Stag2-dependent chromatin remodeling enforces the erythroid-specific Gata1 cistrome

Key Points

• •Stag2 reprograms Gata1 cistrome specificity via chromatin accessibility, a key determinant of transcription factor lineage discrimination.
• •Loss of the Stag2 rheostat causes accessibility-driven Gata1 retargeting, providing a mechanistic framework for dyserythropoiesis in MDS.

Visual Abstract

Abstract

The transcription factor GATA1 has pleiotropic hematopoietic functions, particularly in erythroid and megakaryocytic ontogeny. Although mechanistic investigations have uncovered many facets of GATA1 biology, how GATA1 coregulates divergent cell fates remains incompletely characterized. We previously described that loss of Stag2, a member of the cohesin complex and a recurrent mutational target in myelodysplastic syndrome (MDS) and myeloid leukemia of Down syndrome, results in altered chromatin accessibility, transcription factor function, and cell differentiation. Therefore, we hypothesized that chromatin accessibility determines GATA1 cistrome specificity and lineage fate decisions. To understand the connection between chromatin accessibility and GATA1, we comprehensively studied erythropoiesis in Stag2^Δ mice. Defects in Stag2-deficient hematopoiesis included reduced numbers of erythroid progenitors (EryPs), impaired terminal erythroid differentiation, increased number of megakaryocyte progenitors, and increased megakaryocytes. RNA- and assay for transposase-accessible chromatin–sequencing of EryPs revealed altered patterns of Gata1 target gene expression with altered accessibility in conjunction with loss of expression of erythroid targets and gain of megakaryocyte targets. Despite unchanged Gata1 expression, Gata1 occupancy was reprogrammed from erythroid to megakaryocyte targets with Fli1 motifs enriched at Stag2^Δ Gata1 binding sites. Functionally, we observed that Stag2-deficient EryPs have a diminished erythroid output and augmented megakaryocyte output in orthogonal differentiation assays, which was partially reversed with Fli1 knockdown. Human models and patients with primary MDS recapitulated the essential phenotypic and molecular features of our in vivo murine MDS model. Collectively, this study establishes chromatin accessibility as a determinant of transcription factor binding specificity, revealing an accessibility-driven Gata1 retargeting mechanism underlying MDS dyserythropoiesis.

Anemia is the most common presenting symptom of myelodysplastic syndromes (MDS), clonal myeloid malignancies characterized by ineffective hematopoiesis and myelodysplasia. STAG2 mutations in MDS are linked to severe anemia, erythroid dysplasia, and erythropoietin resistance, yet how STAG2 regulates erythropoiesis has been unclear. Sudunagunta and colleagues reveal that STAG2-dependent cohesion-mediated chromatin remodeling modulates the erythroid specificity of the GATA1 cistrome and its loss creates an accessibility-driven GATA1 retargeting that underlies dyserythropoiesis in MDS.

Introduction

Erythropoiesis is a tightly orchestrated process that transforms multipotent hematopoietic stem cells into mature erythrocytes through successive stages of lineage restriction, nuclear condensation, global transcriptional silencing, and enucleation.^1,2 Erythropoiesis is not only highly efficient but also remarkably adaptable under stress. Central to the regulation of erythropoiesis is a complex network of epigenetic mechanisms that govern chromatin accessibility, transcriptional dynamics, and nuclear architecture.^{3, 4, 5, 6} These mechanisms are mediated by both intrinsic and extrinsic factors, such as cytoskeletal elements,^7,8 histone modifications,^9,10 vesicle trafficking,¹¹ and cell adhesion molecules.¹² Together, these factors ensure a balanced hematopoietic output and reserve capacity during stress.

Despite the substantial erythropoietic reserve, anemia remains the most common presenting symptom of myelodysplastic syndrome (MDS), a clonal myeloid malignancy characterized by ineffective hematopoiesis and myelodysplasia.^13,14 Notably, the genetic landscape of MDS is enriched for mutations affecting epigenetic regulators, including the cohesin complex.^15,16 Mutations in STAG2 account for the majority of cohesin alterations in myeloid malignancies and are particularly enriched in MDS and secondary acute myeloid leukemia.^16,17 Studies from our group and others have highlighted key clinical and pathological features specific to STAG2-mutant MDS, including more severe anemia, erythroid lineage nuclear:cytoplasmic dyssynchrony, and erythropoietin resistance, pointing toward a critical but incompletely understood role of STAG2 in both normal erythropoiesis and MDS pathogenesis.^{18, 19, 20}

Although key erythroid transcriptional networks, hemoglobin switching, and immunophenotypic transitions have been well characterized,¹⁰ the mechanisms that confer lineage specificity remain incompletely understood. Transcription factors (TF) such as GATA1 are broadly expressed across hematopoietic lineages yet direct distinct differentiation programs in different cellular contexts. GATA1 is expressed in a variety of stem cell and progenitor populations, yet is also critical in lineage-specific erythroid,^21,22 megakaryocytic,²³ eosinophilic,²⁴ and mast cell²⁵ differentiation. This raises fundamental questions about how lineage specificity is achieved. Although previous work has identified combinatorial TF binding and cooperativity with lineage-specific cofactors, emerging evidence suggests that TFs also rely on epigenetic context, such as chromatin accessibility and remodeling complexes.^{26, 27, 28, 29, 30, 31, 32} Notably, erythropoiesis requires epigenetic remodeling at multiple stages, particularly during terminal erythroid differentiation (TED), which involves progressive heterochromatin formation characterized by peripheralization of H3K9me3 and topologically associated domain disruption.^6,33,34 Given that STAG2 loss disrupts chromatin accessibility, TF activity, and lineage differentiation, we hypothesized that STAG2 acts as a chromatin accessibility rheostat that enforces the erythroid-specific GATA1 cistrome.¹⁹ Therefore, loss of STAG2 would be expected to produce accessibility-driven Gata1 retargeting to megakaryocyte loci. This mechanism has not been defined previously and would explain dyserythropoiesis in STAG2-mutant MDS. To validate this hypothesis, we leveraged our transgenic Stag2^Δ mouse model, human models of erythropoiesis, and data from patients with STAG2-mutant MDS, to examine how altered chromatin accessibility influences erythroid TF function and transcriptional output.

Methods

Mice

All animals involved in this study were housed at Columbia University Irving Medical Center (CUIMC) and were used for experimental purposes in accordance with approved Institutional Animal Care and Use Committee protocols. Mx1-Cre Stag2^fl/fl mice were previously described.¹⁹ Recombination was induced by 4 intraperitoneal doses of polyinosinic:polycytidylic acid (pIpC; 20 mg/kg) starting at age 6 weeks. All mice in this study received pIpC. Terminal end points for all mice were decided to be no earlier than 4 weeks after the last dose of pIpC. Stag2 excision was verified at the time of terminal analyses, as previously described.¹⁹ Mice were assigned to experiments, agnostic of sex.

Patients

This study was approved by the CUIMC Institutional Review Board (AAAU4817, AAAU8073). Diagnostic bone marrow hematopathology reports from patients with MDS treated at CUIMC were analyzed for descriptors of erythrodysplasia. For a subset of patients, digital peripheral blood smear images were analyzed using CellaVision. Patient characteristics are described in supplemental Table 1, available on the Blood website.

Murine liquid culture

A total of 5000 erythroid progenitors (EryPs) were sorted into 100 μL of fluorescence-activated cell sorting buffer, washed once with phosphate-buffered saline, and resuspended in 200 μL of StemSpan Serum-Free Expansion Media II (catalog no. 09655, StemCell Technologies) supplemented with 1% penicillin/streptomycin, rm-stem cell factor (100 ng/mL; Peprotech), and rh-erythropoietin (4 U/mL; Peprotech).^35,36 After 48 hours, cells were collected and analyzed using flow cytometry for absolute counts and surface markers.

Murine colony assays

A total of 20 000 (erythroid-directing; M3334) or 10 000 (pan-myeloid; M3434) EryPs were seeded in duplicates into 1 mL MethoCult (StemCell Technologies) supplemented with 1% penicillin/streptomycin and no additional cytokines. Colonies were scored after 48 hours (M3334) using brightfield/immunofluorescent microscopy or flow cytometry on day 8 (M3434).

Human CD34⁺ hematopoietic stem and progenitor cell (HSPC) liquid culture and STAG2 knockdown

CD34⁺ cells were isolated from umbilical cord blood, transduced with short hairpin RNA, and cultured as previously described.³⁷ Cells were collected on days 5 to 7, 11, and 15 of culture, stained as previously described, and analyzed using flow cytometry for cell surface markers.^38,39 Additional details are provided in the supplemental Methods.

RNA-sequencing data analysis

The normalized count matrix was downloaded from GSE138006¹⁹ and analyzed as described in the supplemental Methods.

Single cell assay for transposase-accessible chromatin (scATAC) sequencing

HSPCs (lineage-negative) or Ter119⁺ cells were sorted, nuclei were isolated, and prepared using the 10X Genomics workflow, sequenced, and analyzed as described in the supplemental Methods.

Gata1 and Stag2 CUT&RUN

Cleavage under targets and release using nuclease (CUT&RUN, EpiCypher) was performed on sorted EryPs and analyzed as described in the supplemental Methods.

Hi-C

Hi-C was performed on sorted EryPs (Arima High Coverage Hi-C Kit) and analyzed as described in the supplemental Methods.

Statistical analysis

All experiments are reported as N or the number of data points, which indicate the number of independent biological replicates. Statistical significance was evaluated using GraphPad Prism (v.10). Data distribution was assumed to be normal but not formally tested. The robust regression and outlier removal (ROUT, Q = 1%) method was used to evaluate for the presence of and to remove outliers (Figures 1H-K and 2I; supplemental Figures 2B-E,H,L,N and 4F-Q).⁴⁰

Figure 1.

Loss of Stag2 perturbs the balance of erythroid-megakaryocytic fate. (A) Schematic diagram of scATAC-seq for lineage-negative HSPCs. Figure 1A created with BioRender.com. (B) Uniform manifold approximation and projection (UMAP) visualization of scATAC-seq from sorted Stag2^WT and Stag2^Δ HSPCs (N = 3 for both conditions), filtered for inferred cells along erythroid differentiation with cell type annotation (left) and genotypes labeled (right). (C) Quantification of inferred cell types identified in panel B. (D-G) Erythroid module (D), Gypa (E), megakaryocyte module (F), Pf4 (G) gene-body accessibility over erythroid differentiation pseudotime. (H-J) Frequency and representative flow plots of EryP (H), MkP (I), and PreMegE (J) from murine bone marrow after ammonium-chloride-potassium lysis stained with antibodies against lineage markers (see supplemental Table 2), Kit, CD55, CD41, CD105, and CD150. (K) Ratio of EryP to MkP. Each point represents data from an individual mouse; bar graphs depict mean ± standard deviation (SD). ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001 for comparison of data from Stag2^WT and Stag2^Δ assessed using unpaired 2-tailed t test with Welch correction (α = 0.05) after using the ROUT (Q = 1%) method to remove outliers. FSC, forward scatter channel; HSC, hematopoietic stem cell; MPP, multipotent progenitor.

Figure 2.

Stag2^Δ impairs erythroid commitment and TED. (A) Schematic diagram of scATAC-seq for Ter119⁺ erythroblasts. Figure 2A created with BioRender.com. (B) UMAP visualization of scATAC-seq from sorted Stag2^WT and Stag2^Δ Ter119⁺ erythroblasts (N = 3 for both conditions) with cell type annotation (left) and genotypes labeled (right). (C) Quantification of inferred cell types identified in panel B. (D-G) TED module (D), Tfrc (E), megakaryocyte differentiation module (F), Itga2b (G) gene-body accessibility over TED pseudotime. (H) GSEA plot of the TED program (left) and megakaryocyte differentiation program (right). The plot represents the ranked, ordered, nonredundant list of genes with those on the left (red) having a higher correlation with Stag2^Δ, whereas those on the right (blue) have a higher correlation with Stag2^WT. The vertical black lines indicate the position of each gene in the gene set. (I) Frequency of basophilic erythroblasts, PolyE, and OrthoE relative to total Ter119⁺ erythroblasts from bone marrow of Stag2^WT and Stag2^Δ mice after ACK lysis, as determined by flow cytometry. Each point represents data from an individual mouse, bar graphs depict mean ± SD. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001 for comparison of data from Stag2^WT and Stag2^Δ assessed using unpaired 2-tailed t test with Welch correction (α = 0.05) after using the ROUT (Q = 1%) method to remove outliers. BasoE, basophilic erythroblasts; FDR, false discovery rate; NES, normalized enrichment score; ns, not significant; PolyE, polychromatic erythroblasts.

Results

Stag2 coordinates chromatin accessibility during erythroid lineage commitment

To map the dynamic changes in chromatin accessibility throughout early erythroid fate commitment, we performed scATAC sequencing on lineage-negative HSPCs from wild-type and Stag2-deficient bone marrow, henceforth referred to as Stag2^WT and Stag2^Δ (Figure 1A). Cell annotation was performed using lineage-defining genes to identify lineage-specific differentiation trajectories (supplemental Figure 1A-B). Erythroid differentiation was defined by ATAC peak profiling of all cells along the trajectory from HSPCs to lineage-specified EryPs (Figure 1B; supplemental Figure 1C). Consistent with data from previous studies, we observed that Stag2^Δ HSPCs have defective erythroid commitment and differentiation (Figure 1B-C).¹⁹ Moreover, our prior transcriptomic analyses identified a delayed erythroid pseudotime, which was also evident by scATAC sequencing (scATAC-seq) based on the proportions of inferred cell types (Figure 1C).

To determine Stag2-dependent chromatin accessibility changes during erythroid specification, we generated a pseudotime of erythroid differentiation and queried gene-body accessibility at various lineage-defining genes. We observed 2 concurrent chromatin defects in the setting of Stag2 loss: (1) delayed or impaired opening at key erythroid genes (Figure 1D-E) and (2) failure to restrict accessibility at megakaryocyte genes normally silenced during erythropoiesis (Figure 1F-G).

Loss of Stag2 skews MEP differentiation toward megakaryopoiesis

Because previous assessments of HSPCs had not identified a chromatin bias toward megakaryopoiesis, we interrogated lineage-specified megakaryocyte progenitors (MkPs) and EryPs, as previously described, to confirm whether this cell-intrinsic observation resulted in a population skew (supplemental Figure 2A).^41,42 We observed a significant overall reduction in total megakaryocyte-erythroid progenitors (MEPs), consistent with prior work (supplemental Figure 2B).¹⁹ Stag2 loss induced a significant contraction of EryPs and an expansion of MkPs and premegakaryocyte-erythroid progenitors (Figure 1H-J; supplemental Figure 2C-E). To understand the balance between lineage-specified erythroid and megakaryocytic fate, we calculated a ratio of EryP to MkP and observed that Stag2 loss results in a reduction in this ratio, indicating a shift away from erythroid and toward megakaryocytic fate (Figure 1K). Overall, loss of Stag2 impaired erythroid fate specification because of unbalanced output of lineage-specified progenitors.

Stag2 is required for TED and enucleation

To understand whether altered lineage specification within Stag2^Δ HSPCs affected terminal lineage commitment, we profiled Ter119⁺ and Kit⁻CD41⁺ cells to capture irreversibly erythroid- and megakaryocyte-committed cells (supplemental Figure 2F-G). Consistent with our prior findings,¹⁹ Stag2^Δ mice had reduced Ter119⁺ erythroblasts (supplemental Figure 2H-I). We also observed a previously unappreciated expansion of Kit⁻CD41⁺ megakaryocytes (supplemental Figure 2J-K). Given the fate imbalance at the level of lineage-specified progenitors, we sought to determine whether the effect of Stag2 loss on chromatin accessibility was restricted to lineage specification or if chromatin accessibility was also perturbed during TED (Figure 2A). In contrast to the HSPC scATAC-seq data, we observed separate and parallel distributions of Stag2^WT and Stag2^Δ cells in Ter119⁺ scATAC-seq, with no quantitative changes in inferred cell types (Figure 2B-C; supplemental Figure 3A). To ascertain which differences accounted for the distribution of Stag2^WT and Stag2^Δ cells, we examined the differentially accessible genes and noted loss of accessibility at erythroid genes and gain of accessibility at megakaryocyte genes (supplemental Figure 3B). As with our HSPC scATAC-seq analysis, we queried gene-body accessibility throughout TED and observed 2 different synchronous states: (1) key erythroid differentiation genes that have impaired accessibility (Figure 2D-E) and (2) aberrant maintenance of megakaryocyte genes (Figure 2F-G).^{43, 44, 45} We orthogonally observed these concurrent chromatin accessibility defects by gene set enrichment analysis (GSEA, Figure 2H).

Based on these observations, we hypothesized that Stag2 loss may impair TED. Analysis of individual erythroblast populations (supplemental Figure 2F) revealed that Stag2^Δ mice had more basophilic and polychromatic erythroblasts and fewer orthochromatic erythroblasts (OrthoE) (Figure 2I; supplemental Figure 2L-M), with no change in proerythroblast frequency (supplemental Figure 2N-O). Because the polychromatic to orthochromatic erythroblast is hallmarked by nuclear condensation followed by enucleation, we examined genes facilitating these processes.⁴⁶ We observed a striking negative enrichment of these genes in Stag2^Δ erythroblasts (supplemental Figure 4A). Imaging flow cytometry confirmed defective nuclear condensation, showing a significantly increased nuclear to cytoplasmic ratio in Stag2^Δ OrthoEs with reduced enucleation (supplemental Figure 4B-E). These defects, including those observed among progenitors, translated to ineffective erythropoiesis with mild anemia, significant anisocytosis, moderate reticulocytosis, increased nucleated RBCs, elevated serum erythropoietin (EPO), and compensatory splenic erythropoiesis (supplemental Figure 4F-R). Collectively, these findings indicate a pan-erythropoietic requirement for Stag2.

Stag2 loss alters chromatin accessibility and cell type–specific transcriptional fidelity

Because our chromatin and immunophenotypic data pointed toward megakaryocytic skewing, we next aimed to determine whether this effect was evident during erythroid fate specification at the EryP cell stage. At this stage, megakaryocyte differentiation programs are physiologically repressed to facilitate irreversible erythroid commitment.^{47, 48, 49} To determine the role of chromatin accessibility at this critical stage, we interrogated accessibility modules in EryPs that facilitate erythroid differentiation and repress megakaryocyte differentiation by pseudobulking our scATAC-seq data. Among pseudobulked EryPs, we identified differential accessibility at erythroid (eg, Gypc, Hba-a1, Hbq1b) and megakaryocyte genes (eg, Dkk2, Fli1; Figure 3A). Orthogonally, we observed through GSEA that erythroid ontology programs were significantly negatively enriched, whereas megakaryocyte programs were positively enriched (Figure 3B; supplemental Figure 5A).

Figure 3.

Stag2^Δ results in differentially accessible and expressed erythroid and megakaryocyte programs in EryPs because of altered engagement of lineage-specific Gata1 cistromes. (A) Volcano plot representing statistical significance (−log₁₀ FDR) vs magnitude of DNA accessibility changes (log₂ Fold Change) from a comparison of inferred Stag2^WT and Stag2^Δ EryPs with relevant erythroid and megakaryocyte genes (bold and in yellow boxes). (B) Table of selected GSEA pathways enriched in pseudobulked scATAC-seq with erythroid related pathways (in red) and megakaryocyte-related pathways (in blue). (C) Schematic diagram of bulk RNA-seq workflow. Figure 3C created with BioRender.com. (D) Volcano plot representing statistical significance (−log₁₀ FDR) vs magnitude of RNA expression changes (log₂ Fold Change) from a comparison of Stag2^WT and Stag2^Δ EryPs with relevant erythroid and megakaryocyte genes (bold and in yellow boxes). (E) Table of selected GSEA pathways enriched in bulk RNA-seq with erythroid related pathways (in red) and megakaryocyte-related pathways (in blue). (F-G) Lollipop plot representing TF enrichment among downregulated (F) and upregulated (G) genes assessed via bulk RNA-seq of Stag2^WT and Stag2^Δ EryPs using ChIP-X Enrichment Analysis 3. (H-I) GSEA plot of the Gata1 erythroid (left) and Gata1 megakaryocyte (right) cistromes as determined by RNA-seq (H) and pseudobulked scATAC-seq (I). The plot represents the ranked, ordered, nonredundant list of genes with those on the left (red) having a higher correlation with Stag2^Δ, whereas those on the right (blue) have a higher correlation with Stag2^WT. The vertical black lines indicate the position of each gene in the gene set. (J-K) Track plot of RNA-seq and pseudobulked scATAC-seq for Hba-a1 (J) and Itga2b (K). FDR, false discovery rate; HSC, hematopoietic stem cell; HSPC, hematopoietic stem/progenitor cell; MPP, multipotent progenitor; NES, normalized enrichment score.

To assess whether the noted DNA accessibility perturbations resulted in altered transcriptional output, we analyzed our previously existing bulk RNA sequencing (RNA-seq) data from Stag2^WT and Stag2^Δ EryPs (Figure 3C).¹⁹ Indeed, we observed numerous erythroid (eg, Hba-a1, Hbb-b1) and megakaryocyte genes (eg, Plek, F11r) that were differentially expressed with Stag2 loss (Figure 3D). We then repeated GSEA on our RNA-seq data and observed concordant results with significant negative enrichment of erythroid programs and positive enrichment of megakaryocyte transcriptional programs (Figure 3E; supplemental Figure 5B). These findings were orthogonally validated using Gene Ontology (GO) analysis (supplemental Figure 5C-D). Together, our pseudobulked scATAC-seq and RNA-seq data suggest that Stag2 loss results in impaired induction and silencing of erythroid and megakaryocytic programs, respectively.

Gata1 lineage specificity is disrupted by altered chromatin remodeling

To investigate the mechanism by which alterations in DNA accessibility translate into transcriptional changes, we queried the significant genes identified from our RNA-seq data for TF enrichment.⁵⁰ Among the downregulated genes, Gata1, Maz, and CCCTC-binding factor were among the top hits (Figure 3F). Interestingly, Gata2, Ets1, Tal1, and Gata1 were among the top hits within TFs enriched in upregulated genes (Figure 3G). Although Gata1 was identified as one of the top enriched TFs in both downregulated and upregulated genes, the gene targets representing increased and decreased Gata1 engagement were nonoverlapping (supplemental Figure 5E). To assess whether the Gata1 targets represented unique cistromes, we mapped reference Gata1 chromatin immunoprecipitation (ChIP)-seq data sets from erythroid and megakaryocytic cells to create 2 gene sets that are bound and positively regulated by Gata1 in a lineage-specific manner.⁵¹ Differentially expressed genes in Stag2^Δ EryPs were negatively enriched for Gata1 erythroid targets, whereas megakaryocyte-specific Gata1 targets were positively enriched (Figure 3H). We observed the same trend when queried against pseudobulked scATAC-seq EryPs (Figure 3I).

To directly connect Stag2-dependent accessibility changes at Gata1 targets with transcriptomic changes and assess the subset of genes whose transcription was sensitive to accessibility changes, we integrated our RNA-seq and pseudobulked scATAC-seq data sets. Genes involved in erythroid and megakaryocyte ontogeny had concordant changes in expression and accessibility (Figure 3J-K; supplemental Figure 5F). Subsequently, we queried TF enrichment among the integrated genes and observed overrepresentation of Gata1 erythroid targets among genes losing accessibility and expression, whereas Gata1 megakaryocyte targets were enriched among genes gaining both accessibility and expression (supplemental Figure 5G). We also detected overrepresentation of Klf1 (downregulated), Tal1 (downregulated), and Fli1 (upregulated), lineage-specific TFs downstream of Gata1 (supplemental Figure 5G). Collectively, these data point toward altered Gata1 lineage specificity in the setting of Stag2 loss.

Lineage-specific Gata1 cistromes are reprogrammed by chromatin state

Next, we asked whether altered Gata1 stoichiometry affected cistrome engagement. To test this hypothesis, we assayed Gata1 expression levels and found no difference at either the RNA or protein level in Stag2^Δ EryPs (Figure 4A-C). Subsequently, we hypothesized that altered chromatin accessibility results in aberrant Gata1 chromatin occupancy. To evaluate Gata1 chromatin occupancy, we conducted CUT&RUN for Gata1 in Stag2^WT and Stag2^Δ EryPs. Overall, we identified 80 721 and 54 251 Gata1 bound peaks in Stag2^WT and Stag2^Δ EryPs, respectively (Figure 4D). Among the top genes that gained Gata1 occupancy were Mx1, Dip2c, and Rhob, whereas top genes that lost Gata1 occupancy included Pdgfd, Noct, Tpk1, and Ebp42 (Figure 4E-G). GO analysis of differentially bound genes revealed overrepresentation of erythroid- and megakaryocyte-related pathways among the downregulated and upregulated genes, respectively (Figure 4H-I). Together, these analyses provide evidence that Stag2 deficiency alters Gata1 target occupancy in EryPs, which we hypothesized may lead to ectopic maintenance of accessible megakaryocyte targets.

Figure 4.

Stag2^Δ alters Gata1 occupancy at lineage-specific targets. (A) Schematic diagram of Gata1 expression (RNA/protein) and CUT&RUN workflows. Figure 4A created with BioRender.com. (B) Bar plot represents Gata1 expression levels as determined by qPCR in EryPs. (C) Western blot of EryP proteins with antibodies directed against Gata1/Gata1s and vinculin. Gel images were cropped to retain the regions corresponding to Gata1/Gata1s and vinculin, rotated to ensure proper alignment, and outlined with a black line. Gel images were not modified in any other way after acquisition. (D) Heat map of Gata1 binding sites (left) around peak center (center) and venn diagram (right) depicting the number of Gata1 bound peaks called in Stag2^WT and Stag2^Δ EryPs. (E) Volcano plot representing statistical significance (−log₁₀ FDR) vs magnitude of Gata1 occupancy changes (log₂ Fold Change) from a comparison of Stag2^WT and Stag2^Δ EryPs with relevant erythroid and megakaryocyte genes bolded and in yellow boxes. Only the largest peak per gene is labeled. (F-G) Track plots displaying Gata1 occupancy at Epb42 (F) and Rhob (G). (H-I) Bar plot of enriched GO terms among genes losing (H) and gaining (I) Gata1 occupancy in Stag2^Δ EryPs. (J) Plot of magnitude of Gata1 occupancy changes vs magnitude of DNA accessibility changes in Stag2^WT and Stag2^Δ EryPs with relevant erythroid and megakaryocyte genes bolded and in yellow boxes. (K) Venn diagram of overlap of RNA expression, DNA accessibility, and Gata1 occupancy changes per gene. Table to the right represents GO terms enriched among the genes that experience concordant expression, accessibility, and occupancy changes. (L) Track plots displaying DNA accessibility (ATAC), Gata1 occupancy, and RNA expression at Tfrc (left) and Fli1 (right). Each point represents data from an individual mouse, bar graphs depict mean ± SD. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001 for comparison of data from Stag2^WT and Stag2^Δ assessed using unpaired 2-tailed t test with Welch correction (α = 0.05). ns, not significant; qPCR, quantitative polymerase chain reaction.

To determine whether differential accessibility functionally results in megakaryocyte Gata1 target engagement and to assess the subset of genes whose Gata1 occupancy was accessibility-sensitive, we integrated our Gata1 CUT&RUN and pseudobulked scATAC-seq data. We identified numerous erythroid genes and megakaryocyte genes that had concordant losses or gains in accessibility and Gata1 occupancy (Figure 4J). Then, to identify a core set of genes with concordant RNA, ATAC, and Gata1 occupancy changes, we integrated all 3 of our sequencing data sets. We identified 431 genes (eg, Tfrc and Hba-a1) that lost, and 611 genes (eg, Fli1, Cd9) that gained accessibility, Gata1 occupancy, and expression (Figure 4K-L). GO analysis of these consensus gene sets identified erythroid and megakaryocyte pathways among the downregulated and upregulated gene sets, respectively (Figure 4K). Because engagement of lineage-specific cistromes by Gata1 may be accompanied by altered expression of other hematopoietic TFs, corepressors, and/or splice isoforms, we queried RNA expression of these factors and validated altered targets using western blot analysis. We observed no significant alterations in RNA expression of heptad TFs (eg, Gata2 and Erg), cofactors (eg, Zfpm1), corepressors (eg, Eto2 and Ctbp1), or protein expression of Gata1s, except for the megakaryocyte-specific TF Fli1, which was upregulated after loss of Stag2 (Figure 4C; supplemental Figure 6A-B). Indeed, Fli1 protein levels were upregulated, and the Fli1 consensus motif was enriched among Stag2^Δ Gata1 peaks (supplemental Figure 6C-E). Collectively, these data point toward a Stag2-dependent epigenetic mechanism that fine-tunes Gata1 chromatin binding to enable concerted upregulation of an erythroid-specific differentiation program.

Stag2 fine-tunes Gata1 occupancy to facilitate erythroid transcriptional programs

Because of prior data supporting a direct interaction between Stag2 and discrete hematopoietic TFs (eg, Runx1, Erg), we conducted a coimmunoprecipitation to ascertain whether Stag2 and Gata1 directly bind each other.⁵² We did not detect a direct interaction between Stag2 and Gata1 in EryPs (Figure 5A-B). To determine the specific effects of Stag2 on Gata1 occupancy, we conducted Stag2 CUT&RUN in EryPs and intersected Stag2 and Gata1 peaks (Figure 5C-D). Genome-wide overlap analysis demonstrated significant colocalization of Stag2 and Gata1 binding (60.5%) (Figure 5E). When these peaks were dichotomized into the erythroid or megakaryocyte cistrome, we observed that Stag2 binding was strongly enriched at erythroid-specific Gata1 sites (odds ratio [OR], 6.25; P = 1.1E−244) (Figure 5F). Because of the marked effect of Stag2 loss on accessibility, we next intersected Stag2 peaks with ATAC peaks and observed that ∼90% of accessible regions colocalized with Stag2 (Figure 5G-H). When we dichotomized these regions into peaks gaining or losing accessibility, we observed enrichment of Stag2 occupancy at regions losing accessibility (66%), whereas Stag2 occupancy was comparatively depleted at regions gaining accessibility (57%; OR, 0.69; P = 1.23E−276) after Stag2 loss (Figure 5I). Given the marked overlap of Stag2 with regions losing accessibility and Gata1 occupancy, we hypothesized that the presence of Stag2 may positively predict transcriptional output. Indeed, we observed that the presence of Stag2 was moderately associated (r = 0.505) with transcriptional output when Gata1 was a co-occupant, and that this association was attenuated with the loss of Stag2 (r = 0.303), supporting a fine-tuning role for Stag2 (Figure 5J).

Figure 5.

Stag2 indirectly fine-tunes Gata1 binding through maintaining accessibility at erythroid-specific genes. (A) Schematic diagram of Stag2 coimmunoprecipitation and CUT&RUN workflows. Figure 5A created with BioRender.com. (B) Western blot of immunoprecipitated EryP proteins after incubation with an anti-Stag2 antibody. Gel images were cropped to retain the regions corresponding to Stag2 and Gata1, rotated to ensure proper alignment, rearranged to place input, immunoglobulin G, and anti-Stag2 samples adjacent to each other, and outlined with a black line. Gel images were not modified in any other way after acquisition. (C) Heat map of Stag2 binding sites in Stag2^WT and Stag2^Δ EryPs. (D) Heat map of Stag2 binding sites at Gata1 peaks in Stag2^WT and Stag2^Δ EryPs. (E) Pie chart of the fraction of Gata1 peaks colocalizing with Stag2 peaks (calculated as Stag2-Gata1 co-occupant peaks divided by total Gata1 peaks). (F) Bar plot of the percentage of erythroid- and megakaryocyte-specific Gata1 cistromes that colocalize with Stag2 binding (calculated as lineage-specific Stag2-Gata1 co-occupant peaks divided by lineage-specific Gata1 peaks). Table to the right represents OR of Stag2 enrichment within erythroid vs megakaryocyte Gata1 cistromes. (G) Heat map of Stag2 binding sites at ATAC peaks in Stag2^WT and Stag2^Δ EryPs. (H) Pie chart of the fraction of ATAC peaks colocalizing with Stag2 peaks (calculated as Stag2-ATAC peaks divided by total ATAC peaks). (I) Bar plot of the percentage of upregulated and downregulated ATAC peaks that colocalize with Stag2 binding. Table to the right represents the OR of Stag2 enrichment within upregulated vs downregulated ATAC peaks. (J) Plot of log transformed gene expression magnitude (log₁₀TPM) and Gata1 binding strength (log₁₀ mean bigWig signal ± 2 kb from transcription start site [TSS]) at genes with Stag2-Gata1 co-occupancy. Correlation of gene expression and Gata1 occupancy was calculated using Pearson correlation for Stag2^WT (top) and Stag2^Δ (bottom) EryPs. (K) Schematic diagram of Hi-C workflow. Figure 5K created with BioRender.com. (L) Plot of A vs B compartmentalization at 50 kb resolution in Stag2^WT and Stag2^Δ EryPs. E1 represents the first eigenvector of the observed/expected cis-contact matrixes (see supplemental Methods) (M) Quantification of the number of gene bodies (TSS) present in each A/B compartment based on compartment type and differential expression in Stag2^WT vs Stag2^Δ EryPs. Numbers above each bar represent the percentage of genes that are transcriptionally unchanged within each compartment switch category. (N) Venn diagram depicting the number of loops called in Stag2^WT and Stag2^Δ EryPs. (O-P) Trackplots displaying detected loops, Stag2 occupancy, and Gata1 occupancy at Klf1 (left) and Fli1 (right). Black arrows indicate the relevant loop, black rectangles indicate loop anchors, and yellow highlights the corresponding gene body. ORs were calculated using a 2-tailed Fisher exact test (α = 0.05).

To understand the mechanism by which Stag2 orchestrates chromatin accessibility and Gata1 cistrome specificity, we conducted Hi-C in Stag2^WT and Stag2^Δ EryPs (Figure 5K). Despite the canonical role of Stag2-cohesin in genome organization, Hi-C revealed minimal changes in topologically associated domain insulation or compartmentalization, demonstrating that the primary mechanism of STAG2 in EryPs is at local accessibility, not higher order nuclear topology (supplemental Figure 6F-J).^19,34 Although we identified 50 000 compartment switches, these overwhelmingly did not involve differentially expressed genes (Figure 5L-M; supplemental Figure 6K). At 10 kb resolution, we identified nearly 60 000 loops (Figure 5N). However, when we fractionated for loops involving promoters, we observed a weak correlation with transcriptional output (ρ = 0.103; supplemental Figure 6L). When querying for erythroid- and megakaryocyte-specific Gata1 targets, we also observed that upon Stag2 loss, loops at canonical erythroid loci like Klf1 were lost. At these sites, we observed loss of the Stag2 anchor and reduced Gata1 occupancy. In contrast, loops at megakaryocyte genes such as Fli1 were gained (Figure 5O-P). These changes are consistent with failed maintenance of erythroid-specific Gata1 programs and aberrant persistence of megakaryocyte-specific Gata1 programs. Collectively, these data indicate that Stag2 creates the epigenetic landscape that guides Gata1 occupancy and stabilizes transcriptional output at biologically relevant genes, rather than directly recruiting Gata1.

Stag2-deficient EryPs exhibit aberrant lineage fidelity

Finally, to understand the functional implications of altered lineage-specific Gata1 cistrome engagement on lineage fidelity and differentiation capacity, we sorted Stag2^WT and Stag2^Δ EryPs and seeded them into liquid media supplemented with stem cell factor (SCF) and EPO (Figure 6A). We observed fewer cells overall (supplemental Figure 7A-C), and Stag2^Δ EryPs generated fewer Ter119⁺CD71⁺ cells and more CD41⁺CD42d⁺ megakaryocytes (Figure 6B-D; supplemental Figure 7D-E). To validate these findings orthogonally, we assessed colony forming capacity of Stag2^WT and Stag2^Δ EryPs in either erythroid-directing methylcellulose or pan-myeloid–directing methylcellulose (Figure 6E). In erythroid-directing methylcellulose, we observed fewer erythroid colonies generated by Stag2^Δ EryPs (Figure 6F). In pan-myeloid methylcellulose, we observed a reduction in Ter119⁺CD71⁺ cells and an increase in CD41⁺ cells (Figure 6G-H; supplemental Figure 7G-I).

Figure 6.

Stag2-deficient EryPs have diminished erythroid output and increased megakaryocyte output, which is partially reversed with Fli1 knockdown. (A) Schematic diagram of liquid culture differentiation workflow. (B-C) Frequency of Ter119⁺CD71⁺ (B) and CD41⁺42d⁺ (C) cells generated by EryPs after 48 hours in liquid culture as determined by flow cytometry. Figure 6B-C created with BioRender.com. (D) Representative flow cytometry plots identifying Ter119⁺CD71⁺ and CD41⁺42d⁺ cells stained with antibodies against Ter119, CD71, CD41, CD42d, and Kit. (E) Schematic diagram of methylcellulose assay workflow. (F) Absolute count of the number of erythroid colonies generated by EryPs after 48 hours in erythroid-directing methylcellulose identified by brightfield microscopy (left) and immunofluorescence (right). (G) Frequency of Ter119⁺CD71⁺ (left) and CD41⁺ (right) cells generated by EryPs after 8 days in pan-myeloid–directing methylcellulose as determined by flow cytometry. (H) Representative flow cytometry plots identifying Ter119⁺CD71⁺ and CD41⁺ cells stained with antibodies against CD11b, Gr-1, CD41, Ter119, and CD71. (I) Schematic diagram of antisense oligonucleotides (ASO) knockdown of Fli1 in EryPs undergoing erythroid differentiation. Figure 6I created with BioRender.com. (J) Frequency of Ter119⁺CD71⁺ (left) and CD41⁺42d⁺ cells (right) generated by EryPs treated with scramble and Fli1 ASO after 72 hours in liquid culture as determined by flow cytometry. Statistical significance was assessed using a 2-way analysis of variance (α = 0.05) for the presented comparisons only. (K) Representative flow cytometry plots identifying Ter119⁺CD71⁺ and CD41⁺42d⁺ cells stained with antibodies against Ter119, CD71, CD41, and CD42d. Each point represents data from an individual mouse; bar graphs depict mean ± SD. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001 for comparison of data from Stag2^WT and Stag2^Δ assessed using unpaired 2-tailed t test with Welch correction (α = 0.05) unless otherwise stated. BF, brightfield microscopy; IF, immunofluorescence; IL, interleukin; Scr, scramble.

To identify candidates that could ameliorate Stag2^Δ EryP lineage output, we examined TFs downstream of Gata1, given that overexpression of Gata1 has previously been demonstrated to result in a lethal anemia and death of erythroid cells.⁵³ Given the enrichment of Fli1 motifs at Stag2^Δ Gata1 sites, we tested whether reducing Fli1 could restore lineage balance. Using our liquid culture system, we knocked down Fli1 using antisense oligonucleotides and observed complete correction of the augmented megakaryocyte output, with minimal impact on erythroid output in Stag2^Δ EryPs, underscoring that erythroid differentiation is more complex than the expression of a single TF (Figure 6I-K; supplemental Figure 7J-N). Collectively, these data provide a link between the transcriptional, accessibility, and Gata1 occupancy aberrancies to impaired Stag2^Δ EryP function, and demonstrate that EryP function can be partially ameliorated by Fli1 knockdown.

Human STAG2-deficient HSPCs undergo dysplastic erythropoiesis and recapitulate essential features of the murine model

Because STAG2 is a recurrent mutational target in MDS, we sought to determine the generalizability of our murine observations of dyserythropoiesis in human cells and patients with MDS. Using human CD34⁺ HSPCs isolated from umbilical cord blood, we conducted short hairpin RNA knockdown of STAG2 (Figure 7A-B). Flow cytometric analyses identified delayed upregulation of CD36 and gain of CD235, indicating delayed erythroid specification and terminal commitment (Figure 7C-D).^39,54 We also observed that STAG2^KD HSPCs progressed through TED less efficiently and had reduced enucleation (Figure 7E-F). To determine whether the features of erythroid dysplasia were conserved, we conducted cytospins of differentiating STAG2^KD HSPCs and observed features of myelodysplasia such as multinucleated erythroblasts and erythrocytes, unique pathological morphologies described in MDS (Figure 7G-H).¹³ To validate our molecular observations, we leveraged publicly available bulk RNA- and ATAC-seq from STAG2^KD HSPCs (Figure 7I).⁵⁵ Similar to our mouse model, we observed downregulation of erythroid and upregulation of megakaryocyte differentiation programs, dysregulation of lineage-specific GATA1 cistromes, enrichment of GATA1 motifs among downregulated and upregulated ATAC peaks (greater at downregulated) and FLI1 motifs among upregulated ATAC peaks, and detected differential enrichment of lineage-specific TFs among genes experiencing concordant RNA and ATAC changes (Figure 7J-M; supplemental Figure 8A-G). Genome-wide overlap analysis and lineage-specific fractionation demonstrated that STAG2 binding was strongly enriched at erythroid-specific GATA1 sites (OR, 6.31; P = 3.8E−53), despite comparing across different cell types (supplemental Figure 8H-I).

Figure 7.

Human CD34⁺ STAG2 knockdown model of erythropoiesis recapitulates phenotypic and molecular features with Stag2^Δ mice. (A) Schematic diagram of umbilical cord blood–derived CD34⁺ HSPC short hairpin RNA (shRNA) knockdown and erythroid differentiation. Figure 7A created with BioRender.com. (B) Quantification of STAG2 RNA expression after shRNA delivery as determined by quantitative polymerase chain reaction. (C) Representative flow cytometry histogram of CD36 cell surface expression of differentiating HSPCs on day 6 of erythroid differentiation. (D) Representative flow cytometry histogram of CD235 cell surface expression of differentiating HSPCs on day 7 of erythroid differentiation. (E) Representative flow cytometry plots of α4-integrin and band3 expression on D7 of in-vitro differentiation. (F) Representative flow cytometry plots of Syto16 positivity on D15 of in-vitro differentiation. (G) Representative cytospins of shLuciferase and shSTAG2 cells on D11 and D15 of in-vitro differentiation. Arrows indicate dysplastic cells. Original magnification ×40. (H) Bar plots of the quantification of the number of dysplastic cells per high power field (HPF), with the table underneath representing the number of dysplastic cells identified divided by the total number of cells summated across all HPFs. (I) Schematic diagram of umbilical cord blood–derived CD34⁺ HSPC siRNA knockdown and RNA/ATAC-seq. (J) Volcano plot representing statistical significance (−log₁₀FDR) vs magnitude of RNA expression changes (log₂FoldChange) from a comparison of STAG2^Scramble and STAG2^KD HSPCs with relevant erythroid and megakaryocyte genes bolded and in yellow boxes. (K) Volcano plot representing statistical significance (−log₁₀ FDR) vs magnitude of accessibility changes (log₂ Fold Change) from a comparison of STAG2^Scramble and STAG2^KD HSPCs with relevant erythroid and megakaryocyte genes bolded and in yellow boxes. (L-M) GSEA plot of the GATA1 erythroid (left) and GATA1 megakaryocyte (right) cistromes as determined by RNA-seq (L) and ATAC-seq (M). The plot represents the ranked, ordered, nonredundant list of genes with those on the left (red) having a higher correlation with STAG2^KD, whereas those on the right (blue) have a higher correlation with STAG2^Scramble. The vertical black lines indicate the position of each gene in the gene set. Each point represents data from an HPF, bar graphs depict mean ± SD. ∗P < .05; ∗∗P < .01; ∗∗∗P < .001; ∗∗∗∗P < .0001 for comparison of data from STAG2^Luciferase and STAG2^KD assessed using unpaired 2-tailed t test with Welch correction (α = 0.05). FDR, false discovery rate; HSPC, hematopoietic stem/progenitor cell; NES, normalized enrichment score.

Subsequently, we leveraged our annotated cohort of patients with STAG2-mutant MDS to determine whether these patients also had features of abnormal nuclear condensation. We examined bone marrow hematopathology from these patients (N = 18) and found that most patients had altered nuclear morphology (supplemental Figure 9A), and many had nucleated red blood cells and dysplastic erythroblasts in peripheral blood (supplemental Figure 9B).

Finally, to ascertain whether the molecular features we observed in our murine model and in vitro CD34⁺ HSPC model were present in patients, we queried public gene expression data obtained from patients with STAG2-mutant MDS (supplemental Figure 9C).⁵⁶ The differentially expressed genes across all 3 model systems were consistent for erythroid and megakaryocyte ontology pathways among the downregulated and upregulated genes (supplemental Figure 9D). As in our mouse and STAG2^KD models, we observed downregulation of erythroid and upregulation of megakaryocyte differentiation programs, dysregulation of lineage-specific GATA1 cistromes, and altered engagement of lineage-specific TFs downstream of GATA1 (supplemental Figure 9E-H). Collectively, these data support dyserythropoiesis as a common hallmark of STAG2 loss in human cells and in patients with MDS, who share phenotypic and molecular features of impaired erythropoiesis.

Discussion

Our study reveals that Stag2-cohesin–mediated chromatin remodeling enforces the erythroid-specific Gata1 cistrome and maintains balanced erythro-megakaryopoiesis in in vivo murine models, human CD34⁺ HSPCs, and patients with MDS. By integrating multiomics data, we demonstrate that Stag2 loss disrupts chromatin accessibility in EryPs, leading to misallocation of Gata1 binding and altered lineage output. To our knowledge, this is the first study to integrate chromatin accessibility, GATA1 occupancy, and functional amelioration across mouse models, human CD34⁺ HSPCs, and primary MDS samples to define how STAG2 determines GATA1 cistrome specificity. Although GATA1 has been inferred to possess pioneer-like properties in some contexts, our data in EryPs indicate that local chromatin accessibility is the dominant determinant of GATA1 engagement and that STAG2 modulates this accessibility to bias cistrome use, thereby altering lineage output. These findings highlight chromatin accessibility as an active regulatory mechanism directing TF binding specificity and lineage commitment and identify GATA1 as a key TF functionally affected by STAG2 loss in MDS.

A central insight from our work is that Gata1, despite being expressed across multiple hematopoietic lineages, is highly influenced by cell type-specific chromatin accessibility in executing lineage-specific transcriptional programs. The parallels between Stag2 loss and the GATA1 short isoform (GATA1s), both of which produce erythroid loss, megakaryocyte bias, and altered cistromes,^57,58 suggest a convergent mechanism of chromatin-guided GATA1 mislocalization. This concept is exemplified in myeloid leukemia of Down syndrome, where GATA1s mutations and frequent cohesin alterations converge on altered epigenetic regulation, favoring megakaryocytic fate.^{57, 58, 59}

As we and others, establish key mechanisms driving the tumorigenic potential of STAG2 loss, a common observation is strong dysregulation of chromatin accessibility, with more muted effects on DNA looping, transcriptional control, and sister chromatid cohesion, which are the canonical functions of cohesin. These findings demonstrate the mechanistic consequences of altered chromatin accessibility, yet future studies are needed to understand STAG2’s noncanonical control of cell type–specific DNA accessibility.

In summary, our work establishes a model in which chromatin accessibility functions as a molecular gatekeeper that determines TF binding specificity and lineage fate decisions. Our data establishes that STAG2 governs lineage identity by enforcing chromatin conditions that permit erythroid-specific GATA1 engagement. Loss of STAG2 creates an accessibility-driven GATA1 retargeting that underlies dyserythropoiesis in MDS, providing a mechanistic basis for therapeutic resistance¹⁸ and a framework for future chromatin-targeted interventions.

Conflict-of-interest disclosure: A.D.V. is a scientific advisory board member of Arima Genomics and Pixelgen Technologies and reports research support from Kura Oncology outside the submitted work. The remaining authors declare no competing financial interests.