Base-resolution DNA methylome of human MDS hematopoietic stem cell reveals TET2-GFI1 epigenetic axis repressing MDS

Abstract

Dysfunctional hematopoietic stem cells (HSC) drive the initiation of myelodysplastic syndromes (MDS), yet the genome-wide DNA methylation landscape of primitive MDS HSCs and its mechanistic contribution to disease pathogenesis remain poorly defined. Here, we establish single-base resolution DNA methylomes of bone marrow HSCs from MDS patients and healthy donors. We uncover the widespread hypermethylation in CpG islands, alongside hypomethylation in repetitive elements such as Alu. Differentially methylated regions are enriched for genes involved in cancer-related pathways, as well as extrinsic signaling pathways and intrinsic transcriptional networks essential for HSC function. Among these, we identify GFI1 and BMI1 as key targets of DNA methylation dysregulation in MDS. Notably, using either the MDS or a TET2-deficient mouse model, we demonstrate that loss of TET2, a frequently mutated epigenetic regulator in MDS, induces promoter hypermethylation and transcriptional repression of GFI1, contributing to the expansion of the MDS or aged hematopoietic stem and progenitor cell pool. Our study not only charts the base-resolution DNA methylome of human MDS HSCs but also reveals a TET2-GFI1 axis that safeguards HSC homeostasis. These findings provide mechanistic insight into how aberrant DNA methylation drives HSC dysfunction in MDS and offer an epigenomic resource for discovering regulators and therapeutic targets at the stem cell level.

Supplementary Information

The online version contains supplementary material available at 10.1007/s44466-026-00034-4.

Introduction

Myelodysplastic syndromes (MDS) represent a heterogeneous group of clonal hematopoietic disorders characterized by ineffective hematopoiesis, peripheral cytopenias and a heightened risk of progression to acute myeloid leukemia (AML) [1]. Accumulating evidence suggests hematopoietic stem cells (HSCs) as the cellular origin of MDS, where both intrinsic transcriptional dysregulation and extrinsic microenvironmental signals converge to disrupt normal self-renewal and differentiation [2–4]. Among the epigenetic alterations implicated in MDS pathogenesis, DNA methylation abnormalities have emerged as a critical driver, often associated with disease progression and poor prognosis [5]. Previous studies using relative low-resolution or candidate-region approaches have reported widespread methylation changes in MDS, including hypermethylation at tumor suppressor gene promoters [6–8]. However, a whole-genome base-resolution map of DNA methylomes in human primitive MDS HSCs revealing the systemic epigenetic dysregulation at stem-cell level is still lacking.

The ten-eleven-translocation (TET) proteins are α-ketoglutarate- and Fe²⁺-dependent dioxygenases that catalyze the iterative oxidation of both DNA and RNA 5-methylcytosine (5-mC) to 5-hydroxymethylcytosine, 5-formylcytosine, and 5-carboxylcytosine. These oxidized methylcytosines (mC) are key intermediates mediating DNA demethylation through replication-dependent dilution or base excision repair [9]. TET2 plays essential roles in establishing cell-specific function of myeloid cells and their progenitors [10–12]. Recurrent loss-of-function mutations in TET2 are frequently observed in MDS and are associated with increased self-renewal of hematopoietic stem and progenitor cells (HSPC) [13–15]. Nevertheless, whether and how TET2 maintains the expression of key regulators, especially in the context of human MDS HSCs via DNA methylation-centered mechanisms, remains to be investigated.

In this study, we performed whole-genome bisulfite sequencing (WGBS) at single-base resolution to systematically compare the DNA methylome of bone marrow primitive HSCs isolated from MDS patients with thoes from healthy donors. We aimed to investigate the mechanisms of MDS pathogenesis and identify the potential targets for MDS treatment by charting the patterns of aberrant DNA methylation and dysregulated genes in MDS HSCs, and revealing how epigenetic regulators, frequently mutated or dysregulated in MDS, regulate malignant transformation at the HSC level. Our work not only provides a high-resolution epigenetic resource for MDS research but also identifies GFI1 as a critical target through which TET2 maintains HSC homeostasis and suppresses MDS development. These findings advance our mechanistic understanding of epigenetic dysregulation in MDS and highlight potential avenues for stem-cell–directed therapeutic intervention.

Results

Establishing the base-resolution DNA methylome of primitive HSCs from healthy donors and MDS patients

Patients with advanced MDS, i.e. high-risk refractory anemia with excess blasts (RAEB), have hypermethylation variations in several tumor-suppressor gene loci [5, 16]. Hypomethylating agents have become foundation therapies in higher-risk MDS [17, 18]. We mainly focused on MDS patients of REAB-1/2 in this study. To elucidate the specific DNA methylation profiling and discover the potential pathogenic role of DNA methylation in MDS HSCs, we established the genome-wide DNA methylome at single-base resolution using WGBS [19]. In addition to CD34, we used CD133, which is a positive marker of primitive HSC [20], to differentiate HSCs from myeloid progenitors and blasts. The published single-cell RNA-sequencing (RNA-seq) data of bone marrow CD34⁺ cells from MDS patients also confirmed our HSC isolation strategy: CD34⁺CD133⁺ cells are negative for CD38, CD117, and CD123 which are markers of myeloid progenitors and blasts [21] (Fig. S1). DNA samples were extracted from Lin⁻CD34⁺CD133⁺ primitive HSCs that were freshly isolated by sorting from the bone marrow of healthy donors (N) and MDS patients (M) (Table S1). Using chromosome distribution analysis, we also discovered + 8 chromosome variations in two MDS samples (Fig. S2a). Among the total mCs of human HSCs detected in both normal and MDS samples, approximately 97% mCs are in the CG context, unlike the human embryonic stem cell (H1 type) genome, which contains a higher fraction of non-CG mCs (Fig. S2b). Moreover, compared with that in H1 embryonic stem cells, the total level of DNA methylation in normal HSCs is lower, regardless of their sequence context (Fig. S2c). To reveal the genomic patterns of DNA methylation of normal HSCs, we analyzed DNA methylation in the CG context and the CG densities in both whole genome and different genomic elements. We found that, although different functional genomic regions had their own characteristic DNA methylation patterns, the DNA methylation levels decreased when CG densities increased (Fig. S3).

MDS HSCs possess specific DNA methylation variations in gene-associated regulatory regions

We further investigated the DNA methylation differences between normal and MDS HSCs. In the genomic view, we found no significant difference in the proportions and methylation levels of mCs between normal and MDS samples (Fig. 1a, b and Fig. S4a). We then investigated whether DNA methylation variation in the CG context could differentiate MDS HSCs from normal ones in specific genomic elements. After clustering the mean DNA methylation level of each of the genomic elements, we segregated the MDS samples from the normal ones. In the heatmap, we could observe the increased DNA methylation levels in the elements of CpG islands (CGIs), CGI shores, tRNA and 5′-untranslated region (UTR) in at least two MDS samples, compared with those in the normal samples (Fig. 1c). Principal component analysis (PCA) of DNA methylation levels in the DNA elements of 5′-UTR, CGI and tRNA, but not proximal promoters, could differentiate MDS HSCs from normal HSCs, especially in 5′-UTR (Fig. 1d and Fig. S4b). Moreover, upon analyzing the DNA methylation atlas in gene-associated regions, we also observed higher mean DNA methylation levels in the CG context in 5′-UTR in MDS HSCs than in normal HSCs (Fig. 1e). In contrast, in the non-CG context, MDS HSCs had higher DNA methylation levels across the entire gene loci than normal ones, although both showed a low DNA methylation status (Fig. S4c). Notably, such trends were more obvious in M1 and M2 samples, implying heterogeneous DNA methylation patterns across different MDS samples, probably due to the different clinical features. Thus, by establishing the base-resolution DNA methylome of human MDS HSCs, we found increased DNA methylation level in several genomic elements associated with gene expression regulation.

Fig. 1.

Genome-wide trends of DNA methylation of HSCs in healthy donors (N) and MDS patients (M). a The proportion of mCs in CG, CHG, and CHH sequence context. H represents any nucleotide except G. b Distribution of DNA methylation levels of cytosines (C) in the CG context. c Heatmap analysis of two-dimensional hierarchical clustering of mean DNA methylation levels in the CG context of different genomic features. d PCA analysis of DNA methylation levels in the CG context of the indicated genomic elements. e Mean DNA methylation levels for Cs in the CG context across all gene loci. Equal-sized bins are used for calculating the mean DNA methylation level. The part showing the DNA methylation level of 5′-UTR is enlarged. Up, upstream; down, downstream

CGI hypermethylation and repetitive DNA hypomethylation characterize MDS HSCs

Differentially methylated regions (DMRs) play crucial roles in regulating the transcription of key genes in the development of malignancy [22]. Thus, we searched specific DMRs across the whole genome between the two groups (Table S2). DMRs in CGI, which were discovered to be the most DNA methylation aberrant regions in cancers [23], were preferentially hypermethylated, while DMRs in repetitive elements were preferentially hypomethylated, in MDS HSCs compared with normal HSCs (Fig. 2a). Notably, DMRs in 5′-UTR outside the CGIs analyzed here were not preferentially hypermethylated in MDS HSCs. Together with the data in Fig. 1, these results indicate increased DNA methylation levels in 5′-UTR-located CGIs in MDS HSCs. We then separately analyzed the two kinds of DMRs and validated the increased mean DNA methylation levels across CGI-located and hypermethylated DMRs in MDS samples, especially M1 and M2, in profiling analysis (Fig. 2b). Moreover, total DMRs or CGI-located DMRs with increased DNA methylation levels could differentiate MDS HSCs from normal ones, although heterogeneous DNA methylation variations were also observed among the MDS samples (Fig. 2c). Increased DNA methylation levels in hypermethylated DMRs could be observed in the DMR-centered clustering assay in MDS samples, which also showed less increased DNA methylation levels in M3 compared with the other two samples (Fig. 2d).

Fig. 2.

Genomic patterns with increased DNA methylation levels in MDS HSCs. a Distributions of DMRs across different genomic elements. The intragenic regions analyzed here were outside the CGIs. b Profiling analysis of the mean DNA methylation levels across CGI-located hypermethylated DMRs (hyperDMRs) in MDS HSCs compared with normal ones. c PCA of DNA methylation levels in total or CGI-located hyperDMRs. d Clustering analysis of DNA methylation levels across hyperDMRs

We found that DMRs with downregulated DNA methylation levels were enriched in several kinds of repetitive DNA elements especially in Alu (Fig. 3a). We validated the decreased mean DNA methylation levels across Alu-located and hypomethylated DMRs in MDS samples, especially M1 and M3, in profiling analysis (Fig. 3b). Moreover, total DMRs or Alu-located DMRs with decreased DNA methylation levels could differentiate MDS HSCs from normal ones, although heterogeneous DNA methylation variations were also observed among MDS samples (Fig. 3c). Decreased DNA methylation variations in hypomethylated DMRs could also be observed in the DMR-centered clustering assay in MDS samples, which also showed less decreased DNA methylation levels in M2 compared with the other two samples (Fig. 3d). Collectively, these results imply the roles of de novo DNA methylation-associated gene silencing and DNA demethylation-associated genomic instability in dysregulation of MDS HSCs.

Fig. 3.

Genomic patterns with decreased DNA methylation levels in MDS HSCs. a Distributions of DMRs across repetitive elements that were outside the CGIs. b Profiling analysis of DNA methylation levels in Alu-located hypomethylated DMRs (hypoDMR) in MDS HSCs compared with normal ones. c PCA of DNA methylation levels in total or Alu-located hypoDMRs. d Clustering analysis of DNA methylation levels across hypoDMRs

DNA methylation variations in intrinsic and extrinsic regulators in MDS HSCs

To identify the HSC-related key genes dysregulated by aberrant DNA methylation in MDS patients, we performed pathway enrichment analysis on the DMR-associated genes (Fig. 4a and Table S3). Besides the cancer-related pathways that were most enriched with hypermethylated DMR-associated genes, we found some key pathways regulating HSCs maintenance that were also significantly enriched, such as Wnt and mitogen-activated protein kinase signaling pathways. For hypomethylated DMR-associated genes, pathways related to cytokines and chemokines were mostly enriched. Among these pathways, some gene loci, such as interleukin (IL)-6, interferon-γ receptors, granulocyte-macrophage colony-stimulating factor, granulocyte colony-stimulating factor, and IL-2 receptors exhibited hypomethylation abnormalities in MDS HSCs. These genes play important pathological functions in cancer-related inflammation, HSC proliferation, homing, and differentiation [24]. We charted representative DMR-associated genes together with their DNA methylation variations (Fig. 4b).

Fig. 4.

DNA methylation variations of extrinsic and intrinsic regulators in MDS HSCs. a Pathways significantly enriched with DMR–associated genes (P < 0.05). Top panel, hypermethylated DMR; bottom panel, hypomethylated DMR. b Genes in representative pathways. The color of each circle represents the DNA methylation variation levels between normal and MDS samples in the DMRs of each gene (M, MDS samples included in DMR; N, Normal samples). The black line represents the interaction between two genes. c, d Hierarchical clustering of DNA methylation levels of epigenetic regulators and transcription factors. The values show the DNA methylation level of each gene. For determining these, we used a 500 bp window sliding from 10 kb upstream to the 3′-UTR of each gene locus to find the most differentially methylated window between normal and MDS samples. The DNA methylation level of the window for each sample was used here

As the key intrinsic regulators of HSCs, transcription factors and epigenetic regulators are crucial for regulating HSCs' self-renewal and differentiation [3], and they are also largely associated with DMR-associated genes (Table S4). To reveal the correlations between intrinsic regulators and aberrant DNA methylation during MDS development, we analyzed DMR-associated genes in light of previously reported key intrinsic regulators and clustered their DNA methylation variations [5, 23, 24], identifying the distinct DNA methylation differences for these regulators between the two groups. We found that transcription factors, such as GFI1, RUNX1, CEBPB, and ERG, and epigenetic regulators, such as BMI1, histone deacetylase 5 (HDAC5), SMARCA4, DNMT3B, mixed lineage leukemia 3 (MLL3), and PHC1, are associated with DNA methylation variations in at least two MDS samples (Fig. 4c, d). These data imply that DNA methylation could dysregulate key gene transcription regulators, which may lead to extensive epigenetic or transcriptional variations in MDS HSCs.

TET2 promotes GFI1 expression via promoter demethylation to inhibit the expansion of MDS HSCs

The data showed that GFI1 and BMI1 were, respectively, associated with hypermethylated or hypomethylated DMRs in two MDS samples (M1 and M2). The hypermethylated DMRs were located across the 5′-UTR in the GFI1 gene locus, and notably, there were also DNA methylation variations among MDS samples in the GFI1 promoter (Fig. 5a). The hypomethylated DMRs were located upstream of the BMI1 gene locus (Fig. S5a). To verify the universality and clinical correlation of GFI1 and BMI1 with aberrant DNA methylation, we tested another 12 HSCs samples of MDS patients and found the consistence of DNA methylation abnormalities in the other 9 samples of RAEB cases, whereas those of the two transformed AML cases and the single refractory anemia case were similar to the M3 sample, further implicating the heterogeneity among different MDS subtypes. The mRNA levels of GFI1 and BMI1 were also dysregulated, accompanied by the aberrant DNA methylation (Fig. 5b, c and Fig. S5b, c). Due to the small cohort size of MDS subtypes in this study and the high heterogeneity of MDS, we cannot conclude that there is no DNA methylation variation in GFI1- and BMI1-associated DMR in other MDS subtypes. However, according to our data, the two DNA methylation variations were common among patients with the RAEB subtypes.

Fig. 5.

TET2 represses DNA methylation in the GFI1 promoter to restrict the stem pool of MDS. a DNA methylation levels in the CG context across the GFI1 gene in genome browser view. Red boxes show regions with DNA methylation variations in normal (N) and MDS (M) HSCs. b, c MeDIP-qPCR analysis of DNA methylation levels of region 1 in the GFI1 locus and reverse transcription qPCR analysis of mRNA levels of GFI1 in normal (n = 10) and MDS (n = 15) HSCs. d MeDIP-qPCR analysis of the DNA methylation levels of region 2 in the GFI1 locus and reverse transcription qPCR analysis of mRNA levels of GFI1 in MDS samples with increased DNA methylation levels in region 1 compared with normal samples (n = 11). e DNA methylation levels across Gfi1 gene in genome browser view (GSE129691). Red boxes show regions with DNA methylation variations in control (Ctrl) and MDS (NHD13) murine HSCs. f-i Reverse transcription qPCR analysis of mRNA levels of the indicated genes, MeDIP-qPCR analysis of DNA methylation levels in region 2, and colony formation analysis of the serial replating assay of the control and lentivirus-mediated Tet2-KD (shTet2) Lin⁻ HSPCs from NHD13 mice. Data were normalized by input DNA (b, d, h) or β-actin (c, d, g), and compared with normal samples or control groups. Data are the mean ± s.d. g-i, two-tailed unpaired Student’s t-test. *, P < 0.05; **, P < 0.01; ***, P < 0.001

Interestingly, we noticed that, although region 1 of M1 and M2 MDS samples had similar DNA methylation levels, the mRNA level of GFI1 in the M2 sample was higher than that in the M1 sample (Fig. S5d), implying a difference in regulatory mechanisms between M1 and M2 samples in the GFI1 gene loci. We noticed that the DNA methylation level of region 2 in the M2 sample was lower than that in M1 sample (Fig. 5a), implying that the DNA methylation level of region 2 might regulate GFI1 expression in MDS HSCs. We then investigated the correlation between DNA methylation levels of region 2 and GFI1 mRNA levels in MDS samples which had higher DNA methylation levels in region 1 than those in normal samples, and found that they were negatively correlated (Fig. 5d), implying that DNA methylation in region 2, which is located in the GFI1 promoter, could repress GFI1 expression.

TET2 acts as a tumor repressor in myeloid cancers, including MDS. In a Nup98-HoxD13 (NHD13) transgenic mouse model of MDS, TET2 loss was observed to further expand the hematopoietic stem/progenitor pool and accelerates leukemia transformation [25]. We further investigated whether TET2 could promote Gfi1 expression by mediating DNA demethylation in the Gfi1 promoter. Using public data of the DNA methylome of HSPCs from NHD13 mice [26], we first observed decreased DNA methylation level in the Gfi1 promoter, although there was increased DNA methylation level across the 5′-UTR of Gfi1 in HSCs from NHD13 mice compared with the wild-type control (Fig. 5e). We then silenced the Tet2 in HSPCs from NHD13 mice and found that Tet2-knockdown (KD) increased the DNA methylation level in the Gfi1 promoter and decreased the mRNA level of Gfi1 (Fig. 5f-h). Colony-forming cell (CFC) assays revealed that the Tet2-KD in HSCs from NHD13 mice exhibited not only a higher CFC number but also a higher replating capacity than the control HSCs from NHD13 mice (Fig. 5i). These results imply that TET2 promotes promoter demethylation to maintain high expression of GFI1 in HSCs to repress MDS development.

TET2 loss decreases Gfi1 expression via promoter hypermethylation for the expansion of aged HSCs

Given that loss-of-function mutations of Tet2 contribute to MDS development, and aged Tet2-knockout (KO) mice showed enlargement of the HSC compartment and myeloproliferation [13], we further validated that increased expression of Gfi1 through promoter demethylation in HSCs was a key mechanism for TET2-mediated inhibition of MDS development. Using a published dataset of RNA-seq and WGBS analysis of young or aged bone marrow Lin⁻Sca-1⁺c-Kit⁺ stem cells (LSK) from Tet2 catalytic mutant (Mut) and KO mice [27], we analyzed genes with decreased mRNA levels due to TET2 loss in aged but not young LSK. Interestingly, we found that a large fraction of these genes showed increased mRNA levels in aged wild-type control LSK (Fig. 6a), and these genes were enriched in cancer-associated and neural signaling pathways (Fig. 6b and Fig. S6), implying that the impaired transcription increase of these genes in aged HSCs due to TET2 loss may contribute to HSC transformation, which was investigated in aged Tet2-KO mice. Notably, Gfi1 was in the list (Fig. 6c). We further investigated the role of TET2-repressed DNA methylation in regulating the expression of Gfi1. Interestingly, bone marrow LSK and Lin⁻ progenitor cells from both Tet2-Mut and Tet2-KO young mice showed slightly increased DNA methylation levels in the Gfi1 promoter (Fig. 6d), further implying TET2-mediated DNA demethylation in promoting Gfi1 expression against HSC aging.

Fig. 6.

Loss of TET2 leads to DNA hypermethylation-mediated repression of Gfi1 expression in aged HSCs. a-c Heatmap presentation and KEGG pathway enrichment analysis of the genes (including Gfi1, c) with decreased mRNA levels in LSK from aged but not young Tet2 catalytic mutation (Mut) and Tet2-KO mice in RNA-seq analysis (GSE227977). d DNA methylation levels across the Gfi1 gene locus in WGBS analysis of HSPCs from young Tet2-mut, Tet2-KO, and the wild-type control mice (GSE227977). The red box shows DMR. e–g MeDIP-qPCR, reverse transcription-qPCR, and western blot analysis of DNA methylation levels in Gfi1-specific DMR, Gfi1 mRNA, and GFI1 protein in Lin⁻ HSPCs from control and Tet2-KO young and aged mice. h-j Colony formation analysis of the serial replating assay of the control and lentivirus-mediated GFI1-overexpressed (OE) Lin⁻ HSPCs from aged Tet2-KO mice. Data were normalized by input DNA (e) or β-actin (f), and compared with control groups. Data are the mean ± s.d. e, f, j two-tailed unpaired Student’s t-test. *, P < 0.05; **, P < 0.01; ***, P < 0.001

We further experimentally validated the increased DNA methylation levels in the Gfi1 promoter in Tet2-KO Lin⁻ cells, especially from aged mice (Fig. 6e), and decreased mRNA and protein levels of Gfi1 in Tet2-KO Lin⁻ cells from aged mice (Fig. 6f, g). We then observed the effect of rescuing GFI1 expression in aged Tet2-KO HSCs, and found that overexpression of GFI1 significantly reduced the replating potential of Tet2-KO Lin⁻ HSPCs from aged mice in vitro (Fig. 6h-j). These results further indicate that reduced expression of Gfi1 due to DNA hypermethylation caused by TET2 loss contributes to MDS development at the stem-cell level.

Discussion

By establishing single-base resolution DNA methylomes of bone marrow primitive HSCs from healthy donors and MDS patients, we revealed the contribution of aberrant DNA methylation to HSC dysfunction in MDS at both single-gene locus and whole-genome levels. From the perspective of genome-wide DNA methylation variation in MDS, although hypermethylation is associated with the regulatory regions of genes, hypomethylation is found in intergenic regions, including repeat elements. This intergenic hypomethylation may be inherited by the offspring of MDS HSCs and the progenitors, which may increase the genomic variation frequency to accelerate MDS development. It is worthwhile to note that tRNA genes are associated with hypermethylation in MDS HSCs, indicating that decreased protein synthesis efficiency in HSCs may be a pathogenic factor for MDS. Moreover, given that CD34⁺ bone marrow cells, which were used for DNA methylation analysis in most previous studies, cannot discriminate primitive HSCs from both myeloid progenitors and blasts, our study provides valuable data for exploring how DNA methylation variations in primitive HSCs contribute to MDS development. Notably, DNA methylation variation in the M3 case is partially different from that in M1 and M2 cases, indicating that the DNA methylation variations caused by heterogeneity among MDS patients in either the same subtype or among different MDS subtypes should be considered. Moreover, because of the small sample size of our study, future studies are needed to further observe the correlations of our identified DNA methylation variations, gene expression, and clinical features such as blast number and age of MDS cases in the same or different MDS subtypes.

To identify the aberrant DNA methylation-dysregulating genes that are potentially the pathogenic factors or therapeutic targets for MDS, we identified DMR-associated genes between normal and MDS HSCs genome-wide, and found that cancer-related pathways were most enriched by hypermethylated DMR-associated genes. Further exploration of the DNA methylation-dysregulating genes in our DMR data may lead us to discover more cancer-related genes to provide potential biomarkers and targets for the treatment of leukemia transformed from MDS.

Some extrinsic pathways and intrinsic regulators regulating the self-renewal and differentiation of HSCs have been found to be associated with DMRs. For extrinsic pathways, in addition to the reported ones that are important for HSC maintenance, some cytokine-related pathways such as inflammatory and myeloid-specific pathways, were enriched by hypomethylated DMR-associated genes. Dysregulation of these cytokines and their signal transducers may lead to an abnormal microenvironment, causing abnormality of bone marrow HSCs in MDS patients. For intrinsic regulators, epigenetic regulators and transcription factors with DNA methylation variations in at least two MDS samples are or may be involved in MDS development, for example, somatic mutations and chromosomal rearrangements involving RUNX1 are frequently observed in MDS [28]; activation of HOXB4 permits long-term maintenance of functional HSPC [29]; mutation or suppression of the tumor suppressor MLL3 promotes self-renewal capacity of HSCs and impairs the differentiation of HSPCs [30, 31]; and HDAC5 inhibits HSC homing and engraftment [32]. Future studies are needed to investigate how aberrant DNA methylation dysregulates the expression of these genes and their contribution to MDS development at the HSC level.

Moreover, we focused on DNA methylation abnormalities in the GFI1 and BMI1 gene loci, which were verified in additional MDS samples. Zinc-finger repressor GFI1 was reported to restrict the proliferation of HSCs and preserve HSC functional integrity. Loss of GFI1 leads to elevated proliferation rates and a compromised repopulation function of HSCs [33–35]. GFI1 is also important for the differentiation and function of myeloid cells. Loss of GFI1 leads to severe neutropenia, accumulation of immature monocytic cells, and enhanced inflammatory immune response [36, 37]. Reduced expression of GFI1 leads to a fatal myeloproliferative disease in the Gfi1-KD mouse model [38]. GFI1-36N, which encodes a variant GFI1 protein with decreased efficiency to act as a transcriptional repressor, was found to be clearly correlated with an increased risk of MDS patients to develop AML [39, 40]. Expression of GFI1 was also reported to be repressed in MDS samples and low expression of GFI1 is associated with an inferior prognosis of AML patients [41]. For the epigenetic mechanism, GFI1 interacts with the CoREST complex, including HDAC1/2 and lysine-specific demethylase 1, which mediates transcriptional repression by GFI1 [42]. Thus, our data provide an epigenetic mechanism for the repressed expression of GFI1 during MDS development, implying that GFI1-dependent CoREST complex-mediated gene silencing is inhibited by DNA methylation in MDS HSCs. Identifying the target genes under the control of this epigenetic regulation axis needs further investigation. As a component of the polycomb repressive complex 1, BMI1 is essential for generating self-renewing adult HSCs and is required for efficient self-renewal of HSCs and leukemic stem cells [43, 44]. Therefore, silencing GFI1 and overexpressing BMI1 driven by DNA methylation abnormalities in HSCs during MDS development may break the balance of quiescence and self-renewal of HSCs, contributing to the transformation of hematopoietic progenitors.

TET2 is a well-established tumor suppressor for myeloid malignancies. The epigenetic mechanism underlying the role of TET2 in restricting the transformation of HSCs is receiving serious attention. Using both a murine MDS and a Tet2-KO model, we demonstrated that TET2 promoted the expression of GFI1 through promoter DNA demethylation in both MDS and aged HSCs. According to our data, repressed expression or impaired enzymatic activity of TET2, probably owing to gene mutation or negative regulation mechanisms [45, 46], leads to promoter hypermethylation-mediated silencing of GFI1 expression, contributing to clonal hematopoiesis during MDS development and HSC aging. Dysregulation of TET2-GFI1 epigenetic axis is a pathogenic mechanism for age-related hematopoietic diseases, including MDS, as well as myeloproliferative disorders, and myeloid leukemia [27]. Moreover, our study reveals that loss of TET2 in HSCs prevents the increase in expression of genes that are enriched in pathways of cancer, insulin resistance, and neural signaling during aging. These genes may play repressive roles in the development of aging-related diseases, such as GFI1 in repressing the development of MDS and AML at the stem-cell level. TET2 was recently reported to oxidize RNA m5C and antagonize MBD6-dependent H2AK119ub deubiquitination to repress aberrant expansion of myeloid leukemia cells [47]. Loss of TET2 further accelerates MDS development by promoting the occurrence of secondary mutations in an NHD13 mouse model [25]. Our study provides a DNA demethylation-dependent transcription regulation function of TET2 in repressing myeloid cancers, especially during aging. However, more evidences from both in vivo experiments using a Tet2-KO NHD13 mouse model and clinical samples with Tet2 mutations is needed for validating our finding.

Notably, the region of TET2-mediated DNA demethylation in the GFI1 gene locus was not the region significantly de novo methylated in human MDS HSCs compared with normal HSCs, implying that there is another mechanism inhibiting DNA hypermethylation in HSCs beyond TET2 during MDS development, which is waiting to be revealed. In silencing GFI1 expression in MDS HSCs, TET2 loss plays an accelerating role. Thus, TET2 deficiency can further expand the hematopoietic stem and progenitor cell pool and accelerates leukemia transformation, contributing to MDS development. Our study provides a mechanism for regulating DNA methylation for the pathogenic role of either loss-of-function mutations or expression and enzymatic activity inhibition of TET2 in MDS patients.

Materials and methods

Human bone marrow collection and sample preparation

Bone marrow aspirates of 15 MDS patients with only supportive care and 10 healthy donors were collected between 2010 and 2011 and grouped according to the World Health Organization (WHO) classification system for MDS. The study was approved by the ethics committees of the Chinese People’s Liberation Army General Hospital and informed patient consent was obtained. All the bone marrow samples were harvested freshly and processed for flow sorting immediately. Mononuclear cells (MNCs) from the bone marrow samples were isolated by the Ficoll-Paque (GE Healthcare, Uppsala, Sweden) density gradient centrifugation. For purification of HSCs, MNCs were stained with anti-human lineage markers, CD34, and CD133 (BioLegend, CA, USA), and a population of Lin⁻CD34⁺CD133⁺ cells were sorted using the MoFlo XDP cell sorter (Beckman-Coulter, CA, USA). The > 95% purity of the population was confirmed by performing fluorescence-activated cell sorting (FACS) analysis of the sorted cells.

Isolation of genomic DNA and bisulfite high-throughput sequencing

Total DNA was isolated using a QIAamp DNA Mini Kit (QIAGEN, Hilden, Germany) following the manufacturer’s instructions. Bisulfite DNA was prepared and high-throughput sequencing was performed as previously described [19].

Reverse transcription quantitative PCR

Reverse transcription was performed using a reverse transcription system (Toyobo). Products were used for quantitative PCR (qPCR) analysis by LightCycler (Roche) and SYBR RT-PCR kit (Takara) as described previously [11]. Data were normalized by the level of β-actin.

Bisulfite sequencing alignment and identification of mCs

The bisulfite treated reads generated by Illumina sequencing were aligned to Hg18, and mC identification was performed as previously reported [19].

Methylated DNA immunoprecipitation assay

Genomic DNA was extracted using the QIAamp DNA mini kit (QIAGEN) and fragmented into a main band between 200 bp and 400 bp using the Bioruptor from Diagenode. The purified genomic DNA was denatured at 95 ℃ for 10 min and incubated with 5-mC antibody (Zymo Research) at 4 ℃ overnight. Protein A/G magnetic beads were added to pull-down antibody-DNA complexes for 2 h at 4 ℃. Beads were washed and digested by protease K. Eluted DNA was extracted with phenol–chloroform and ethanol precipitation. The obtained DNA from the methylated DNA immunoprecipitation (MeDIP) assay was subjected to qPCR analysis.

Identification of DMRs

All the shared cytosines (C) in the CG context that had a coverage of over 1 × in each of the six samples were selected for DMR analysis. DMRs were identified using sliding windows containing 5 CGs each through the following steps: step 1, windows that had a mean depth of over 10 and 80% CGs coverage were used; step 2, windows having DNA methylation rates with at least double difference (Fisher test, P ≤ 0.01) and with either of the DNA methylation level being higher than 50% was identified as differentially methylated windows; step 3, each identified differentially methylated window was extended unless two CGs were at a distance of over 200 bp or the window no longer met the above standards for a differentially methylated window. The extended window was the DMR.

DMRs between each normal and disease sample were first analyzed. If the DNA methylation variation of the DMR was significant among three normal samples for a 3 × 2 Fisher test, the DMR was discarded. If the DMR appeared spontaneously in two or three MDS samples, it was used for the analysis. The gene (from 10 kb upstream to the transcription end site) with at least one DMR was defined as a DMR-associated gene.

DNA methylation pattern discrepancy between MDS and normal samples

To investigate the DNA methylation pattern discrepancy between MDS and normal samples, the DNA methylation level was computed as the average for every C in the CG or non-CG context in a specific genomic element. Alternatively, DNA methylation levels in the CG context were used for PCA and DMR-based heatmap analysis using deepTools.

Pathway enrichment analysis

DMRs-associated genes were subjected to pathway enrichment analysis based on the DAVID database.

Mice

Tet2-KO mice with a C57BL/6 × 129/SvEv background were provided by Dr. R. L. Levine. NUP98-HOXD13 (NHD13) transgenic mice were obtained from Jackson Laboratory [48]. All mice were maintained under specific pathogen-free conditions. All animal experiments were performed according to the National Institutes of Health Guide for the Care and Use of Laboratory Animals, with the approval of the Scientific Investigation Board of Naval Medical University, Shanghai, China.

Lentivirus-mediated gene knockdown or overexpression

Short hairpin RNAs (shRNA) specifically targeting Tet2 were generated by GenePharma (Shanghai, China). Lin⁻ HSPCs were isolated from the bone marrow of Tet2-KO or NHD13 mice using the EasySep™ negative selection kit (STEMCELL Technologies) and infected with lentivirus expressing Tet2-specific shRNA, Gfi1, or controls. Cells were infected with a multiplicity of infection of 50 for at least 4 days in serum-free expansion medium (09650; STEMCELL Technologies), or methyl-cellulose media (M3434; STEMCELL Technologies) with myeloid differentiation cytokines.

Colony-forming assays

Lin⁻ HSPCs isolated from the bone marrow of Tet2-KO or NHD13 mice, infected with lentivirus for Tet2 silencing or GFI1 overexpression, and were seeded into methyl-cellulose media (M3434; STEMCELL Technologies) with myeloid differentiation cytokines. After incubation for 7 days, colonies were counted, and then sequentially replated for another 7 days for the replating assay.

Statistical analysis

Error bars displayed throughout the paper represent s.d. and were calculated from triplicate technical replicates. Sample sizes were chosen using standard methods to ensure adequate power, and no randomization of weight and sex or blinding was used for animal studies. Data shown are representative of three independent experiments. No statistical method was used to predetermine the sample size. Statistical significance was determined using unpaired two-sided Student’s t-test, with P < 0.05 considered statistically significant.

Authors’ contributions

X.C., Q.Z., Y.H. and L.H. conceived the study and designed the experiments; Y.H. and Y.G. sorted the HSCs from human bone marrow samples; L.H., Y.G., Y.L., N.L. and B.Z. selected and characterized the samples, and provided disease-specific expertise in data analysis; J.L. provided the NHD13 transgenic mice; L.H. and Q.S. performed the experiments; Q.Z. conducted the bioinformatic analysis; X.C., Q.Z. and Y.H. analyzed the data and wrote the paper.

Funding

This work was supported by grants from the National Natural Science Foundation of China (82388201, 32270961, and 82271799), Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences (2024-I2M-ZD-005 and 2021-I2M-1–017), and Shanghai Leading Talent Program of Eastern Talent Plan (LJ2023115).

Data availability

The data for the DNA methylomes of MDS and normal HSCs have been deposited into the Gene Expression Omnibus under accession code GSE323172. All other study data are included in the article and Supplemental information. All data are available in the main text or the supplementary materials.

Declarations

Competing interests

The co-corresponding author Xuetao Cao is the Editor-in-Chief of this journal Immunity & Inflammation. However, he was not involved in the peer-review or decision-making process for this manuscript. The authors declare no other competing interests.