD38high cell population in OCI-AML2 sgControl or sgESCO2 cells, either untreated or treated with Dox to induce the expression of NPM1c (n = 3). E Quantitative FACS analysis of CD38 expression in control and sgESCO2 OCI-AML2 cells, either untreated or treated with Dox to induce the expression of NPM1c (n = 3). F The effects of ESCO2 deficiency and NPM1c mutation on cell proliferation were detected using CCK-8 assays (n = 3). G Average H3K27ac occupancy at self-renewal gene loci in sgESCO2#1 OCI-AML3 cells following the indicated treatments (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h), n = 3 biologically independent experiments. H Quantitative RT-PCR analysis of HOXA9/10 expression in sgESCO2#1 OCI-AML3 cells following the indicated treatments (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h). Statistical analysis was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD (n = 3). I Bubble diagram showing the enrichment of GO pathways in transcripts that were upregulated and downregulated by combinational treatment versus KPT-8602 mono-treatment (50 nM, 48 h) in sgESCO2#1 OCI-AML3 cells. Statistical significance of GO term enrichment was evaluated using the hypergeometric test, and P-values were adjusted for multiple comparisons using the Benjamini–Hochberg method, n = 3 biologically independent experiments. J Flow plot for CD11b in control and sgESCO2#1/2 OCI-AML3 cells following the indicated treatments (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h). Representative results from three independent experiments with similar results are shown. Statistical analysis in panels C–F was performed using two-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD (n = 3). Source data are provided as a Source Data file. To further evaluate the generalizability of our findings, we ectopically expressed NPM1c and depleted ESCO2 in an additional AML cell line, KG-1α, which is intrinsically negative for NPM1c (Supplementary Fig. 7G), and examined their effects on colony-forming capacity and HOXA9/10 expression. The results showed that either NPM1c overexpression or ESCO2 knockdown individually enhanced clonogenic potential and elevated HOXA9/10 transcript levels. Importantly, the combination of NPM1c expression and ESCO2 depletion resulted in a synergistic increase in both colony formation and HOXA9/10 expression (Supplementary Fig. 7H, I). These findings align with our observations in OCI-AML3 cells and in the doxycycline-inducible NPM1c system in OCI-AML2. XPO1 mediates the cytoplasmic transport of NPM1c, and XPO1 inhibitor treatment results in nuclear relocalization of NPM1c2. XPO1 inhibition displaces NPM1c from chromatin, suppresses HOXA gene expression and induces the differentiation and apoptosis in NPM1-mutated AML cells6–8. First, we confirmed that monotreatment with XPO1 inhibitor KPT-8602 induced nuclear retention of the NPM1c mutant (Supplementary Fig. 8A) and downregulated HOXA9/10 gene expression (Supplementary Fig. 8B). We next evaluated the combination of an HDAC8 inhibitor PCI-34051 with KPT-8602. Previous study revealed that KPT-8602 represses the expression of HOXA genes by reducing H3K27ac level in OCI-AML38. Our results showed that the combination therapy further diminished H3K27ac levels at self-renewal gene loci and consequently suppressed HOXA9/10 gene transcription (Fig. 5G, H). Consistent with these findings, combined treatment also suppressed HOXA9/10 expression in ESCO2-knockdown KG-1α cells overexpressing NPM1c (Supplementary Fig. 9A). Transcriptomic profiling revealed that the combination treatment markedly suppressed key biological processes in AML cells, including transcription and translation (Supplementary Fig. 9B). Notably, the dual-agent treatment elicited substantially enhanced apoptotic and differentiation response compared to KPT-8602 monotherapy (Fig. 5I). Co-treatment further enhanced surface CD11b expression in ESCO2-depleted AML cells (Fig. 5J). These findings demonstrate that the combination of HDAC8 and XPO1 inhibitors remodels the epigenetic landscape at self-renewal gene loci and triggers the differentiation of AML cells. Co-targeting HDAC8 and XPO1 demonstrates anti-leukemia efficacy in both in vitro and in vivo studies To evaluate the cytotoxic efficacy of the combination of HDAC8 inhibitor PCI-34051 and XPO1 inhibitor KPT-8602 in NPM1-mutated AML, we assessed the synergistic anti-leukemia effect of these two small molecules in vitro (Fig. 6A). The results showed that combination treatment significantly enhanced apoptosis induction, exerted stronger inhibitory effects on clonogenic capacity, and effectively reduced cellular proliferation in both OCI-AML3 cells and KG-1α cells with ESCO2 depletion and NPM1c overexpression (Fig. 6B–D, Supplementary Fig. 10A). In ESCO2-deficient cells, treatment with KPT-8602 alone resulted in an IC₅₀ of 59.8 nM, as determined by cell viability assays. Notably, co-treatment with PCI-34051 markedly enhanced cellular sensitivity to KPT-8602, leading to a pronounced reduction of the IC₅₀ to 17.5 nM (Fig. 6E). Quantitative drug interaction analysis using SynergyFinder revealed a synergistic effect between KPT-8602 and PCI-34051 (Fig. 6F). In AML CDX models, ESCO2-deficient OCI-AML3 cells were transplanted into immunodeficient recipient mice, and treatment was initiated with vehicle control, PCI-34051 (40 mg/kg, 5 times a week), and/or KPT-8602 (15 mg/kg, 3 times a week) (Fig. 6G). Remarkably, the combination of PCI-34051 and KPT-8602 significantly extended the survival of mice compared to monotherapy (Fig. 6H). Consistently, the combination treatment alleviated the symptoms of splenomegaly and resulted in a more rapid clearance of leukemia cells from spleen and bone marrow (Fig. 6I–K, Supplementary Fig. 10B, C). We isolated primary AML cells from patients harboring NPM1 mutation (Fig. 6L) and found that ESCO2 protein level in these patient-derived blasts was significantly lower than in healthy control PBMCs (peripheral blood mononuclear cells) (Supplementary Fig. 10D). We next investigated the effect of combined treatment on primary AML cells from patients with NPM1 mutation. The results showed that combined treatment promoted cell apoptosis, reduced colony-forming capacity and suppressed HOXA9/10 expression compared with mono-treatment (Fig. 6M–O). In the ex vivo assays using primary NPM1-mutant AML samples, we observed a relatively high baseline level of apoptosis in control cultures, a phenomenon frequently encountered in such experimental systems. Nevertheless, the combination of PCI-34051 and KPT-8602 reproducibly induced a further and statistically significant increase in apoptosis compared with either agent alone. Finally, we assessed the therapeutic potential of the combo in a patient-derived xenografts in vivo model. The combined treatment significantly extended survival (Fig. 6P). These studies demonstrate the in vivo efficacy of combined therapy and support this approach as a promising strategy for patients with NPM1-mutated AML.Fig. 6The combination of HDAC8 and XPO1 inhibitors demonstrates anti-leukemia efficacy both in vitro and in vivo.A schematic diagram illustrating the detection of the efficiency of combining HDAC8 and XPO1 inhibitors (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h). Created in BioRender. Fu, J. (2026) https://BioRender.com/p3es5dp. B–D Percentage of apoptotic cells, cell proliferation rate and colony formation numbers in control and sgESCO2#1/2 OCI-AML3 cells following the indicated treatments (n = 3). E Half-maximal inhibitory concentrations (IC₅₀) were calculated by fitting dose–response curves using nonlinear regression analysis with a four-parameter logistic model. Cell viability data were normalized to controls (n = 3). F Drug interaction analysis was performed using the SynergyFinder+ platform. Synergy scores were calculated based on dose–response matrices generated from cell viability data. Representative western blot images from three independent experiments with similar results are shown. G A schematic diagram illustrating the in vivo efficiency of combining HDAC8 and XPO1 inhibitors. Created in BioRender. Fu, J. (2026) https://BioRender.com/pka7opp. H Kaplan–Meier survival curves for control (n = 5), PCI-34051 (n = 5), KPT-8602 (n = 5), and combination treatment (n = 5) mice. P values calculated using the log-rank test. I Spleen weight quantification from transplanted mice (n = 5). J Representative H&E staining images showing leukemic infiltration in the spleen. Scale bars, 500 μm (n = 5). The cells stained bluish-purple are infiltrating AML cells, while the red cells represent normal tissue cells. K Representative H&E staining images showing leukemic infiltration in bone marrow (n = 5). Scale bars, 50 μm. L Schematic illustrating the experimental design using primary AML cells treated with PCI-34051 and/or KPT-8602 for evaluation. Created in BioRender. Fu, J. (2026) https://BioRender.com/x6qz64s. M–O Percentage of apoptotic cells, colony formation capacity and HOXA9/10 expression level in primary AML cells following the indicated treatments (n = 3). P Kaplan–Meier curves comparing overall survival in PDX mice across four treatment groups: untreated controls (n = 5), PCI-34051 monotherapy (n = 5), KPT-8602 monotherapy (n = 5), and combination treatment (n = 5). P-values calculated using the log-rank test. Statistical analysis in panels B–D, I and M–O was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD. Source data are provided as a Source Data file. The combination of HDAC8 and XPO1 inhibitors demonstrates anti-leukemia efficacy both in vitro and in vivo. A schematic diagram illustrating the detection of the efficiency of combining HDAC8 and XPO1 inhibitors (PCI-34051: 50 μM, 48 h; KPT-8602: 50 nM, 48 h). Created in BioRender. Fu, J. (2026) https://BioRender.com/p3es5dp. B–D Percentage of apoptotic cells, cell proliferation rate and colony formation numbers in control and sgESCO2#1/2 OCI-AML3 cells following the indicated treatments (n = 3). E Half-maximal inhibitory concentrations (IC₅₀) were calculated by fitting dose–response curves using nonlinear regression analysis with a four-parameter logistic model. Cell viability data were normalized to controls (n = 3). F Drug interaction analysis was performed using the SynergyFinder+ platform. Synergy scores were calculated based on dose–response matrices generated from cell viability data. Representative western blot images from three independent experiments with similar results are shown. G A schematic diagram illustrating the in vivo efficiency of combining HDAC8 and XPO1 inhibitors. Created in BioRender. Fu, J. (2026) https://BioRender.com/pka7opp. H Kaplan–Meier survival curves for control (n = 5), PCI-34051 (n = 5), KPT-8602 (n = 5), and combination treatment (n = 5) mice. P values calculated using the log-rank test. I Spleen weight quantification from transplanted mice (n = 5). J Representative H&E staining images showing leukemic infiltration in the spleen. Scale bars, 500 μm (n = 5). The cells stained bluish-purple are infiltrating AML cells, while the red cells represent normal tissue cells. K Representative H&E staining images showing leukemic infiltration in bone marrow (n = 5). Scale bars, 50 μm. L Schematic illustrating the experimental design using primary AML cells treated with PCI-34051 and/or KPT-8602 for evaluation. Created in BioRender. Fu, J. (2026) https://BioRender.com/x6qz64s. M–O Percentage of apoptotic cells, colony formation capacity and HOXA9/10 expression level in primary AML cells following the indicated treatments (n = 3). P Kaplan–Meier curves comparing overall survival in PDX mice across four treatment groups: untreated controls (n = 5), PCI-34051 monotherapy (n = 5), KPT-8602 monotherapy (n = 5), and combination treatment (n = 5). P-values calculated using the log-rank test. Statistical analysis in panels B–D, I and M–O was performed using one-way ANOVA followed by Tukey’s multiple comparison test. Data are presented as mean ± SD. Source data are provided as a Source Data file. Discussion Our study elucidates the synergistic interactions between ESCO2 deficiency and NPM1c mutant in the progression of AML. Our findings reveal the interdependent and cooperative epigenetic mechanisms that regulate the expression of leukemia self-renewal-associated genes, mediated through the cohesin complex and NPM1c mutant. ESCO2-mediated acetylation of SMC3 stabilizes the cohesin complex and the NuRD complex, which suppresses the expression of leukemia self-renewal genes by deacetylating H3K27ac at these loci. Loss of ESCO2 upregulates leukemia self-renewal genes and promotes AML progression (Fig. 7). Notably, we report a promising therapeutic strategy for NPM1-mutated AML, offering potential treatment avenues for this specific subtype of acute myeloid leukemia.Fig. 7Proposed model depicting the function and mechanism of ESCO2 in NPM1-mutated AML progression.ESCO2 deficiency destabilizes the cohesin complex on the promoters of self-renewal genes, thereby upregulating their expression. Targeting ESCO2-mediated acetylation of SMC3 and nuclear exporter XPO1 can induce differentiation of NPM1-mutated AML cells and enhance their elimination. Created in BioRender. Fu, J. (2026) https://BioRender.com/hnsjmj3. Proposed model depicting the function and mechanism of ESCO2 in NPM1-mutated AML progression. ESCO2 deficiency destabilizes the cohesin complex on the promoters of self-renewal genes, thereby upregulating their expression. Targeting ESCO2-mediated acetylation of SMC3 and nuclear exporter XPO1 can induce differentiation of NPM1-mutated AML cells and enhance their elimination. Created in BioRender. Fu, J. (2026) https://BioRender.com/hnsjmj3. The cohesin complex genes are among the most frequently mutated genes in clinical AML patients27–29, and partially dysfunction of cohesin blocks differentiation of HSPCs27,30–32,57. Here, we demonstrated the role of ESCO2 in NPM1-mutated AML by enhancing the binding of cohesin to chromatin. ESCO2 deficiency not only contributes to AML progression and poor prognosis but also presents a potential therapeutic vulnerability. ESCO2-deficient AML cells exhibit hypersensitivity to selective HDAC8 inhibitors. Given the function of cohesin in genomic stability, loss-of-function mutations in cohesin complex genes were previously thought to confer a proliferative disadvantage58. However, both in vivo genetic screening and in vitro assays showed that partial dysfunction of cohesin in hematopoietic stem cells leads to excessive proliferation of immature myeloid cells. This is caused by altered chromatin accessibility of genes involved in myelomonocytic maturation and differentiation30–32. Yet, the mechanism underlying cohesin mutation-driven chromatin structure changes remains obscure. Our findings demonstrate that ESCO2-mediated stabilization of the cohesin complex cooperates with the NuRD complex to suppress the transcription of myeloid self-renewal signature genes, including HOXA/B. In OCI-AML3 cells, these self-renewal genes are already transcriptionally active due to NPM1c-driven programs6,7. ESCO2 depletion does not initiate their expression de novo, but rather augments this pre-existing transcriptional state, thereby enhancing self-renewal capacity and promoting a more aggressive leukemic phenotype. Mechanistically, the NuRD complex deacetylates H3K27ac through HDAC1/2 and plays a critical role in restraining excessive transcriptional activation of these loci. Notably, our results indicating that ESCO2 depletion enhances leukemic self-renewal appear to differ from the growth-inhibitory effects predicted by DepMap screens in OCI-AML3 cells. Both our findings and several independent studies30–32 have demonstrated that loss of cohesin components can increase the self-renewal capacity and proliferative potential of HSPCs and AML cells, supporting a context-dependent, pro-leukemic role for cohesin dysfunction. The absence of karyotypic abnormalities in AML patients with cohesin loss-of-function mutations suggests the role of cohesin in AML progression may not be primarily mediated by its canonical function in mitosis27. Moreover, we also identified a subset of genomic loci that exhibit increased SMC3 occupancy following ESCO2 depletion (Supplementary Fig. 4B). This observation likely indicates compensatory mechanisms involving other cohesin components or regulators, such as NIPBL and ESCO1, which may promote transcriptional activation by enhancing chromatin looping, potentially in cooperation with other epigenetic regulators. Further studies are necessary to directly validate this hypothesis in the relevant biological context. Several inhibitors can overcome myeloid differentiation blockade and induce cell differentiation in NPM1-mutated, KMT2A-rearranged, MLL-rearranged, and IDH1/2-mutated AML59–61. However, a critical therapeutic gap persists in other AML subsets. Our results showed that stabilizing the cohesin complex on chromatin with the HDAC8 inhibitor PCI-34051 re-suppresses HOXA/B gene expression and promotes the differentiation of ESCO2-depleted AML cells. Previous study showed that HDAC8 deacetylates p53 and activates apoptosis in inv (16)+ AML cells62. The alignment of these findings with our results conclusively demonstrates the critical role of HDAC8 deacetylase activity across different AML subtypes. The mutation profiles of clinical AML patients indicate a strong correlation between mutations related to the cohesin complex and NPM1 mutations27,63. Our findings support a model of functional convergence. NPM1c has been reported to promote aberrant activation of self-renewal programs by interfering with HDAC1 activity at target loci8. In parallel, ESCO2 deficiency destabilizes cohesin-NuRD chromatin association, further impairing HDAC1-mediated repression. Thus, NPM1c and ESCO2 loss appear to converge on a common regulatory node, HDAC1-dependent control of H3K27ac, through distinct yet complementary mechanisms. The reproducibility of this cooperative effect across endogenous, inducible, and independent NPM1c-expressing AML models further supports the broader relevance of this epigenetic crosstalk. In this context, cohesin dysfunction, whether caused by mutations in cohesin subunits or impaired cohesin stabilization due to ESCO2 loss, further amplifies these transcriptional programs by additionally elevating H3K27ac levels at these loci. This cooperative interaction provides a potential mechanistic explanation for the frequent co-mutation of cohesin genes and NPM1 in AML, and highlights the context-dependent epigenetic vulnerability associated with perturbation of SMC3 acetylation in NPM1-mutated AML. PCI-34051 and KPT-8602, when used alone, have shown promising therapeutic efficacy against AML3,4,56. This highlights the importance of understanding treatment resistance mechanisms and developing drug combinations to improve the safety and efficacy of clinical treatment. By elucidating the synergistic functions of cohesin and NPM1c in AML, we showed that the combination of these two inhibitors can regulate leukemic transcription, leading to profound anti-leukemic effects both in vitro and in vivo. However, surface biomarkers for ESCO2-deficient cells within NPM1-mutated AML remain to be elucidated, which is a critical prerequisite for advancing therapeutic combinations into clinical practice. In conclusion, our findings uncover the molecular mechanism by which ESCO2 deficiency-induced hypoacetylation of SMC3 fuels AML, and provide a rational basis for the potential co-administration of HDAC8 and XPO1 inhibitors in the treatment of NPM1-mutated AML. Methods Ethics statement All experiments were conducted in accordance with relevant ethical regulations. Human peripheral blood samples were collected from Peking University People’s Hospital. This study was approved by the Ethics Committee of Peking University People’s Hospital (permit number: 2025PHB043-001) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to sample collection. Animal experiments were approved by the IACUC of the Center for Experimental Animal Research (China) and Peking University Laboratory Animal Center (IACUC No. LSC–ZhengX–1). Cell culture OCI-AML3 (RRID: CVCL_1844), OCI-AML2 (RRID: CVCL_1619) and KG-1α (RRID: CVCL_1824) cell lines were kindly provided by Jianwei Wang at School of Pharmaceutical Sciences, Tsinghua University, Bing Xu at the First Affiliated Hospital of Xiamen University and Hsiang-Ying Lee at School of Life sciences, Peking University. The identity of the cell lines was verified by the providers, and cells were used at low passage numbers after receipt. All cell lines were routinely tested and confirmed to be free of mycoplasma contamination by PCR analysis. Cells were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium with 10% FBS at 37 °C in a humidified incubator with 5% CO2. All cell lines were confirmed to have no mycoplasma contamination by PCR analysis. Plasmids Human SMC3 fragment was amplified by PCR and cloned into the pLVX-zsgreen vector. Wild-type NPM1 and NPM1c were FLAG-tagged and then cloned into the pTRE-EGFP lentiviral vector with the T2A-EGFP fragment. All plasmids were verified by DNA sequencing. Plasmids were transfected with Suspension Cell Special Transfection Reagent (MF388) according to the manufacturer’s instructions. Antibodies, chemicals and drugs Anti-ESCO2 (23525-1-AP, RRID: AB_2879292; WB: 1:1000), Anti-SMC3 (14185-1-AP, RRID: AB_2192625; WB: 1:1000, CUT&Tag: 1 μg), Anti-HDAC1 (10197-1-AP, RRID: AB_2118062; WB: 1:1000, CUT&Tag: 1 μg), Anti-HDAC2 (12922-3-AP, RRID: AB_2118516; WB: 1:1000, CUT&Tag: 1 μg), Anti-MTA2 (17554-1-AP, RRID: AB_10638601; WB: 1:1000, CUT&Tag: 1 μg), Anti-HDAC8 (17548-1-AP, RRID: AB_2116926; WB: 1:1000), Anti-H4ac (39244, RRID: AB_2793201; WB: 1:1000), Anti-H3K14ac (65702; WB: 1:1000) polyclonal antibodies were purchased from Proteintech (Wuhan, China). Anti-Flag (F3165, RRID: AB_259529; WB: 1:10000) mouse monoclonal antibodies were purchased from Sigma-Aldrich (Maryland, USA). Anti-H3K27ac (A22264; CUT&Tag: 1 μg), Anti-H3K9me3 (A22295, RRID: AB_3066640; IF: 1:100, CUT&Tag: 1 μg) rabbit polyclonal antibody was purchased from Abclone (Wuhan, China). FITC-conjugated goat anti-mouse IgG (ZF-0312, RRID: AB_2716306; IF: 1:200) and TRITC-conjugated goat anti-rabbit IgG (ZF-0316, RRID: AB_2728778; IF: 1:200) were purchased from ZSGB-BIO (Beijing, China). Doxycycline (HY-N0565, MCE), PCI-34051 (HY-15224, MCE), and KPT-8602 (HY-100423, MCE) were reconstituted in DMSO, stored at −80 °C and diluted. An equivalent proportion of DMSO was added to all control samples. Mice 5 × 106 OCI-AML3 cells were collected and injected through the tail vein into 6-week-old female NCG (NOD/ShiLtJGpt-Prkdcem26Cd52Il2rgem26Cd22/Gpt, Gempharmatech) mice. Only female mice were used in this study due to a higher incidence of donor cell engraftment. Mice were housed in a specific pathogen-free (SPF) facility under controlled environmental conditions with a 12-h light/12-h dark cycle, an ambient temperature of 22 ± 2 °C, and a relative humidity of 50–60%. Food and water were provided ad libitum. Animal welfare was monitored at least once daily, with assessments including body weight, posture, mobility, fur condition, and any signs of distress or illness. The maximum permitted tumor burden was defined as either a body weight loss exceeding 20% of the initial weight or the presence of severe clinical symptoms requiring humane euthanasia. Humane endpoints were implemented when mice exhibited signs of severe disease, such as impaired mobility, hunched posture, or other indicators of distress. Euthanasia was performed via CO₂ inhalation followed by cervical dislocation to ensure death. These mice are irradiated at a dose of 2.5 Gy within 12 h prior to the intravenous cell injection. The presence of leukemic cells in the spleen and bone marrow was subsequently examined using hematoxylin and eosin (H&E) staining. The maximum allowable disease burden, as defined by the Institutional Animal Care and Use Committee (IACUC) of the Center for Experimental Animal Research (China) and the Peking University Laboratory Animal Center, was set at either a body weight loss not exceeding 20% of the initial weight or the presence of severe clinical symptoms necessitating humane euthanasia. These predefined limits were not exceeded in this study. Human primary AML samples Peripheral blood samples were collected from Peking University People’s Hospital. The study cohort consisted of two female AML patients and three healthy donors (two females and one male). The sex of participants was recorded based on clinical records. Sex was not considered as a biological variable in the statistical analyses due to the limited sample size. Mononuclear cells were isolated from the peripheral blood samples using Ficoll-Paque Premium density gradient centrifugation with Septube (Stemcell, Canada). The samples were resuspended in serum-free cell-freezing medium (Yeasen, China) and stored at −80 °C until further use. In vivo drug treatment For in vivo experiments, PCI-34051 was dissolved in a solution containing 40% PEG 300, 10% DMSO, 5% Tween 80, and 45% saline. Meanwhile, KPT-8602 was dissolved in a solution composed of 0.6% (w/v) Pluronic F-68, 0.6% (w/v) PVP-K29/32 in sterile water. To evaluate the drug efficacy of the drugs, mice were administered either the vehicle control (saline) or the test compounds via intraperitoneal injection with PCI-34051 (40 mg/kg, 5 times per week) and/or oral gavage with KPT-8602 (15 mg/kg, 3 times per week) for two weeks. Stable expression cell lines construction To generate stable cell lines with human SMC3 overexpression, after 48 h of infection, OCI-AML3 cells were sorted using BD FACSAria™ III to obtain stable cell lines. To generate stable cell lines with Dox-inducible NPM1 expression, NPM1 wild-type and NPM1c were cloned into a pTRE-T2A-EGFP vector with an N-terminal FLAG tag. The NPM1c construct harbors the AML-associated C-terminal mutation that introduces a nuclear export signal. Both constructs were generated using the same vector backbone to ensure comparable expression levels. pTRE-NPM1wt-T2A-EGFP or pTRE-NPM1c-T2A-EGFP together with psPAX2 and pMD2.G plasmids were co-transfected into 293 T cells using PEI according to the manufacturer’s instructions (Polyscience). OCI-AML2 cells were then transduced by incubating with the appropriate volume of lentiviral particles for 24 h. Thereafter, 100 ng/ml Dox was added to the medium for 24 h. Next, cells were sorted using BD FACSAria™ III to obtain stable cell lines with inducible expression. Cells were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium with 10% FBS at 37 °C in a humidified incubator with 5% CO2. Generation of ESCO2 knockdown cell lines by CRISPR-cas9 system To investigate the function of the human ESCO2 gene, sgRNAs targeting different regions within its mRNA (sgESCO2#1: 5’-GAGAGCTCGCCGGGCTCTGA-3’; sgESCO2#2: 5’- TAAGTGGTACCTCAATCCAC-3’) were designed and cloned into a lentiviral sgRNA vector (sgControl: 5’-CACCGGCACTACCAGAGCTAACTCA-3’) containing an mCherry selection marker using the Golden Gate cloning method. Subsequently, cells were co-transfected with the sgRNA and Cas9 vectors. After 48 h of transfection, mCherry-positive cells were sorted by FACS (MOFLO, Cytomation). Single-cell clones were isolated following a 10-day selection period. The knockdown efficiency of ESCO2 was confirmed by immunoblotting. Additionally, ESCO2 gene mutations were verified through PCR and sequencing analyses. Sanger sequencing of sgRNA target regions was analyzed using ICE (Synthego) to quantify indel percentages and knockout scores. siRNA-mediated knockdown of HDAC8 and ESCO2 Small interfering RNA (siRNA) targeting human histone deacetylase 8 (HDAC8) was used to suppress HDAC8 expression. A pool of three to five target-specific siRNA duplexes against HDAC8 (HDAC8 siRNA (h), sc-35548, Santa Cruz Biotechnology, Dallas, TX, USA) was employed for transient knockdown. KG-1α cells were transfected with a pool of siRNAs targeting human ESCO2 (siESCO2-1, siESCO2-2 and siESCO2-3 mixed at an equal molar ratio). For transfection in a 6-well plate, the pooled siRNAs were diluted in Opti-MEM to a final total siRNA concentration of 50 nM. The pooled ESCO2-targeting siRNAs used were: siESCO2-1: forward: 5’-CCUGCAUUGCUCUCAAUAATT-3’; siESCO2-2: forward: 5’-GCAAAUCAAGGCUCACCAUTT-3’; siESCO2-3: forward: 5’-CUCUUAGACCAGGAUUAUCTT-3’. A non-targeting control siRNA (Santa Cruz Biotechnology) was used as a negative control. RNA isolation, RNA-seq and quantitative real-time PCR (qRT-PCR) Total RNA was extracted using the TRIzol Reagent (15596026, Thermo). Sequencing libraries were constructed from 500 ng of total RNA. The libraries were sequenced on the Illumina PE150 platform at Fulgent Genetics to generate paired-end reads of 2 × 150 bp. The RNA-seq raw reads were aligned to the human reference genome (hg38) using the STAR aligner (version 2.7.11) with default parameters. Gene-level read counts were generated and used for downstream analysis in R. Differential expression analysis was performed using DESeq2. Variance-stabilizing transformation was applied to the count matrix for visualization. Principal component analysis was performed to assess sample clustering, and hierarchical clustering heatmaps were generated based on sample-to-sample distances and the expression profiles of differentially expressed genes. Gene Ontology biological process enrichment analysis of upregulated and downregulated genes was carried out using clusterProfiler with the org.Hs.eg.db annotation package, with all filtered genes used as the background set. Reverse transcription was conducted using a reverse transcription kit (AG11705, Accurate Biology). Quantitative real-time PCR was subsequently carried out using specific primers (Supplementary Table 1) with a SYBR Green premixed qPCR reagent (AG11701, Accurate Biology). CUT&Tag and data analysis 1 × 106 cells were collected, and chromatin tagmentation was performed using the CUT&Tag Kit (N259-YH01, Novoprotein). The resulting CUT&Tag samples were sequenced on the Illumina PE150 platform. Sequencing reads were aligned to the human reference genome (hg38) using Bowtie2 v2.2.5 with default parameters. The aligned reads were processed using samtools v1.6 to convert them into BAM format, followed by sorting and indexing. Peak calling was performed using MACS2 v2.2.8 for all CUT&Tag datasets. For differential binding analysis, MACS2 peak files from all samples were analyzed using DiffBind in R. The chromatin distribution of the identified peaks was analyzed using the HOMER (RRID: SCR_010881) module annotatePeaks.pl v4.11. For visualization, the filtered, sorted, and indexed BAM files were converted to bigwig files using bamCoverage from deeptools v3.5.0 and normalized by reads per genomic content (RPGC). Heatmaps were generated using the computeMatrix and plotHeatmap from deeptools v3.5.0. Coverage plots were generated using the computeMatrix and plotProfile from deeptools v3.5.0. ATAC-seq and data analysis 1 × 106 cells were collected, and chromatin tagmentation was performed using the High-Sensitivity Open Chromatin Profile Kit (N248, Novoprotein). Clean reads were then aligned to the human reference genome (hg38) using bwa-mem, followed by the removal of PCR duplicates with Picard. Genome browser tracks in bigWig format were generated from merged replicates using samtools. Peaks were called using MACS2 with the following parameters: “-f BAMPE --nomodel --shift −100 --extsize 200 –nomodel”. The bigWig data were visualized using the Integrative Genomics Viewer (IGV) software (Broad Institute). Peak visualizations were prepared using Deeptools. Cytometric analysis Cells were stained with antibodies against human CD34 (clone 378603, RRID: AB_2941552, Biolegend; 0.25 µg per million cells in 100 µL), CD38 (clone 102705, RRID: AB_312926, Biolegend; 0.25 µg per million cells in 100 µL), CD11b (clone 982614, RRID: AB_2924634, Biolegend; 0.25 µg per million cells in 100 µL) lineage and were incubated for 45 min at room temperature. Flow cytometry analyses were conducted using a CytoFlEX LX instrument (Beckman Colter), and the data were analyzed using FlowJo software (v.10.4, RRID: SCR_008520). Cell proliferation assay Cells were seeded at a density of 1 × 104 per well into a 96-well plate. Cell proliferation was measured at 48 and 72 h using CCK-8 reagent. Following a 2-h incubation at 37 °C, the optical density at 450 nm was measured using an enzyme-labeled meter (Spectramax M3, Molecular Devices). Half-maximal inhibitory concentrations (IC₅₀) were calculated by fitting dose–response curves using nonlinear regression analysis with a four-parameter logistic model. Cell viability data were normalized to vehicle-treated controls and analyzed using GraphPad Prism software (Version 9.5.0). Colony formation assay 1 × 103 cells were suspended in 0.35% (w/v) agarose and plated over a base layer of 0.7% (w/v) agarose. Both layers contained complete RPMI 1640 medium. Cells were cultured at 37 °C for 15 days. Colony formation was monitored using an inverted fluorescence microscope (IX73, Olympus). The number and size of colonies in each field (at 20 × magnification) were recorded. True limiting dilution assay CD34-CD38high cells isolated from control and ESCO2-depleted OCI-AML3 cells were serially diluted to 2000, 1000, 500 and 100 cells per well and plated in 18 replicates for each density. After 7 days at 37 °C, 5% CO2, clusters were counted by microscopy. Frequencies of leukemia-initiating cells were determined with ELDA software using the Poisson statistics and the χ2-test64. Co-immunoprecipitation (co-IP) Cells were cultured in 10-cm dishes and transfected with the indicated plasmids. After 48 h, cells were washed with 1 mL of pre-chilled PBS and lysed for 60 min in 1 mL NP-40 lysis buffer (50 mM Tris–HCl, 150 mM NaCl, 1 mM EDTA, 1% NP-40, pH 7.4) containing 10 μL of protease inhibitor cocktail. Next, 1 μg of antibody or IgG was used to bind to the bait proteins for 4 h, and then incubated with 30 μL protein G for an additional 3 h. Finally, the protein G was washed with 1 mL of NP-40 lysis buffer three times and then heated for 10 min at 96 °C with 30 μL 2 × SDS loading buffer. Samples were analyzed by SDS-PAGE and western blotting. Immunoprecipitation tandem mass spectrometry OCI-AML3 cells were lysed in RIPA buffer, sonicated, and precleared with protein G beads. The supernatants were incubated with anti-SMC3 affinity antibody and protein G beads at 4 °C for 4 h. The eluates were precipitated with TCA and subjected to mass spectrometric analysis. The MS data were aligned with the Human Reviewed Swiss-Prot database using Proteome Discoverer 2.2 software (RRID: SCR_014477). Proteins were considered to be major hits (positive) when matching the following criteria: (i) not found in the negative control group, (ii) high Protein FDR Confidence (FDR < 0.01), and (iii) peptides ≥5. Immunofluorescence (IF) microscopy Cells were cultured in cell dishes and suspended on L-lysine–coated object slides. Cells were fixed with anhydrous methanol at room temperature for 10 min and blocked with PBS containing 1% BSA at 4 °C for 40 min. Next, cells were sequentially incubated with primary antibodies at 4 °C for 12 h, followed by incubation with secondary antibodies at room temperature for 1 h, and mounted with mounting medium with DAPI (ZSGB-BIO) at room temperature for 30 min. Images were captured using Carl ZEISS LSM710 confocal microscope (LSM 710 NLO & Duo Scan System). Duolink proximity ligation assay (PLA) PLAs were performed to examine the in situ interaction between H3K9me3 and wild-type Flag-NPM1/NPM1c. The assay was performed using the Duolink® In Situ PLA® kit (DUO92101, Sigma-Aldrich). Wright Giemsa stain Cells were harvested, washed once with PBS, and smeared onto poly-L-lysine–coated glass slides. The slides were air-dried, fixed in methanol for 5 min, and subsequently stained with Wright–Giemsa solution (Beyotime) following the manufacturer’s instructions. After staining, slides were rinsed with distilled water, air-dried, and examined under a light microscope at 100 × magnification. Statistics & reproducibility Statistical analyses were performed using GraphPad Prism and R. Data are presented as mean ± SD. unless otherwise indicated. P-values were determined using two-tailed Student’s t-test, one-way or two-way ANOVA as appropriate based on multiple independent biological replicates. The Kaplan-Meier method and the log-rank test were used to estimate the survival rates. No statistical method was used to predetermine sample size. The sample sizes used in this study are similar to those generally employed in the field. The sample sizes used were sufficient to detect biologically meaningful differences with the statistical tests applied. No data were excluded from the analyses. Animals were randomly assigned to experimental groups. The investigators were not blinded to allocation during experiments and outcome assessment. All experiments were repeated independently at least three times with similar results unless otherwise stated. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting summary Transparent Peer Review file Supplementary Information Reporting summary Transparent Peer Review file Source data Source data Source data