The METTL3-YTHDC1 axis mediates architectural RNA m⁶A modification to modulate the integrity of chromatin TADs in MLLr + AML genome

Abstract

Background

The RNA methyltransferase METTL3 as a key regulator of acute myeloid leukemia (AML) contributes to malignant transformation. Chromatin topologically associating domains (TADs) are critical for maintaining AML genome integrity, but the mechanism by which METTL3 facilitates TADs integrity in AML progression remains unclear.

Methods

To determine whether METTL3 is transcriptionally activated by MLL in MLL-rearranged (MLLr+) AML cells, we analyzed MLL ChIP-seq data. Additionally, we performed a multi-omics approach—including RNA-seq, immunoprecipitation-mass spectrometry (IP-MS) for METTL3 and YTHDC1, DNA: RNA hybrid immunoprecipitation sequencing (DRIP-seq), METTL3 ChIP-seq, CTCF ChIP-seq, H3K4me3 and H3K27ac ChIP-seq—to delineate the functional interplay among METTL3-YTHDC1 axis, R-loops, and CTCF in the MLLr + AML genome. Furthermore, METTL3-RIPseq, YTHDC1 RIPseq, CTCF RIPseq, m⁶A-seq, and Hi-C-seq assays were conducted to elucidate the function of the METTL3-YTHDC1 axis-mediated m⁶A modification of architectural RNAs (arcRNAs) in regulating CTCF-dependent TAD boundary activity.

Results

METTL3 is transcriptionally activated by MLL and forms a complex with YTHDC1 and CTCF, colocalizing at promoters and enhancers in MLLr + AML cells. METTL3 depletion disrupts CTCF binding sites (CBSs) and reduces chromatin accessibility at key leukemic genes (e.g. MYB and RUNX1). Hi-C analysis further reveals that YTHDC1 loss compromises CTCF-dependent 3D genome organization. METTL3-mediated m⁶A modification stabilizes arcRNAs and R-loops, which are crucial for maintaining TAD integrity at leukemic loci. Mechanistically, YTHDC1 recognizes m⁶A-modified arcRNAs (e.g. MALAT1) to enhance R-loop formation, thereby sustaining CTCF-mediated TAD activity in the MLLr + AML genome.

Conclusions

Our study identifies the METTL3-YTHDC1-CTCF axis as a critical regulator of AML signature gene expression by orchestrating 3D genome organization. These findings provide novel insights into AML pathogenesis and reveal new therapeutic targets for this kind of aggressive disease.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12943-025-02545-x.

Background

Acute myeloid leukemia (AML) is a complex disorder of hematopoietic stem/progenitor cell (HS/PC) characterized by the sequential acquisition of driver gene mutations that promote the self-renewal of leukemic stem cells (LSCs) and induce a maturation arrest, ultimately impairing differentiation into functional myeloid cells [1–5]. AML cases involving the rearrangements of lysine methyltransferase 2 A (KMT2A), previously known as mixed-lineage leukemia (MLL), are associated with particularly poor prognosis [6]. Mechanistically, MLL-rearranged (MLLr) AML is driven by aberrant expression of HOX genes and their cofactor MEIS1 [7].

The CCCTC-binding factor (CTCF) is recognized as a key regulator of mammalian three-dimensional (3D) genome organization to regulate gene transcription [8–10]. Remodeling of CTCF-mediated chromatin TAD boundaries alters chromatin domains, driving aberrant transcription of leukemia-related genes in the MLLr + AML genome [11, 12]. CTCF collaborates with the cohesin complex to establish chromatin loops and topologically associated domains (TADs) [13], defining the TAD boundaries that constrain regulatory interactions within specific genomic neighborhoods [8, 14]. The CTCF/Cohesin complex stabilizes enhancer-promoter interactions within TADs to regulate cell type-specific transcription and maintain cell identity [15]. Beyond Cohesin [16], CTCF also mediates enhancer-promoter interactions through homodimerization or cooperation with other transcription factors (TFs) [13], including MYC [17], NPM1C [18] and MAZ [19].

Emerging studies have highlighted the importance of RNA-interacting domains within CTCF for facilitating enhancer-promoter interactions and high-order genome organization [20, 21]. For example, the long non-coding RNA (lncRNA), such as HOTTIP, interacts with CTCF to coordinate with CTCF-dependent TADs, thereby regulating leukemia-associated genes and WNT signaling targets in MLLr + AML [11, 22]. Similarly, the HOXBLINC lncRNA mediates CTCF-driven chromatin interactions to modulate the expression of leukemia signature genes, such as HOXB4 and STAT1 [18, 23]. These findings position chromatin-architectural RNAs (arcRNAs) as critical regulators of gene expression by defining CTCF-mediated TAD boundaries in the AML genome [24, 25],

N6-methyladenosine (m⁶A), a dynamic and reversible RNA modification present on mRNAs and lncRNAs, plays a key role in AML pathogenesis [26–28]. METTL3, the core catalytic subunit of the m⁶A methyltransferase, is responsible for the majority of m⁶A deposition [29, 30]. Dysregulated METTL3 expression contributes significantly to AML development and progression [31]. Aberrant transcription of METTL3 enhances hematopoietic stem cell (HSC) self-renewal, and drives AML-like disease by sustaining leukemic transcription programs (e.g., MYC, BCL2, and PTEN) [31–34]. Notably, beyond its cytoplasmic role in RNA metabolism, nuclear METTL3 can directly bind chromatin and modulate epigenetic pathways to maintain the leukemic state [35]. YTHDC1, a nuclear m⁶A reader protein, recognizes m⁶A-modified RNAs and regulates leukemia stem cell (LSC) self-renewal in AML by targeting leukemia-associated genes, such as MCM and BCL2 [36]. However, the mechanisms by which the METTL3-YTHDC1 axis mediates m⁶A modification of chromatin-associated arcRNA to regulate CTCF-dependent TAD boundaries involved in establishing AML-specific chromatin signatures and the leukemic transcription programs remain poorly understood. Therefore, this research aims to investigate how the METTL3-YTHDC1 axis facilitates m⁶A modification of arcRNAs to reinforce leukemic TAD boundary activity, alter 3D genome organization, and sustain the leukemic transcription program in AML.

Materials and methods

Cell lines

The human cell line HEK293T was obtained from the American Type Culture Collection (Manassas, VA, USA), while several human leukemia cell lines (MOLM13, HL60, MV4-11, OCI-AML3, OCI-AML2 and THP1) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). HEK293T cells were cultured in DMEM supplemented with 10% bovine serum (FBS) and 1% penicillin/streptomycin, MOLM13, HL60 and THP1 AML cells were cultured in RPMI 1640 supplemented with 10% FBS and 1% penicillin/streptomycin. MV4-11 and OCI-AML3 cells were cultured in IMDM supplemented with 10% FBS and 1% penicillin/streptomycin. OCI-AML2 cells was cultured in MEMα supplemented with 10% FBS and 1% penicillin/streptomycin. All cells were maintained at 37 °C with 5% CO₂ in a humidified incubator.

Generation of METTL3- or YTHDC1-depleted AML cells using pLKO.1-shRNA system

Human shRNAs targeting METTL3 and YTHDC1 were designed using the shRNA design tool (https://portals.broadinstitute.org/gpp/public/gene/search). The shRNA sequences were subcloned into the RNAi Consortium (TRC)-based short hairpin RNA lentiviral vector, pLKO.1-TRC (Addgene #10878), with scramble shRNA (Addgene #1864) as the control. Lentiviral particles were produced in HEK293T cells using pLKO.1-shRNA or shScramble vector, pMD2.G (Addgene #12259) and psPAX2 (Addgene #12260) vectors. MOLM13 AML cells were transduced with the lentivirus for 48 h, followed by selection with 2 µg/mL puromycin for an additional 48 h. Subsequently, RNAs were extracted and purified using the TRIzol reagent (TIANGEN, cat# DP424), and the knockdown efficiency was validated by RT-qPCR.

Generation of dCas9-RNaseH, dCas9-APEX and dCas13-ALKBH5 vector system

All guide RNAs (gRNAs) were subcloned into the pLKO5.sgRNA.EFS.tRFP vector (Addgene #57823). To generate the dCas9-RNaseH fusion, GFP-tagged RNaseH (Addgene #65784) was subcloned into the pHR-dCas9 vector (Addgene #46911). sgRNA was designed to target RUNX1 or MYB promoter loci. A non-targeting control sgRNA was designed to bind an intergenic region (chr21: 36,512,097 − 36,512,116) with no known R-loop formation or leukememogenic relevance. For dCas9-APEX2 (plasmid #97421) system [37], sgRNAs were designed to target three regions within the RUNX1 promoter: approximately − 150 bp, -10 bp and + 50 bp relative to the transcription start site. A control sgRNA targeting a non-coding intergenic region (chr21: 36,512,097 − 36,512,116) with no known arcRNA associations as the control. For dCas13-ALKBH5 (dm6A) system construction [38], gRNAs were designed to target the m⁶A modification site (chr11: 65,267,935 − 65,267,962) within the MALAT1 transcript.

RNA isolation, RT-qPCR, and RNA sequencing

Total RNA was purified from cells using the TRIzol^® reagent (TIANGEN, Cat# DP424), followed by the standard chloroform phase separation and alcohol precipitation. cDNA was synthesized from 1 µg total RNA, and reverse transcription was performed using the HiScript II Q RT SuperMix kit (Vazyme, cat#R222-01) according to the manufacturer’s instructions. cDNA was then analyzed by quantitative PCR, and relative expression differences were calculated using the 2^-ΔΔCt method.

RNA sequencing libraries were prepared from 1 µg total RNA using the VAHTS^® Universal V8 RNA-seq Library Prep Kit for Illumina (Vazyme, Cat# NR605-01/02) following the manufacturer’s instructions. In brief, polyadenylated mRNA was enriched using poly-T oligo magnetic beads. The enriched RNA was then purified and fragmented into ~ 200–300 bp fragments, a size range optimized for short-read sequencing. Subsequently, the fragmented RNAs were used as templates for the first- and second- strand cDNA synthesis, followed by PCR amplification and indexing with sequencing adaptors. Library quality was assessed using the Qubit and Agilent Bioanalyzer. Libraries were subjected to paired-end sequencing with a read length of 150 bp on the Illumina NovaSeq 6000 platform.

RNA-seq analysis

RNA-seq raw reads were trimmed using Cutadapt (http://cutadapt.readthedocs.io, version 1.2.0) [39] and aligned to the hg19 human genome using TopHat (version 2.0) and Bowtie2 [40–42]. Read quality was assessed using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). FPKM (paired-end fragments per kilobase of exon model per million mapped reads) values were calculated using Cufflinks v2.2.1. Differential expression analysis was performed using the DESeq2 package in R (V4.3.1) with GENCODE annotations (https://www.gencodegenes.org/). A heatmap of differential expression genes (DEGs) was generated using cluster 3.0 and Java Treeview based on log₂-transformed FPKM values [43]. Gene set enrichment analysis (GSEA) utilized gene sets from the Molecular Signatures Database (http://software.broadinstitute.org/gsea/doc/GSEAUserGuideFrame.html) [44]. GO analysis of DEGs was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) tool (https://david.ncifcrf.gov/) [45]. Sequence reads have been deposited in the NCBI GEO under accession number GSE287825.

Gene set enrichment analysis

Gene set enrichment analysis was performed using GSEA software (http://www.broadinstitute.org/gsea/) [44] and R statistical software. Leukemic-associated gene sets were obtained from the Molecular Signatures Database (MSigDB, http://software.broadinstitute.org/gsea/msigdb/search.jsp) [46, 47].

Co-Immunoprecipitation (Co-IP) and Immunoprecipitation-Mass Spectrometry (IP-MS) assays

For Co-IP assay, 2 × 10⁶ MOLM13 cells were washed with PBS and lysed in RIPA lysis buffer containing protease inhibitors on ice for 30 min. The lysates were incubated with the target-specific antibodies (Anti-CTCF, Cell Signaling Technology Cat# 2622; anti-METTL3, Abcam Cat# ab195352; anti-METTL14, Proteintech Cat# 26158-1-AP; Anti-YTHDC1, Abcam Cat#ab122340; Anti-DHX9, Abclonal Cat# A17955; Anti-DDX24, Proteintech Cat#15769-1-AP; Anti-GAPDH, Proteintech Cat#10494-1-AP) overnight at 4 °C, followed by the addition of protein A/G magnetic beads (Vazyme, cat# PB101-02) for 2 h. Beads were washed three times with washing buffer and boiled in SDS buffer for 10 min. Proteins were resolved by SDS-PAGE and analyzed by immunoblotting. For IP-MS assay, eluted proteins were separated by SDS-PAGE and stained with the Coomassie blue. Protein pellets were centrifuged at 16,000 g at 4 °C for 30 min, washed with cold acetone, air-dried for 1 min, and solubilized in 1× Laemmli sample buffer. Immunoglobulin light/heavy chain bands were excised separately. Peptides were analyzed by nano-liquid chromatography-tandem mass spectrometry (nano-LC-MS/MS). Peptide and protein identification were used MaxQuant (version 1.5.3.30) [48] against the Uniprot database (https://www.uniprot.org/). For interactome analysis, proteins with more than 4 unique peptides and four-fold enrichment were retained. Differentially expressed proteins were filtered at log2 (fold change) > 2 and adjusted p value < 0.01 (Benjamini-Hochberg correction).

Molecular docking

The three-dimensional structures of the human CTCF RNA-binding domain (PDB: 8SSQ) and the YTH domain of YTHDC1 (PDB: 6YNP) were retrieved from the Protein Data Bank (https://www.rcsb.org/). Molecular docking was performed using AutoDock Vina (v1.2.3) ( [49]) for protein-protein interaction prediction. Structural visualization and interaction analysis used PyMOL(https://www.pymol.org). Structures were prepared using PyMOL (https://www.pymol.org) and AutoDock Tools by removing non-protein atoms, adding hydrogens, and assigning Gasteiger charges. Molecular docking was performed using AutoDock Vina with a 40 ų grid with a 0.375 Å spacing box centered on predicted interaction interfaces. Docking was run with an exhaustiveness of 32, and the top-ranking pose was selected based on the lowest binding affinity score. The top-scoring pose (by AutoDock Vina affinity score) was selected, with consistency across ≥ 75% of top 10 poses. In silico prediction of a direct CTCF-YTHDC1 interaction was validated by co-immunoprecipitation in AML cells.

In vitro protein-protein interaction pulldown assay

Full-length cDNAs encoding CTCF and YTHDC1 were subcloned into the pGEX-5X vector (GST-tag; Addgene #27-4584-01) and pET-His vector (His-tag; Addgene #29653), respectively. These recombinant GST-tagged CTCF and His-tagged-YTHDC1 proteins were expressed in E. coli strain BL21 (DE3) and affinity-purified using glutathione sepharose beads (Thermo Fisher Scientific) for GST-CTCF and Ni-NTA magnetic Beads (Thermo Fisher Scientific) for His-YTHDC1. Protein purity and integrity were verified by SDS-PAGE followed by Coomassie blue staining. For in vitro interaction studies, purified GST-CTCF was incubated with His-YTHDC1 in a GST pull-down assay. After extensive washing, bound proteins were eluted from glutathione sepharose beads, resolved by SDS-PAGE, and detected by immunoblotting using anti-His (Abclonal, Cat#AE086), anti-YTHDC1 (Abcam, Cat#ab122340), and anti-GST (Proteintech, Cat#10000-0-AP) antibodies.

Cell cycle analysis

Cells were harvested and washed with ice-cold phosphate buffered saline (PBS). The washed cells were fixed by adding 70% ethanol drop wise to the pellet with vortexing and incubated overnight at 4 °C. After fixation, cells were washed with ice-cold PBS twice. The PBS washed cells were treated with the staining buffer (RNase A, Triton X-100) and then incubated at 37 °C for 30 min. Cells were then stained with 50 µg/mL PI in the dark for 15 min at room temperature. Stained samples were proceeded on the Thermo Attune NxT (Thermo Fisher Scientific), and cell cycle data analysis was performed using FlowJo program. Triplicate experiments were performed for each sample.

Colony-forming assay

Colony-forming unit (CFU) assays were conducted as previously described [50]. Briefly, primary patient-derived AML cells were isolated using CD34⁺ magnetic beads (Miltenyi Biotec, Cat. No. 130-046-703) and pretreated for 24 h with either DMSO (Ctrl), 0.5 µM menin-MLL inhibitor (MLLi) alone, 0.5 µM cytarabine (Ara-C) alone, or MLLi in combination with Ara-C. Subsequently, 2.5 × 10⁴ cells per plate were plated in triplicate in methylcellulose medium (MethoCult™ H4434, STEMCELL Technologies) supplemented with recombinant cytokines: interleukin 3 (IL-3, 20 ng/mL, PeproTech, Cat# 200-03), interleukin 6 (IL-6, 20 ng/mL, PeproTech, Cat# 200-06), erythropoietin (Epo, 20 ng/mL, MCE cat#HY-P7164), thrombopoietin (Tpo, 20ng/mL, MCE cat#HY-P70637) and stem cell factor (SCF, 100 ng/mL, PeproTech Cat# 300-07). Colonies were quantified on day 7. For the serial replating experiments, cells were harvested from the primary plates, washed, counted and replated at an identical density (2.5 × 10⁴ cells/plate) in fresh methylcellulose medium. This replating process was repeated every 7 days for a total of three sequential passages.

APEX-mediated proximity RNA labeling

The APEX-mediated proximity RNA labeling assay was conducted as described previously with minor modifications ( [51, 52]). Briefly, 500 ng/mL doxycycline was added to activate the inducible Cas9-APEX2 expression system for 18–24 h. Subsequently, 500 µM biotin-tyramide phenol (prepared in DMSO) was added directly to cell culture medium and incubated for 30 min at 37 °C, followed by addition of 1 mM hydrogen peroxide to initiate biotinylation. After 60 s of gentle agitation, the medium was rapidly decanted, and the cells were washed three times with 15 ml of ice-cold PBS containing 100 mM sodium azide, 100 mM sodium ascorbate, and 50 mM TROLOX (6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid). Cells were harvested into 15-ml Falcon tubes with ice-cold PBS, centrifuged at 500 x g for 3 min, flash-frozen in liquid nitrogen, and stored at -80 °C.

Purification and sequencing biotinylated RNA

After the APEX labeling reaction, cells were quenched with 2X quenching solution (10 mM Trolox and 20 mM sodium ascorbate in DPBS) once, followed by a 1X quenching solution wash. One 1X quenching solution wash was used to resuspend cells, which were then gently pelleted and lysed in 800 µL of lysis buffer containing 1X quenching reagents (1% Triton, 0.1% SDS, 20 mM Tris-HCl pH 7.4, 150 mM NaCl, 5 mM MgCl2, 5 mM trolox, 10 mM sodium ascorbate, and one tablet (per 10 ml) of cOmplete Mini Protease Inhibitor Cocktail. Lysates were clarified by centrifugation for 10 min at 20,000 x g, 4 °C. Streptavidin beads were equilibrated with lysis buffer for a total of two washes. Lysate was mixed with streptavidin beads at room temperature (RT) for 1 h. Beads were washed twice with lysis buffer. Then, RNA was extracted by adding 500 µL of TRIzol, followed by 200 µL of Chloroform, and then precipitated from the aqueous phase. RNAs were eluted in nuclease free H₂O. RNA sequencing libraries were prepared using the VAHTS^® Universal V8 RNA-seq Library Prep Kit for Illumina (Vazyme, Cat# NR605-01/02). The library was purified, quantified, and sequenced as 150 bp paired-end reads on the Illumina NovaSeq 6000 platform.

RNase treatment of AML cells

In brief, 5 × 10⁶ AML cells were harvested, washed with 1X PBS, and centrifuged at 500 X g for 3 min at room temperature (RT). Cells were gently permeabilized by resuspending cell pellets in PBST (0.05% Tween-20 in PBS) for 10 min on ice. After permeabilization, cells were treated with 150 U RNase H and 200 µg/mL RNase A on rotator for 30 min at RT. These permeabilized cells were then used for 4 C, ATAC, Hi-C, DRIP and ChIP assays.

Xenotransplantation of Patient-Derived Xenografts (PDX)

For the patient-derived xenografts (PDX) assay, primary MLLr + AML cells transduced with control or METTL3-targetting shRNA were injected via tail vein into the 6-8-week- old NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG) mice at a density of 1.8 × 10⁵ cells per mouse. Peripheral blood (PB) was collected and subjected to red blood cell lysis using ammonium chloride. Bone marrow (BM) was harvested from the tibias, femurs, and pelvis. Human CD45 + chimerism in BM and PB was assessed by flow cytometry (Thermo Attune NxT, Thermo Fisher Scientific), and data were analyzed using FlowJo software.

RNA stability assay

To examine RNA stability, cells were cultured with 5 µg/mL of actinomycin D and harvested at the indicted time points. Residual gene expression levels were measured by RT-qPCR, and mRNA half-life was calculated using GraphPad Prism software (GraphPad Software).

Chromatin Immunoprecipitation (ChIP) assay and sequencing data analysis

ChIP assays were performed as previously described [22]. In brief, 5 × 10⁶ cells were crosslinked in 1% formaldehyde for 10 min and quenched with 125 mM glycine for 5 min at RT. Cells were lysed in lysis buffer (50 mM Tris-HCl, pH 8, 10 mM EDTA, 1% SDS, protease inhibitor cocktail), and chromatin DNA was sheared by sonication (Bioruptor). Lysates were incubated overnight at 4 °C with the following target-specific antibodies (Anti-CTCF, Cell Signaling Technology, Cat# 2622; anti-METTL3, Abcam Cat# ab195352; Anti-RAD21, Cell Signaling Technology, Cat# 4321; Anti-YTHDC1, Abcam, Cat#ab122340; Anti-MLL, Thermo Fisher Scientific, Cat#A300-086 A; Anti-AF9 antibody, Thermo Fisher Scientific, Cat#PA5-27797; Anti-H3K4me3, Abclonal, Cat#A22146; Anti-H3K27ac, Abclonal, Cat#A7253) in ChIP buffer (20 mM Tris-HCl pH 8.1, 150 mM NaCl, 2 mM EDTA, 1% TritonX-100 and 0.05% SDS) overnight at 4 °C, with a 10% of aliquiot of each lysate used as input. Each sample was incubated with 50 µL protein A/G magnetic beads (Vazyme, cat# PB101-02) for 2 h at 4 °C and sequentially washed with low-salt washing buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS), high-salt washing buffer (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS), lithium chloride washing buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA, 1% Triton X-100, 250 mM LiCl, 1% sodium deoxycholate), and TE buffer (50 mM Tris-HCl, pH 8.0, 10 mM EDTA). Elution and reverse crosslinking were performed in elution buffer (100 mM NaHCO₃, 1% SDS) with 2 µL proteinase K (10 mg/mL) at 65 °C for 5 h. Chromatin DNA was purified and analyzed by quantitative PCR according to the manufacturer’s guidelines. Results are represented as a percentage of input, with error bars indicating standard deviations (S.D.) from triplicate experiments.

ChIP-DNA libraries were prepared using the Illumina’s VAHTS Universal Pro DNA Library Prep Kit according to the manufacturer’s instructions (Vazyme, cat# ND608-01/02). Briefly, PCR products were treated with the End Prep Enzyme Mix for end-repair, 5’ phosphorylation and dA-tailing in a single reaction, followed by T-A ligation to add adaptors. Adaptor-ligated DNA was purified, ligated with adapter indexes, and amplified. The library was purified, quantified, and sequenced as 150 bp paired-end reads on the Illumina NovaSeq 6000 platform.

For ChIP-seq data analysis, raw reads were trimmed and filtered using Cutadapt (http://cutadapt.readthedocs.io, version 1.2.0) [39] and aligned to the hg19 human reference genome using Bowtie2 with default parameters [41]. PCR duplicates were removed using the samtools-rmdup [53]. SAM files were converted to BAM files and sorted using Samtools. Genome coverage bigwig files were generated for heatmap and aggregation plot using DeepTools (v3.5.2) bamCoverage with the parameter “-normalizeUsing RPKM -binSize” [54]. Genome coverage bigwig files were visualized by the Integrated Genomic Viewer (IGV) [55]. Peak calling was performed using MACS2 [56]. Peak annotation, binding distribution, and de novo motif analysis were conducted using HOMER (v4.10) [57]. Promoters were defined as H3K27ac⁺/H3K4me3⁺ regions within ± 1 kb of RefSeq TSSs, while enhancers were defined as H3K27ac⁺/H3K4me3⁻ regions located ≥ 2 kb away from TSSs, based on MACS2-called peaks. The heatmaps and the co-binding analyses on the genome (including promoter and enhancer regions) were generated with DeepTools (v3.5.2) computeMatrix and plotHeatmap programs (https://deeptools.readthedocs.io/en/develop/index.html). Gene ontology (GO) analysis was performed using DAVID (Version 6.8, https://david.ncifcrf.gov) [45, 58]. Differential peaks analysis was conducted using the DiffBind (v3.2.5) and DESeq2 packages (Cut-off: ≥ 2 fold change; P value ≤ 0.05) in R [59]. Overlapping peak analysis was performed using the BEDTools (v2.29.2) with the “bedtools intersect” program [60]. All these datasets generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE287825.

Chromatin Isolation by RNA Immunoprecipitation (ChIRP) assay

The Chromatin Isolation by RNA Immunoprecipitation (ChIRP) assay was deducted as previously described with modifications [61]. Briefly, 20 million cells were cross-linked in PBS containing 1% glutaraldehyde for 10 min at room temperature with gentle shaking, quenched with 125 mM glycine for 5 min, and washed twice with ice-cold PBS. Cells were lysed in 1mL of cell lysis buffer (50 mM Tris-HCl pH 7.0, 10 mM EDTA, 1% SDS) supplemented with fresh PMSF, DTT, protease inhibitors (P.I.), and Superase-In per 100 mg of cells pellet. DNA was fragmented by sonication with a Bioruptor UCD-200 (Diagenode). Chromatin was diluted in hybridization buffer (750 mM NaCl, 1% SDS, 50 mM Tris-HCl pH 7.0, 1 mM EDTA, 15% formamide) supplemented with fresh DTT, PMSF, P.I, and Superase-In, followed by hybridization with 100 pmole of biotinylated DNA probes targeting MALAT1, with LacZ probes as a negative control. DNA/RNA hybrids were precipitated using 100uL of Streptavidin-magnetic C1 beads and washed five times with washing buffer (2x SSC, 0.5% SDS). Pull-down DNA was isolated via Phenol: Chloroform extraction and ethanol precipitation. ChIRP probes were designed using the Biosearch Technologies ChIRP Designer (https://www.biosearchtech.com/Account/Login?return=/chirp-designer). CHIRP-seq library was prepared for sequencing as described for ChIP-seq library.

Chromosome conformation capture (3C) and circular chromosome conformation capture (4C) assays

The 3–4 C-seq assays were conducted as described previously [62, 63] with minor modifications. For 3 C assay, 1 × 10⁷ cells were cross-linked with 2% formaldehyde for 10 min at RT and quenched with 0.125 M glycine for 5 min. Cells were washed twice with cold PBS and re-suspended in digestion buffer containing 0.3% SDS overnight at 37 °C. Next, 1.8% Triton X-100 was added to sequester the SDS for 1.5 h at 37 °C with shaking. Chromatin was digested with 800U DpnII enzyme (MCE, Cat#HY-KE7012) overnight at 37 °C. The digested chromatin was incubated at 65 °C for 20 min and ligated with 240U T4 DNA ligase (Vazyme, cat# C301-01) at 16 °C for ~ 18 h. Reverse crosslinking was performed with 3U proteinase K (MCE, Cat# HY-108717) at 65 °C for 4 h. Chromatin DNA was purified by phenol extraction, ethanol precipitation, and dissolved in ddH₂O. The 3 C-ligated DNA was analyzed with 3 C-qPCR. For 4 C assays, the purified DNA was treated with 800U EcoR I (MCE, Cat#HY-KE7014) overnight at 37 °C with shaking. The digested DNA was ligated in ligation buffer containing 250U of T4 DNA ligase at overnight 16 °C. 4 C DNA was extracted with phenol: chloroform, purified using a Qiagen PCR kit, and amplified by inverse PCR using bait-specific primers. 4 C-seq libraries were generated by adding barcoded Illumina adapters to each primer (Table S1). The DNA libraries were purified, quantified, and sequenced as 150 bp paired-end reads on the Illumina Novaseq 6000 platform.

4C-seq data analysis

4 C-seq raw data were processed with Cutadapt (http://cutadapt.readthedocs.io, version 1.2.0) to trim the adaptors and remove low-quality reads [40]. Filtered reads were aligned to the human hg19 reference genome using Bowtie2 [64]. The 4 C-seq data were analyzed with the 4cseq_pipeline [65] and normalized with the DESeq2 package in R [66]. Differential chromatin interactions based on “bait” were identified with the DESeq2 package in R language. Spearman correlations for each condition was calculated using R package. The 4 C-seq data have been deposited in the NCBI GEO database (GSE287825).

DNA: RNA Immunoprecipitation (DRIP) and DNA: RNA Immunoprecipitation followed by cDNA (DRIPc) assays

DRIP and DRIPc assays were conducted as previously described [67–69]. Briefly, 5 × 10⁶ AML cells were crosslinked with 1% formaldehyde for 10 min at RT and quenched with 125 mM glycine for 5 min. The chromatin was lysed in lysis buffer (50 mM HEPES-KOH pH 7.5, 140 mM NaCl, 1 mM EDTA pH 8.0, 1% Triton X-100, 0.1% Na-Deoxycholate, 1% SDS) and fragmented by sonication to ~ 300 bp. The fragmented chromatin was treated with proteinase K (10 mg/mL) overnight at 65 °C. Chromatin DNA was extracted by phenol-chloroform, ethanol precipitation, and resuspended in 5 mM Tris-HCl pH 8.5. 2% of each sample was retained as input. RNAseH-treated samples served as a negative control. Purified genomic DNA was incubated with 10 µg of S9.6 antibody (Kerafast, Cat#ENH001) in DRIP buffer (50 mM Hepes/KOH at pH 7.5; 0.14 M NaCl; 5 mM EDTA; 1% Triton X-100; 0.1% Na-Deoxycholate, ddH₂O) and BSA-blocked protein A/G beads overnight at 4 °C with rotation. After washing, DNA was recovered for DRIP-qPCR over R-loop-associated loci. Alternatively, RNA was recovered and purified using RNA TRIzol reagent, then reverse transcription was performed using random hexamers for DRIPc-qPCR.

RNA Immunoprecipitation (RIP) assay

The RIP assays were conducted as previously described [22]. Briefly, 2 × 10⁷ MOLM13 AML cells were crosslinked with 1% (v/v) formaldehyde at RT for 15 min, then quenched with 125 mM glycine for 5 min. Cell pellets were washed with cold 1x PBS, resuspended in lysis buffer (50 mM HEPES-KOH at pH 7.5, 150 mM NaCl, 1 mM EDTA at pH 8.0, 1% (v/v) Triton X-100, 0.1% (v/v) sodium deoxycholate, 5 mM DTT, and 1x protease inhibitor), and incubated on ice for 10 min. The lysed cells were sonicated, and the supernatants were collected by centrifugation at 14,000 g for 10 min at 4 °C. The lysates were incubated overnight at 4 °C with 2–10 ug anti-YTHDC1, METTL3, CTCF, m⁶A (MeRIP), or IgG (Anti-CTCF, Cell Signaling Technology Cat# 2899; Anti-METTL3, Abcam Cat# ab195352; Anti-m⁶A, Abcam Cat# ab151230; Anti-YTHDC1, Abcam Cat#ab122340; Anti-IgG, Abclonal Cat# A1971). Next, 50 µL of prewashed protein A/G magnetic beads (Vazyme, cat# PB101-02) were added to the lysates and incubated at 4 °C for 2 h with rotation. The complexes were washed three times with ice-cold wash buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl,1 mM MgCl2, 0.5% (v/v) NP-40, RNase inhibitor) and eluted with 500 µL SDS-EDTA (50 mM Tris pH 8.0, 100 mM NaCl, 10 mM EDTA, 1% SDS) for 10 min at 65 °C. RNA samples were extracted using TRIzol Reagent, precipitated with isopropanol, washed with 70% ethanol, and eluted in nuclease-free H₂O. The purified RNAs were reverse-transcribed to generate cDNA for RIP-qPCR and RIP-seq. The RIP-seq and MeRIP-seq datasets were deposited in the NCBI GEO under accession number GSE287825.

Assay for Transposase-Accessible Chromatin using Sequencing (ATAC-seq)

The ATAC-seq assay was performed as previously described [70]. Briefly, 5 × 10⁴ cells were washed twice with PBS and lysed in lysis buffer containing 10 mM Tris-HCl (pH 7.4), 10 mM NaCl, 3 mM MgCl2, and 0.1% NP-40. The lysate was washed in PBS and treated with Tn5 transposase at 37 °C for 30 min. DNA was purified using the DNA Purification Kit (TIANGEN, Cat# DP214). DNA library was generated using the Hyperactive ATAC-Seq Library Prep Kit for Illumina (Vazyme, cat# TD711-01). Quality control was performed using the Qubit and Agilent Bioanalyzer, and the libraries were subjected to paired-end sequencing with a read length of 150 bp on the Illumina Novaseq 6000 platform.

ATAC-seq data analysis

Raw sequencing reads were trimmed and filtered with Cutadapt (http://cutadapt.readthedocs.io/, version 1.2.0), PCR duplicates were removed using Picard MarkDuplicates (version 2.0.1), and mitochondrial reads were removed using Samtools [71]. High-quality reads were aligned to the human (hg19) reference genome using Bowtie2 with default parameters [64]. SAM files were used to BAM files and sorted using Samtools. ENCODE blacklist regions were used to filter the data (https://sites.google.com/site/anshulkundaje/projects/blacklists). Peak calling was performed using MACS2 with parameters (-g hs -p 1e-9 –nolambda -f BAMPE –nomodel –shiftsize 100 –extsize 200) [56]. Genome coverage bigwig files for heatmaps and aggregation plots were generated suing DeepTools (v3.5.2) bamCoverage with the parameter “-normalizeUsing RPKM -binSize” [54]. Sequencing tracks were visualized using the Integrated Genomic Viewer [55]. Peak annotation was performed using HOMER [57]. Differentially accessible sites were identified using DEseq2 (Benjamini-Hochberg adjusted p < 0.05; FoldChange (FC) > 2) in R. The ATAC-seq datasets were deposited in the NCBI GEO under accession number GSE287825.

The high-resolution chromatin conformation capture (Hi-C) assay

In situ Hi-C was performed as described previously with some modifications [72]. Briefly, 5 × 10⁶ cells were crosslinked with 1% formaldehyde for 10 min and quenched with 125 mM glycine for 5 min at RT. Cells were lysed in lysis buffer (10 mM Tris-HCl (pH 8.0 at 25 °C), 10 mM NaCl, 0.5% NP-50, 0.5% SDS, and 1x protease inhibitor) at 60 °C for 5 min and then immediately placed on ice. For chromatin digestion, 800U DpnII was added to the lysate and incubated overnight at 37 °C. Digested fragments were biotinylated at 37 °C for 4 h by adding biotin-14-dATP, 10 mM dCTP, 10 mM dGTP, 10 mM dTTP, and DNA polymerase I (Klenow fragment) to the reaction. The lysate was incubated with 1xT4 ligase buffer, 10% Triton-X-100, 12 µL of 10 mg/mL BSA, 5 µL T4 DNA ligase (Vazyme, cat# C301-01) at RT for 4 h. For reverse-crosslinking, 100 µL of 10 mg/ml proteinase K, 120 µL of 10% SDS, and 130 µL of 5 M NaCl were added to the digested chromatin and incubated at 68 °C overnight. The DNA was isolated with ethanol precipitation and dissolved in Tris-buffer (10 mM Tris-HCl, pH = 8.0) for sonication. To purify the biotin-labeled ligation junctions, the samples were incubated with pre-washed the VAHTS CA-28 Streptavidin Beads (Vazyme, cat# N512-01/02) for 45 min in 2× binding buffer (10 mM Tris-HCl (pH = 7.5), 1 mM EDTA, 2 M NaCl). The beads were washed twice in 1× Tween Wash Buffer (5 mM Tris-HCl pH = 7.5, 0.5 mM EDTA, 1 M NaCl, 0.05% Tween 20) at 55 °C for 2 min. A-tailing was then performed by incubating the beads with dATP and Klenow exo at 37 °C for 30 min. The adapters were ligated by incubating the beads in a mixture of 1×ligation buffer, T4 ligase, and 3 µL of Illumina indexed adapter for 15 min at RT. Library preparation was performed using the DNA Library Prep Kit, amplified for 10–12 cycles, and size-selected with AMPure XP beads (Beckman Coulter, #A63881). Libraries were subjected to paired-end sequencing with a read length of 150 bp on the Illumina Novaseq 6000 platform.

Hi-C sequencing data analysis

Raw sequencing reads were trimmed using HOMER (version 4.10) [57], and PCR duplicates were removed using Picard MarkDuplicates (version 2.0.1). High-quality filtered reads were aligned to the human reference genome (hg19) using Bowtie2 with parameters ‘‘-n 1 -m 1 -p 8’’ [64]. Aligned sequencing reads were used to generate a contact matrix using HOMER (version 4.10). Contact matrix normalization, Hi-C correlation matrix generation, principal component analysis (PCA), and significant interaction identification were performed as previously described [73]. In brief, a normalized and visualizable interaction matrix was generated using the HOMER (version 4.10) analyzeHiC program with default parameters “-res 10,000 -superRes 20,000 -pos chromosome location”. PCA analysis of the normalized interaction matrix were conducted using the runHiCpca.pl program with the parameter “-res 10,000 -cpu 8 -genome hg19” in R. TAD domain scores were normalized by subtracting the mean of all TADs, and quantile normalization was applied to domain scores using the HOMER program (v4.10). Differential TADs with significantly different domain scores were identified using ANOVA (cutoff: Bonferroni-corrected p value < 0.05). Differential chromatin interactions (loops) were identified using HiCExplorer (version 3.5.3) [74]. The Hi-C heatmaps of the normalized and visualized chromatin interaction matrices were generated using juicer (version 1.5.5) and visualized with Juicebox [75] and Java TreeView [76]. All Hi-C datasets were deposited in the NCBI GEO under accession number GSE287825.

Quantification and statistical analysis

Statistical analyses were performed using the Student’s t-test or analysis of variance (ANOVA), followed by appreciate multiple comparison tests. Significance levels are denoted as follows: *p < 0.05, **p < 0.01, and ***p < 0.001*. For in vivo experiment, sample size chosen was determined using the generalized linear model with Bonferroni correction for multiple comparisons. Animals were randomly assigned to experimental groups. For in vitro experiments, a minimum of three independent biological replicates per condition/genotype were performed, each including technical replicates where applicable.

Results

METTL3 interacts with the CTCF/Cohesin complex and R-loop-associated proteins in AML cells

Recent studies have demonstrated that METTL3 is closely associated with AML progression; however, its specific contributions to genome organization and regulation remain poorly understood [33, 35, 77]. To investigate the function of the nuclear m⁶A writer METTL3 in AML cells, we isolated METTL3-associated proteins using immunoprecipitation followed by liquid chromatography - tandem mass spectrometry (LC-MS/MS) in MOLM13 AML cells, using IgG-MS as a negative control (Fig. 1A; Table S2). METTL3 was found to specifically interact with CTCF, the cohesin complex (RAD21, SMC1A), m⁶A modifiers (METTL14, YTHDC1, WTAP), and R-loop-associated proteins (DHX9, PARP1, DHX15) (Fig. 1B). Gene ontology (GO) and protein enrichment analyses indicated that METTL3-associated proteins are primarily involved in chromatin structure, genome organization, cohesin complex assembly, RNA stabilization/splicing, cell cycle regulation, DNA/RNA binding, and methyltransferase activity (Fig. 1C, D). Co-immunoprecipitation (CoIP) followed by Western blot (WB) analysis confirmed the specific binding of METTL3 to YTHDC1, METTL14, CTCF, and DHX9, while no interaction was observed for the negative control GAPDH (Fig. 1E).

Fig. 1.

METTL3 interactome identified by immunoprecipitation-Mass Spectrometry (IP-MS) in AML cells. A. Coomassie blue staining of proteins purified from MOLM13 cells using METTL3 or IgG antibodies. B. Partial list of METTL3-associated proteins identified by LC-MS/MS in AML cells, with IgG as a negative control. C. GO analysis of METTL3-associated proteins in AML cells. D. Overrepresentation analysis of METTL3-associated proteins showing the enrichment classified by the DAVID database. Ratios indicate protein proportions in the METTL3-associated proteome; statistical significance ranked by Benjamini-Hochberg-corrected p value ( ≦ 0.05). E. Validation of METTL3-associated proteins by immunoprecipitation followed by western blot (IP-WB), with GAPDH as a negative control

Notably, the METTL3 interactome demonstrated strong enrichment of YTHDC1, which plays a critical role in AML leukemogenesis through m⁶A-dependent RNA modification. To further investigate the role of YTHDC1 in AML, we conducted the YTHDC1 IP-MS assay to identify the YTHDC1 interactome in MOLM13 cells, with IgG serving as the negative control (Fig. S1A; Table S2). Analysis of the YTHDC1 interactome reveals that YTHDC1 specifically interacts with CTCF, the cohesin complex (RAD21, SMC3, SMC1A), m⁶A modifiers (METTL3, METTL14, YTHDF1), and R-loop-associated proteins (DHX9, DHX16, DDX24) (Fig. S1B). These enriched proteins are linked to chromatin organization, the cohesin complex assembly, RNA splicing, cell cycle regulation, RNA binding, and WNT signaling (Fig. S1C, D). Strikingly, YTHDC1 exhibits a strong interaction with CTCF, a key architectural protein that regulates oncogenic TADs and drives the expression of leukemic signature genes, such as HOXA9 and CTNNB1 [11, 22]. Additionally, this interaction was further supported by STRING database analysis and molecular docking, which identified the putative binding domains between YTHDC1 and CTCF (Fig. S1E, F). Next, to biochemically validate the interaction between METTL3 and CTCF, we performed in vitro GST pull-down assays. Using recombinant proteins, we found that GST-tagged-CTCF weakly precipitated His-tagged METTL3 in the absence of YTHDC1, while GST-tagged empty vector as the control (Fig. S1G). However, robust co-precipitation was observed upon the addition of His-tagged YTHDC1, confirming that YTHDC1 mediates the METTL3-CTCF interaction (Fig. S1G). Furthermore, a direct protein-protein interaction between CTCF and YTHDC1 was demonstrated, as GST-CTCF successfully pulled down His-YTHDC1 in the absence of other cellular factors, while GST-tagged empty vector as the control (Fig. S1H). These findings suggest that METTL3 collaborates with YTHDC1 to regulate the expression of leukemic signature genes by interacting with CTCF/Cohesin complex and R-loop-associated proteins, thereby modulating genome organization in AML.

METTL3 plays a crucial role in regulating the leukemic transcription program and AML cell proliferation

METTL3, the core methyltransferase in the m⁶A modification system [33, 78], governs the hematopoietic stem/progenitor cells (HSPCs) differentiation and drives leukemic transformation through its oncogenic activity [31, 35]. To investigate the role of METTL3 and YTHDC1 in HSPCs and leukemic stem cells (LSCs), we reanalyzed the NCBI GEO dataset (GSE74246) to examine their expression patterns across the hematopoietic stem cell (HSC) differentiation hierarchy and in AML leukemic stem/progenitor cells [79]. We observed that the expression of METTL3 and YTHDC1 is highly upregulated in early-stage hematopoietic cells (HSCs, MPPs), but decreases in committed hematopoietic progenitors (CMPs, MEPs, and GMPs) (Fig. S2A). Strikingly, the expression of METTL3 and YTHDC1 is markedly elevated in pre-leukemic HSCs (pHSCs) and LSCs compared to their levels in more differentiated leukemia blasts (Fig. S2B). Although the expression of METTL3 and YTHDC1 was modestly lower in LSCs than in pHSCs, it remained significantly higher in both MOLM13 AML cells and primary MLLr + AML LSCs (CD34⁺CD38⁻ fractions) relative to normal CD34 + HSCs (Fig. S2C, S2D). This expression pattern suggests that METTL3 and YTHDC1 may play critical roles in regulating both hematopoiesis and AML leukemogenesis.

To further explore the role of METTL3 and YTHDC1 in AML, we analyzed the TCGA-LAML datasets and observed significantly higher the expression of METTL3 and YTHDC1 in AML patients compared to normal healthy individuals (Fig. S2E). Notably, the expression levels of METTL3 and YTHDC1 showed positively correlated with established AML signature genes, such as MYB and RUNX1, within this AML cohort (Fig. S2F). Critically, AML patients with lower METTL3 or YTHDC1 expression exhibited significantly longer overall survival compared to those with higher expression of these genes (Fig. S2G).

To assess the impact of METTL3 and YTHDC1 on AML cell proliferation, we knocked down (KD) METTL3 using METTL3-specific shRNA in human AML cells, as well as performed YTHDC1 KD. Our findings revealed that depletion of METTL3 or YTHDC1 significantly suppresses cell proliferation in MOLM13 cells (Fig. S2H, I) and MV4-11 cells (Fig. S2J) compared to scramble shRNA controls (Ctrl). Importantly, the proliferative defect in YTHDC1-depleted cells was rescued by re-expression of METTL3 (Fig. S2K), suggesting functional interplay. Furthermore, depletion of METTL3 or YTHDC1 induces a distinct cell cycle arrest in the G1 phase and significant reduction in the G2/M phase population in both MOLM13 cells (Fig. 2A) and primary AML cells (Fig. 2B). We next assessed the effect of METTL3 or YTHDC1 on the self-renewal. METTL3 or YTHDC1 depletion severely impairs the serial replating capacity of MOLM13 cells (Fig. 2C) and primary AML cells (Fig. 2D) over three rounds, and significantly reduces the proliferation of primary AML cells in vitro (Fig. 2E). Critically, this translated to a reduced disease burden in vivo. Mice transplanted with METTL3-depleted primary AML cells showed a significant reduction of hCD45 + chimerism in the bone (BM) and peripheral blood (PB) of recipient mice (Fig. S2L). Consistently, while all mice transplanted with control cells succumbed to disease within ~ 30 days, those receiving METTL3-knockdown cells exhibit a significant survival advantage (Fig. 2F). Collectively, these data indicate that depletion of METTL3 and YTHDC1 is critical for AML cell proliferation, self-renewal, and in vivo leukemogenesis.

Fig. 2.

Loss of METTL3 or YTHDC1 disrupts leukemic transcriptional programs and AML cell proliferation. A. Cell-cycle profiles of Ctrl, shMETTL3, and shYTHDC1 MOLM13 cells. B. Cell-cycle profiles of Ctrl, shMETTL3, and shYTHDC1 primary AML cells. C. Serial replating colony-forming unit (CFU) assays of control or shMETTL3 MOLM13 AML cells are shown. Colonies were replated every 7 days for 3 rounds. D. Serial replating colony-forming unit (CFU) assays of control or shMETTL3 primary AML cells are shown. Colonies were replated every 7 days for 3 rounds. E. Proliferation curves of primary MLLr + AML cells transduced with control (shScramble, Ctrl) or two independent METTL3-targeting shRNAs. F. Kaplan-Meier survival analysis of recipient mice receiving control or shMETTL3 primary AML cells (p < 0.05, log-rank test; n = 4 mice per group). G. Heatmap of RNA-seq analysis showing the differentially expressed genes (DEGs) in Ctrl vs. shMETTL3 or shYTHDC1 MOLM13 AML cells (two independent experiments). Arrows highlight DEGs. H. RT-qPCR validation of the expression of the leukemia-associated genes in control, shRNA-mediated METTL3- or YTHDC1-depleted MOLM13 cells. I. Western blotting validation of METTL3 expression in Ctrl and shMETTL3 MOLM13 cells. J. Western blotting validation of YTHDC1 expression in Ctrl and shYTHDC1 MOLM13 cells. K. Venn diagram of overlap DEGs between shMETTL3 or shYTHDC1 vs. Ctrl MOLM13 AML cells. L. GO analysis of these overlapping DEGs altered upon METTL3 or YTHDC1 depletion. M. GSEA enrichment analysis showing the downregulated AML-associated pathways and cell cycle process in overlapping DEGs of shMETTL3 or shYTHDC1 vs. Ctrl. N. RT-qPCR analysis of the expression of the leukemia-associated genes in Ctrl and shMETTL3 MOLM13 cells. Data are represented as the mean ± SD of total viable cells from three independent experiments; two-tailed Student’s t-test; *P < 0.05, **P < 0.01, ***P < 0.001

To elucidate the impact of METTL3 or YTHDC1 depletion on genome-wide transcriptome changes, we performed RNA-seq on control (Ctrl), shYTHDC1 and shMETTL3 MOLM13 cells (Fig. 2G). Depletion of METTL3 or YTHDC1 significantly downregulates the transcription of AML signature genes, including MYC, MYB, CTNNB1, RUNX1, and STAT5B, a finding validated by RT-qPCR (Fig. 2H). Western blotting analysis confirmed the efficient knockdown of METTL3 or YTHDC1 at the protein levels (Fig. 2I, J). Loss of METTL3 or YTHDC1 substantially suppresses AML cell growth and induces cell cycle arrest, with concurrent downregulation of leukemia signature genes (e.g., MYB, RUNX1, HOXA9) that are known to promote proliferation and cell cycle progression. These findings suggest that downregulation of such signature genes may contribute to the observed phenotypes. Notably, we observed a substantial overlap in the differentially expressed genes (DEGs) between shMETTL3 and shYTHDC1 conditions compared to control (Ctrl) MOLM13 AML cells (Fig. 2K). Gene Ontology (GO) analysis of these shared DEGs revealed enrichment for terms related to chromatin structure, DNA/RNA binding, AML pathogenesis, cell cycle regulation, pathways in cancer, Wnt signaling, and PI3K-AKT signaling (Fig. 2L). Consistence with this, Gene Set Enrichment Analysis (GSEA) showed significant downregulation of gene sets linked to cell cycle process, AML progression, methylation, and WNT signaling in METTL3- or YTHDC1-depleted cells (Fig. 2M; Fig. S2M). The downregulation of key leukemic drivers such as MYB, RUNX1, and HOXA9 genes, which are known to directly promote proliferation and cell cycle progression [80, 81], thereby providing a plausible mechanistic explanation for observed phenotypes of suppressed cell growth and G1/S arrest.

To determine the dependency on METTL3’s catalytic activity, we expressed a well-characterized catalytic dead mutant (METTL3-D395A; METTL3-MUT) [82]. While re-expression of wide-type METTL3 (METTL3-WT) in shMETTL3 cells fully rescues the expression of AML signature genes, the METTL3-MUT construct fails to do so (Fig. S2N), underscoring the requirement for its methyltransferase function. Finally, we confirmed the broad relevance of this axis by demonstrating that depletion of METTL3 or YTHDC1 also significantly reduces the expression of these key AML signature genes (MYC, MYB, RUNX1, CDK6 and CTNNB1) in MV4-11 cells (Fig. S2O) and, most importantly, in primary MLLr + AML cells (Fig. 2N). Collectively, these results demonstrate that METTL3 and YTHDC1 are critical for sustaining the expression of a leukemic transcriptional program and that its loss substantially suppresses AML cell growth and cell cycle progression through the coordinated downregulation of critical oncogenic drivers.

MLL regulates the expression of METTL3 and YTHDC1 in MLLr + AML cells

METTL3 and YTHDC1 play critical roles in regulating the expression of leukemia-associated genes in the AML genome (Fig. 2). AML with rearrangements of lysine methyltransferase 2 A (KMT2A), previously known as mixed-lineage leukemia (MLL), is associated with poor prognosis [6]. We observed that the expression of METTL3, YTHDC1 and MLL (KMT2A) exhibits highly correlated expression across early-stage hematopoietic cells (HSCs, MPPs), pre-leukemic HSCs (pHSCs) and LSCs (Fig. S2A), suggesting shared transcriptional regulation in both normal and malignant hematopoiesis. To determine whether this co-expression of METTL3 and YTHDC1 translates to specific overexpression in MLL-rearranged (MLLr+) AML, we analyzed transcriptomic data from the TCGA-LAML and TARGET-AML datasets. METTL3 and YTHDC1 are significantly higher in MLLr + AML patients compared to both healthy donors and patients with AML subtypes (e.g., NPM1C+, FLT3-ITD, RUNX1-ETO and other MLLr- AML) (Fig. 3A; Fig.S3A). This MLLr + specific upregulation is validated in primary patient samples by RT-qPCR (Fig. S3B, 3 C; Table S3). Furthermore, the expression levels of METTL3 and YTHDC1 are highly elevated in MLLr + AML cell lines compared to MLLr- AML cell lines within the DepMap database (https://depmap.org/portal/) (Fig. S3D). Consistent with a critical role in this subtype, the genome-wide CRISPR screening data (DepMap database) reveals that depletion of METTL3 or YTHDC1 more significantly impairs the proliferation of MLLr + AML cell lines (e.g., MOLM13 and THP-1) than that of MLLr- AML cell lines (Fig. S3E).

Fig. 3.

METTL3 and YTHDC1 are transcriptionally activated by MLL in MLLr + AML cell. A. The expression levels of METTL3 was analyzed across MLLr+, NPM1C+, FLT3-ITD+, RUNX1-ETO, other MLLr- AML, and healthy samples from the TCGA-LAML and TARGET-AML datasets. B. The MLL ChIP-seq tracks at the promoters of METTL3 and YTHDC1 promoters in MOLM13 AML cells from NCBI GEO dataset (GSE127507). C. Proliferation curves were generated for primary MLLr + AML cells treated with DMSO (Ctrl), Cytarabine (ara-C), Menin-MLL inhibitor (MLLi), or a combination of ara-C and MLLi. D. The number of colonies were calculated in primary MLLr + AML cells treated with DMSO (Ctrl), ara-C, MLLi, or a combination of ara-C and MLLi. E. RT-qPCR analysis showing METTL3 expression in MLLr+ (MOLM13, THP1, MV4-11) and MLLr- (HL60, OCI-AML2, and OCI-AML3) AML cell lines treated with 0.5µM MLLi for 16 h. F. RT-qPCR analysis assessing the expression of leukemia-related genes in primary MLLr + AML cells treated with DMSO (Ctrl), ara-C, MLLi, or a combination of ara-C and MLLi. G. RT-qPCR analysis of METTL3 expression in primary MLLr + AML patient samples treated with DMSO or 0.5µM MLLi for 16 h. The black bars and lines represent the mean value of each group. H. RT-qPCR analysis of METTL3 expression in primary MLLr- AML patient samples treated with DMSO or 0.5µM MLLi for 16 h. The black bars and lines represent the mean value of each group. I Cell proliferation analysis of control (Ctrl) MOLM13 cells and overexpression of MLL-WT or MLL-AF9 MOLM13 cells. J. The expression levels of METTL3 and YTHDC1 in Ctrl, MLL-WT- or MLL-AF9-overexpressed MOLM13 cells. K. ChIP-qPCR showing MLL binding at the promoters of METTL3 and YTHDC1 in Ctrl, MLL-WT- or MLL-AF9-overexpressed MOLM13 cells. Data are presented as mean ± SD; statistical significance is determined by the Student’s t-test (*p < 0.05, **p < 0.01, ***p < 0.001, n = 3 independent experiments)

Previous studies have indicated that Menin, a critical oncogenic cofactor of MLL, regulates oncogene expression by binding to the leukemia-associated gene promoters in the MLLr + AML genome [83, 84]. The interaction of Menin and MLL can be therapeutically disrupted by the Menin inhibitor Revumenib (SNDX-5613) in MLLr + AML cells [83, 84]. To investigate whether METTL3 and YTHDC1 are transcriptionally regulated by MLL in MLLr + AML, we reanalyzed MLL ChIP-seq data from Gene Expression Omnibus (GEO) dataset (GSE127507) [85]. Remarkably, MLL is highly enriched at the promoters of METTL3 and YTHDC1 genes (Fig. 3B), suggesting direct transcriptional regulation in MLLr + MOLM13 cells. Consistent with this, inhibition of the Menin-MLL interaction with a small-molecule inhibitor (MLLi) significantly suppresses MOLM13 AML cell proliferation (Fig. S3F). We next asked if targeting this axis could synergize with standard chemotherapy. The combination of cytarabine (ara-C) with MLLi exhibits synergistic anti-leukemic activity, reducing primary AML cell growth and colony formation more effectively that either agent alone (Fig. 3C, D). In contrast to their robust effects on MLLr + AML cells, ara-C or MLLi monotherapy exhibits minimal impacts on the proliferation and colony formation of normal CD34 + HSPCs (Fig. S3G, H). The combination treatment of ara-C and MLLi shows only a modestly inhibitory effect on CD34 + HSPCs (Fig. S3G, H). As expected, MLLr + AML cells (MOLM13, MV4-11, and THP-1) exhibit higher basal METTL3 expression and greater sensitivity to MLLi treatment compared to MLLr- AML cells (HL60, OCI-AML2, and OCI-AML3) (Fig. 3E). Mechanistically, treatment with MLLi alone reduces the expression of key AML signature genes (e.g., RUNX1, MYB, MYC, CCND1, and CDK6) in primary AML cells, with the combination treatment yielding the greatest suppression (Fig. 3F). Crucially, MLLi treatment significantly downregulates the expression of METTL3 in the primary MLLr + AML patient samples (Fig. 3G), but has minimal effects in primary MLLr- AML samples (Fig. 3H). To further investigate the specific role of the MLL fusion protein in AML, we transduced MOLM13 cells with Flag-tagged constructs (empty vector, MLL-WT, and MLL-AF9). Successful ectopic expression was confirmed by Western blot (Fig. S3I). We found that ectopic expression of the leukemogenic MLL-AF9 fusion, but not wild-type MLL(MLL-WT), promotes MOLM13 cell proliferation (Fig. 3I) and upregulates both METTL3 and YTHDC1 expression (Fig. 3J). MLL ChIP-qPCR demonstrates that this effect is mediated by the highly direct binding of MLL-AF9 to the METTL3 and YTHDC1 promoters (Fig. 3K). Critically, AF9 ChIP-qPCR data indicates the specific recruitment of the MLL-AF9 fusion protein to the promoters of METTL3 and YTHDC1, but not wild-type MLL(MLL-WT) and Ctrl groups (Fig.S3J). Consequently, MLLi treatment substantially reduces the expression of METTL3, YTHDC1, RUNX1, and MYB in MOLM13 cells (Fig. S3K) and impairs METTL3 chromatin occupancy at the promoters and enhancers of RUNX1 and MYB, as measured by METTL3 ChIP-qPCR (Fig. S3L). Collectively, these findings demonstrate that METTL3 and YTHDC1 are transcriptionally activated by MLL in MLLr + AML, establishing a causal link between the oncogenic driver and the m⁶A machinery to promote leukemogenesis.

METTL3 directly occupies a subset of CTCF binding sites (CBSs) in an R-loop-dependent manner in the MLLr + AML genome

Although METTL3 interacts with CTCF/Cohesin complexes (Fig. 1B-D), the mechanism by which it regulates CTCF-dependent chromatin organization in the AML genome remains unclear. To investigate how METTL3 influences chromatin organization and leukemia-associated gene expression, we performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) for METTL3, CTCF, RAD21, H3K4me3, and H3K27ac in the MLLr + MOLM13 cells. Functional regulatory elements are defined using H3K4me3 (a marker of active promoters) and H3K27ac (a marker of active enhancers) ChIP-seq peaks [86, 87]. By integrative analysis of these ChIP data, we found that METTL3 binding sites in AML genome are predominantly localized to functional regulatory regions, including promoters (43%) and enhancers (32%), with a smaller fraction in gene bodies (Fig. 4A). These METTL3-bound sites are primarily associated with protein coding genes (79%) and long non-coding RNAs (lncRNAs, 13%), with minimal overlap with small nuclear RNAs (snRNAs,1%) or microRNAs (miRNAs, 3%) (Fig. S4A). Notably, Gene Ontology (GO) analysis indicates that METTL3-bound targets are mainly enriched in processes critical for leukemogenesis, including cell-cycle regulation, RNA splicing, DNA/RNA binding, methyltransferase, DNA/RNA helicase activity, chromatin organization, and WNT signaling (Fig. S4B). To determine whether METTL3 directly influence the leukemic transcription program through its genomic binding, we performed de novo motif analysis of METTL3 binding sites (Fig. 4B). Strikingly, the top-identified transcription factor (TF) motifs central to hematopoietic malignancy, including GATA, ETS1, RUNX1, MYB, E-box, and notably, CTCF motifs, are associated with the leukemic gene regulation (Fig. 4B). This suggests that METTL3 potentially co-binds with CTCF to mediate the transcription of leukemia-associated genes. Critically, the top METTL3 binding motifs also contain NFY and KLF9 (Fig. 4B), consistent with prior report [35].

Fig. 4.

METTL3 co-localizes with a subset of CTCF binding sites (CBSs) in an R-loop-dependent manner in the AML genome. A. Genome-wide distribution of METTL3 binding in the MLLr + AML genome. B. De novo motif analysis of METTL3 ChIP-seq peaks; significance determined using a hypergeometric test. C. Integrative analysis of METTL3, CTCF, RAD21, H3K4me3, H3K27ac (ChIP-seq), and R-loop (DRIP-seq) in MOLM13 cells, with IgG as the negative control. D. ChIP-seq and DRIP-seq tracks showing METTL3, CTCF, RAD21, H3K4me3, H3K27ac, and R-loop binding profiles at the RUNX1 locus in MOLM13 cells. E. ChIP-seq and DRIP-seq tracks showing METTL3, CTCF, RAD21, H3K4me3, H3K27ac and R-loop binding profiles at the MYB locus in MOLM13 cells. F. Pie chart showing the classification of R-loop-associated genes in the MLLr + AML genome. G. Overlapping peaks of CTCF, METTL3, and R-loop-occupied sites in MOLM13 cells. H. GO enrichment analysis of genes associated with overlapping CTCF, METTL3, and R-loop occupied sites in MOLM13 cells. I. CTCF ChIP-qPCR analysis of CTCF binding at the RUNX1, MYB, and CDK6 loci in control (Ctrl) and shMETTL3 MOLM13 cells. J. CTCF ChIP-qPCR analysis of CTCF binding at the RUNX1, MYB, and CDK6 loci in DMSO (Ctrl) and MLLi-treated MOLM13 cells. K. CTCF ChIP-qPCR analysis of CTCF binding at the promoters of RUNX1 and MYB in Ctrl, MLL-WT- or MLL-AF9-overexpressed MOLM13 cells. L. CTCF ChIP-qPCR analysis of CTCF binding at the MYB and RUNX1 promoters in Ctrl, shYTHDC1 cells, and shYTHDC1-OE-METTL3 (shYTH-OE-MET) cells. M. DRIP-qPCR analysis of R-loops at the promoter and enhancer of the RUNX1 locus in Ctrl, shMETTL3, and shYTHDC1 MOLM13 cells. N. DRIP-qPCR analysis of R-loops at the promoters of RUNX1 and MYB in Ctrl and shYTHDC1 MOLM13 cells. O. Co-IP followed by western blotting showing the interaction between CTCF and METTL3 in Ctrl and shYTHDC1 MOLM13 cells. Data are presented as the mean ± SD; two-tailed t-test; *P < 0.05, **P < 0.01, ***P < 0.001; n = 3 independent experiments

To define functional regulatory elements, we classified ATAC-seq peaks using H3K4me3 and H3K27ac ChIP-seq data, categorizing them as active promoters (H3K4me3⁺/H3K27ac⁺) and active enhancers (H3K4me3⁻/H3K27ac⁺). Subsequently, by integrative analysis of ChIP-seq data from METTL3, CTCF, RAD21, H3K4me3, and H3K27ac, we found that METTL3 co-localizes with CTCF and RAD21 at the promoter and enhancer regions of the AML genome (Fig. 4C). Notably, METTL3 shows strong co-occupancy with CTCF and RAD21 binding sites at the gene promoters and enhancers of key leukemia-associated genes, such as RUNX, MYB, STAT5B, GSK3B, CDK6, and CDK1 (Fig. 4D, E; Fig. S4C).

RNA-mediated DNA-RNA triple-stranded hybrid complexes, known as R-loops, play critical roles in transcriptional regulation in the AML genome [88–90]. Given that METTL3 binds promoters and enhancers of the annotated genes linked to DNA/RNA helicase activity (Fig. 1C, D; Fig. S4B), we hypothesized that METTL3 mediates chromatin organization through facilitating the formation of R-loops at the regulatory elements (e.g., promoters or enhancers) of leukemia-associated genes in AML. To test this, we performed DNA: RNA hybrid immunoprecipitation followed by sequencing (DRIP-seq) in MOLM13 cells. The majority of R-loop peaks are predominantly localized to the promoter and enhancer regions, highlighting their association with active coding genes and lncRNAs (Fig. 4F; Fig. S4D).

Through integrative analysis of a significant proportion of METTL3 binding peaks (39.01%, 1,594 out of 4,086) co-localized with both CTCF binding sites (CBSs) and R-loops (Fig. 4G), suggesting a functional interplay at shared genomic loci in the AML genome. GO enrichment analysis of genes associated with these overlapping CTCF/METTL3/R-loop peaks reveals their involvement in chromatin structure, RNA binding, AML pathogenesis, cell cycle regulation, DNA/RNA helicase activity, methyltransferase activity, and Wnt signaling pathway (Fig. 4H), further supporting the role of METTL3 in CBS boundary regulation via R-loop formation.

Next, we assessed the functional consequence of this co-localization. Depletion of METTL3 or treatment with MLLi significantly reduces CTCF occupancy at the key leukemic gene promoters (RUNX1, MYB and CDK6) in MOLM13 and MV4-11 cells (Fig. 4I and J; Fig. S4E). Furthermore, METTL3 loss also decreases Cohesin complex (RAD21) binding at these loci in MOLM13 cells (Fig. S4F). Notably, expression of the oncogenic MLL-AF9 fusion, but not MLL-WT, markedly enhances CTCF binding to the RUNX1 and MYB promoters (p < 0.05), indicating a fusion-specific mechanism for transcriptional activation (Fig. 4K). The relationship between METTL3 and its reader protein YTHDC1 was found to be codependent but functionally distinct. METTL3 overexpression fails to rescue CTCF occupancy at MYB and RUNX1 promoter in YTHDC1-depleted cells (Fig. 4L), indicating that YTHDC1’s role in stabilizing the CTCF complex is independent of METTL3’s catalytic function. Conversely, METTL3 chromatin recruitment requires YTHDC1, as YTHDC1 knockdown significantly reduces METTL3 binding at the MYB and RUNX1 promoters (Fig. S4G). Reciprocally, depletion of METTL3 also impaired YTHDC1 binding at these sites (Fig. S4H), demonstrating a codependent relationship for their chromatin occupancy.

Next, we examined whether METTL3 occupancy is functionally linked to R-loop formation. Depletion of METTL3 or YTHDC1 significantly reduces the R-loop levels at the RUNX1 and MYB loci in MOLM13, MV4-11 and primary MLLr + AML cells (Fig. 4M, N; Fig. S4I). Critically, this R-loop deficiency is fully rescued by re-expression of wild-type METTL3 (METTL3-WT) in shMETTL3 cells, but not a catalytically dead mutant (METTL3-D395A), establishing the strict dependency on METTL3’s methyltransferase activity (Fig. S4J). Importantly, YTHDC1 depletion significantly disrupts the physical interaction between METTL3 with CTCF (Fig. 4O), positioning YTHDC1 as the critical molecular bridge within this complex.

The METTL3-YTHDC1 axis facilitates CTCF-dependent chromatin interactions and TADs to regulate leukemia-associated genes in the MLLr + AML genome

Our data indicate that METTL3 specifically interacts with YTHDC1 and CTCF in AML cells (Fig. 1B-E), and YTHDC1 interactome data (IP-MS) further implicate it in leukemic gene regulation via cohesin complex in MLLr + AML (Fig. 2D; Fig. S1B-D). To investigate the impact of the METTL3-YTHDC1 axis on 3D genome organization and its role in establishing the leukemic transcription profile, we performed high-throughput chromosome conformation capture sequencing (Hi-C-seq) in MLLr + MOLM13 cells. Although depletion of YTHDC1 or METTL3 did not dramatically disrupt global TAD organization, it significantly altered local topology, resulting in the loss of 58 and gain of 44 TADs upon YTHDC1 knockdown, and the loss of 67 and gain of 53 TADs upon METTL3 knockdown (Fig. 5A; Fig. S5A). Strikingly, 83 (81.3%) of these TADs are commonly altered by METTL3 and YTHDC1 knockdowns (Fig. S5B), demonstrating a convergent role for the METTL3-YTHDC1 axis in maintaining genome topology (Fig. S5B). Furthermore, GO analysis of genes within these altered TADs shows the enrichment in pathways related to cell cycle regulation, cell differentiation, RNA splicing, chromatin structure, transcription dysregulation, WNT signaling, AML pathogenesis, and methyltransferase activity (Fig. 5B). Furthermore, depletion of METTL3 or YTHDC1 significantly disrupts TAD boundaries at key leukemia signature gene loci, including MYB, RUNX1, CDK6, and STAT5B (Fig. 5C, D; Fig. S5C), underscoring the critical role of the METTL3-YTHDC1 axis in oncogenic 3D genome organization and gene expression.

Fig. 5.

The METTL3-YTHDC1 axis maintains CTCF-dependent TAD integrity via R-loops in the MLLr + AML genome. A. Differential TADs identified via Hi-C in control (Ctrl) and YTHDC1-depleted MOLM13 cells. The domain scores of altered TADs were normalized with Hi-C signals. ANOVA analysis with Bonferroni-corrected p value ≦ 0.05. B. GO enrichment analysis of genes encompassed by the TADs commonly altered by METTL3 and YTHDC1 knockdowns. C. Hi-C interaction map showing the MYB locus in Ctrl, shMETTL3, and shYTHDC1 cells. CTCF-bound TAD boundaries are indicated by black arrows. D. Hi-C interaction map showing the RUNX1 locus in Ctrl, shMETTL3, and shYTHDC1 cells. CTCF-bound TAD boundaries are highlighted with red arrows. E. Gain or loss of enhancer-promoter (E-P) interactions in Ctrl and shYTHDC1 MOLM13 cells are shown. F. 3 C-qPCR analysis of the enhancer-promoter (E-P) interactions at the RUNX1 locus in Ctrl, shMETTL3 and shYTHDC1 cells. G. 3 C-qPCR analysis of the enhancer-promoter (E-P) interactions at RUNX1 locus in Ctrl, shMETTL3, METTL3-WT- or METTL3-MUT-overexpressing in shMETTL3 (shMET + OEMET-WT or shMET + OEMET-MUT) MOLM13 cells. H. 3 C-qPCR analysis of the enhancer-promoter (E-P) interactions at RUNX1 locus in DMSO (Ctrl) and MLLi-treated MOLM13 cells. I. YTHDC1 depletion (shDC1) reduces CTCF binding, R-loop formation, and chromatin accessibility at the enhancer and promoter regions. J. CTCF, R-loop, ATAC-seq, H3K4me3, and H3K27ac profiles at the MYB (Left) and RUNX1 (Right) loci in Ctrl vs. shYTHDC1 MOLM13 cells. Data are presented as mean ± SD; statistical significance was determined by the Student’s t-test (*p < 0.05, **p < 0.01, n = 3 independent experiments)

We next investigate the role of METTL3 or YTHDC1 in regulating finer-scale chromatin interactions. METTL3 and YTHDC1 knockdown significantly disrupts 15.3% (1,987/12,956) and 14.9% (1,842/12,383) of all chromatin loops, respectively (Fig. S5D, E). Enhancer-promoter (E-P) loops are particularly vulnerable, with 27.1% (1,055/3,892) and 26.6% (989/3,714) significantly weakened upon YTHDC1 or METTL3 loss (Fig. 5E; Fig. S5F). Importantly, a core set of 784 E-P loops are co-dependent on both factors in AML cells (Fig. S5G).

This computational prediction was functionally validated using 3 C and 4 C assays, which confirmed that METTL3 and YTHDC1 depletion significantly impaired E-P interactions at the RUNX1 locus in MOLM13 and MV4-11 cells (Fig. 5F; Fig. S5H). Notably, the impaired enhancer-promoter interaction in METTL3-depleted cells is fully rescued by re-expression of wild-type METTL3 (METTL3-WT; MET-WT) but not a catalytically dead mutation (METTL3-D395A; METTL3-MUT), establishing the strict dependency on m6A methylation activity (Fig. 5G). Furthermore, depletion of YTHDC1 significantly reduces the enhancer-promoter loops in primary AML cells at leukemic loci, including RUNX1, MYB, MEIS1, CDK6 and STAT5B (Fig. S5I). Treatment with MLLi in MOLM13 cell phenocopies this effect at the RUNX1 locus (Fig. 5H), positioning the axis downstream of the oncogenic driver.

Mechanistically, although YTHDC1 interacts with CTCF and Cohesin complex (Fig. S1B-D), its functional link to chromatin binding and R-loop formation remains unclear. We observed that depletion of YTHDC1 significantly reduces CTCF occupancy, R-loop formation, and chromatin accessibility (by ATAC-seq) at the promoter and enhancer regions (Fig. 5I) and specifically at key leukemic gens, including MYB, RUNX1, STAT5B, and CDK6 (Fig. 5J; Fig. S5K). Integrative analysis shows that differential ATAC-seq peaks at MYB or RUNX1 loci highly overlapped with H3K4me3 (promoters) and H3K27ac (enhancers), both of which are reduced in YTHDC1-depleted cells (Fig. 5I and J). This integration suggests that YTHDC1 depletion impacts chromatin accessibility at active regulatory elements marked by these histone modifications (Fig. 5I, J). Collectively, these data establish that the METTL3-YTHDC1 axis drives leukemogenesis by orchestrating CTCF-dependent chromatin architecture. It stabilizes R-loops at active enhancers and promoters to maintain TAD integrity and E-P interactions, thereby enforcing a pro-leukemic transcriptional state. This work reveals a critical mechanistic link between epitranscriptomic regulation and 3D genome organization in MLLr + AML.

Chromatin architectural RNAs (arcRNAs) are required for YTHDC1-mediated genome organization in MLLr + AML

Chromatin-associated RNAs are well-established regulators of 3D genome organization and carcinogenesis [25, 91]. Intriguingly, recent studies suggest that RNA-interacting domains of CTCF are critical for its role in mediating enhancer-promoter interactions and chromatin looping [20, 21]. Given that CTCF-bound regions are enriched with R-loop structures [92], we hypothesized that chromatin architectural RNAs (arcRNAs) and their associated R-loops are key contributors to oncogenic genome organization and transcriptional regulation. R-loops (DNA: RNA hybrid) can be degraded by RNase H treatment [92]. To test this, we treated MLLr + MOLM13 cells with RNase H to degrade R-loops and performed integrated multi-omics analysis of ChIP-seq, ATAC-seq, and DRIP-seq data. Integrative analysis identifies 847 overlapping peaks at promoter/enhancer regions from differential CTCF binding, R-loops formation, and chromatin accessibility according to CTCF ChIP-seq, DRIP-seq, ATAC-seq, H3K4me3 ChIP-seq, and H3K27ac ChIP-seq data by comparing control (Ctrl) with RNase-treated MOLM13 cells (Fig. 6A). GO enrichment analysis of genes associated with these overlapping peaks reveals their enrichment in RNA binding, cell cycle regulation, myeloid differentiation, RNA splicing, chromatin organization, WNT signaling, AML pathogenesis, and DNA/RNA helicase activity (Fig. S6A). Loss of R-loops reduces CTCF binding, chromatin accessibility, and H3K4me3/H3K27ac enrichment at RUNX1 and MYB regulatory regions (Fig. 6B; Fig. S6B). Consistent with the patterns at RUNX1 and MYB, a meta-density plot of all 847 overlapping regions confirmed that R-loop degradation significantly reduces CTCF occupancy and chromatin accessibility at the leukemic loci (Fig. 6C).

Fig. 6.

Chromatin architectural RNAs (arcRNAs) are required for chromatin organization and R-loop formation in the AML genome. A. Overlap analysis showing differential CTCF binding, R-loops, and chromatin accessibility using CTCF ChIP-seq, DRIP-seq, ATAC-seq, H3K4me3, and H3K27ac ChIP-seq data in control (Ctrl) vs. RNase-treated MOLM13 cells. B. Loss of R-loops affects METTL3 binding, CTCF binding, R-loops, and chromatin accessibility at the RUNX1 locus as shown by METTL3 ChIP-seq, CTCF ChIP-seq, DRIP-seq, ATAC-seq, H3K4me3, and H3K27ac ChIP-seq in Ctrl vs. RNase-treated MOLM13 cells. C. Heatmap of CTCF binding, R-loop formation, and chromatin accessibility upon 847 overlapping peaks in Ctrl and RNase-treated MOLM13 cells. D. Differential TADs identified by Hi-C in Ctrl vs. RNase-treated MOLM13 cells. (ANOVA, Bonferroni-corrected p ≤ 0.05). E. Hi-C interaction maps at the RUNX1 locus in Ctrl vs. RNase-treated MOLM13 cells. CTCF-bound TAD boundaries are marked by red arrows. F. CoIP followed by western blot analysis showing R-loop disruption reduces YTHDC1 interactions with CTCF, METTL3, DDX24, and DHX9, with GAPDH as a negative control. G. Overlap binding analysis of YTHDC1-, CTCF-, and R-loop-associated arcRNAs using YTHDC1 RIP-seq, CTCF RIP-seq, and DRIPc-seq in MOLM13 AML cells. H. Pie chart categorizing the overlapping arcRNA-associated genes in the AML genome. I. De novo motif analysis of the top motifs from m⁶A RIP-seq, METTL3 RIP-seq, YTHDC1 RIP-seq and CTCF RIP-seq, with significance determined using a hypergeometric test. J. CTCF RIP-seq, YTHDC1 RIP-seq and m⁶A RIP-seq tracks showing the binding profiles of m⁶A, YTHDC1 and CTCF at the MALAT1 lncRNA locus in control (Ctrl) and YTHDC1-depleted MOLM13 cells. K. RT-qPCR analysis of MALAT1 and PVT1 expression in Ctrl, shYTHDC1, and shMETTL3 MOLM13 cells. L. RT-qPCR analysis of RUNX1 and MYB expression in Ctrl and shMALAT1 MOLM13 cells. M. DRIP-qPCR analysis of R-loops at the promoters of RUNX1 and MYB in Ctrl, shMALAT1, and MALAT1-dm6A MOLM13 cells. N. ChIP-qPCR analysis of CTCF binding at the promoters of RUNX1 and MYB in Ctrl, shMALAT1, and MALAT1-dm6A MOLM13 cells. O. 3 C-qPCR analysis of the enhancer-promoter interactions at the RUNX1 locus in Ctrl, shMALAT1, and MALAT1-dm6A MOLM13 cells. P. MALAT1 ChIRP-seq binding profiles at the RUNX1 promoter in MOLM13 cells. Q. m⁶A-qPCR analysis of m⁶A levels at the MALAT1 and PVT1 in MLLr + AML cells and CD34 + HSC cells. R. RT-qPCR analysis of the expression of MALAT1 and PVT1 in DMSO (Ctrl) and MLLi-treated MOLM13 cells. S. m⁶A-qPCR analysis of m⁶A levels at the MALAT1 and PVT1 in Ctrl, shMETTL3, and MLLi-treated MOLM13 cells. Data are represented as the mean ± SD; two-tailed Student’s t-test; *P < 0.05, **P < 0.01, n = 3 independent experiments

To further investigate whether arcRNA-mediated R-loops are essential for CTCF- dependent 3D chromatin organization in the AML genome, we analyzed Hi-C seq data from control (Ctrl) and RNase-treated cells. Loss of R-loops alters TAD structures, significantly decreasing 318 TADs and increasing 256 TADs in the AML genome (Fig. 6D). Critically, TAD topology at RUNX1, MYB, and STAT5B loci is significantly disrupted in RNase-treated cells (Fig. 6E; Fig. S6C), highlighting the critical role of arcRNA-mediated R-loops in maintaining oncogenic genome organization.

Mechanistically, CoIP followed by western blotting results demonstrated that R-loops disruption diminishes YTHDC1 interactions with METTL3, CTCF, DDX24, and DHX9, with GAPDH serving as a negative control (Fig. 6F). To clarify whether the specific arcRNAs are involved in CTCF-dependent genome organization and R-loop formation in the MLLr + AML genome, we conducted performed METTL3 RIP-seq, YTHDC1 RIP-seq, CTCF RIP-seq, and DRIPc-seq in MOLM13 cells. Notably, METTL3 RIP-seq identifies 12,686 METTL3-associated RNAs in MOLM13 cells, of which 3,564 (70.1%) are shared with YTHDC1-associated RNAs from YTHDC1 RIP-seq (Fig. S6D). YTHDC1 RIP-seq identifies 5087 YTHDC1-associated RNAs (62.7% protein-coding RNAs, 25.15% lncRNAs, 2.44% snRNAs, 5.31% miRNAs) (Fig. 6G; Fig. S6E). CTCF RIP-seq identifies 3980 CTCF-associated RNAs, including 58.21% protein-coding RNAs, 29.23% lncRNAs, 2.84% snRNAs, 5.40% miRNAs (Fig. 6G; Fig. S6F). DRIPc-seq identifies 6,254 R-loop-associated RNAs, composed of protein-coding RNAs (75.48%), lncRNAs (14.10%), snRNAs (1.51%), miRNAs (5.73%) (Fig. 6G; Fig. S6G). Cross-analysis of these datasets identifies554 overlapping arcRNAs that are R-loop-associated, bound by YTHDC1/METTL3, and bound by CTCF (Fig. 6G). These arcRNAs are predominantly protein-coding RNAs (50.67%) and lncRNAs (36.77%) (Fig. 6H) and are linked to chromatin organization, cell cycle regulation, WNT signaling, PI3K-AKT signaling, R-loop processing, and chromatin looping (Fig. S6H).

To further investigate whether these arcRNAs are functionally m⁶A-modified, we conducted m⁶A-specific RNA immunoprecipitation followed by high-throughput sequencing (MeRIP-seq) in MOLM13 AML cells. MeRIP-seq analysis revealed that METTL3-mediated m⁶A modification is enriched in these arcRNAs, and notably, 69.9% (387/554) of m⁶A-modified arcRNAs are co-bound by both METTL3 and YTHDC1 (Fig. S6I; Table S4), demonstrating tight functional coordination within this axis. de novo motif analysis of m⁶A, CTCF and YTHDC1 binding sites identified the m⁶A core motif “GGACU” (Fig. 6I), suggesting that m⁶A modification facilitates the co-binding of METTL3/YTHDC1/CTCF to arcRNAs (Fig. 6I). R-loops passively associated with enhancers and insulators due to transcription activity [93]. Depletion of YTHDC1 reduces CTCF binding and m⁶A enrichment at two specific arcRNA loci: MALAT1 and PVT1 lncRNAs (Fig. 6J; Fig.S6J). In addition, depletion of METTL3 or YTHDC1 significantly decreases the expression of MALAT1 and PVT1 (Fig. 6K).

We next investigated whether MALAT1, as an arcRNA, is required for AML genome. MALAT1 depletion specifically impairs the expression of MYB and RUNX1 genes (Fig. 6L), reduces CTCF occupancy, disrupts R-loop structures, and weakens chromatin interactions at the promoter and enhancer regions of MYB and RUNX1 (Fig. 6M-O). To determine whether MALAT1 regulates the expression RUNX1 or MYB by directly associating with their promoters, we performed Chromatin Isolation by RNA Purification (ChIRP) assay. MALAT1 ChIRP data indicates that MALAT1 directly binds at the promoters of RUNX1 and MYB in MOLM13 cells (Fig. 6P; Fig. S6K, S6L). Critically, the m⁶A mark itself is essential for this function. To further examine whether m⁶A modification is required for MALAT1 function in AML genome, we performed the gRNA-guided dcas13-ALKBH5 (dm6A) system [38] to specifically remove m⁶A modification in MALAT1 RNA (Fig. S6M). Notably, loss of m⁶A in MALAT1 significantly reduces the CTCF occupancy, R-loop structures, and chromatin interactions at the MYB and RUNX1 loci (Fig. 6M-O), phenocopying the effects of MALAT1 depletion. Furthermore, the oncogenic MLL-AF9 fusion increased m⁶A modification on MALAT1 and PVT1 (Fig. S6N), and m⁶A levels were higher in primary MLLr + AML cells than in normal CD34 + HSCs (Fig. 6Q). Depletion of METTL3 or treatment with MLLi distinctly reduces the expression and m⁶A modification of MALAT1 and PVT1 arcRNAs (Fig. 6R, S).

Collectively, these findings establish a novel model in which the METTL3-YTHDC1 axis recognizes m⁶A-modified arcRNAs and collaborates with the CTCF/Cohesin complex to stabilize R-loops, thereby regulating oncogenic TAD boundaries and maintaining the leukemic transcriptional program for MLLr + AML.

Targeting arcRNA-mediated R-loop structure alters AML genome organization

To investigate the role of chromatin arcRNA-mediated R-loops at promoters of key leukemia-associated genes, MYB and RUNX1, we employed a dCas9-RNaseH (dCas9-RH) to specifically target and disrupt R-loop structures at the MYB promoter (referred to as MYB^RH) or RUNX1 promoter (referred to as RUNX1^RH) in MLLr + MOLM13 cells (Fig. 7A; Fig. S7A). We also generated a dCas9-RNaseH(D210N) mutant, which retains DNA-binding ability but lacks nuclease activity, served as a critical control [94]. Targeting R-loops with the active RNase H (RH-WT) significantly reduces the expression of the key leukemia-related genes, including RUNX1, MYB, MYC, CTNNB1, and CDK6, while the dead mutant (RH-D210N) has minimal effect (Fig. S7B). Functionally, disruption of the R-loop at either the MYB promoter (MYB^RH) or RUNX1 promoter (RUNX1^RH) significantly inhibits cell proliferation compared to non-targeting sgRNA controls (Fig. 7B; Fig. S7C). Mechanistically, MYB^RH distinctly reduces the expression of MYB and its target genes, CDK6 and CCND1 (Fig. 7C). This is accompanied by a significant loss of R-loop formation, reduced CTCF and METTL3 binding, and impaired enhancer-promoter interactions at the MYB locus, as measured by ChIP-qPCR, DRIP-qPCR, and 3 C-qPCR (Fig. 7D-G). A similar effect was observed at RUNX1 locus, where RUNX1^RH significantly decreases the expression of RUNX1, CDK6, and CCND1 (Fig. S7D), reduces R-loops and METTL3 binding (Fig. S7E, 7 F), and disrupts long-range chromatin interactions as shown by 4 C-seq (Fig. S7G).

Fig. 7.

Targeting arcRNA-mediated R-loop at the MYB locus suppresses MLLr + AML proliferation. A. Schematic representation of dCas9-RNaseH (dCas-RH) targeting the MYB promoter. B. Proliferation curves of MYB^RH-targeted (sgMYB-RH-WT-#1 and sgMYB-RH-WT-#2) MOLM13 cells and controls (Ng-Ctrl, non-sgRNA; sgNT-Ctrl, sgRNA targeting a non-specific locus) MOLM13 cells. C. RT-qPCR analysis of the expression of MYB and its target genes in controls (Ng-Ctrl, sgNT-Ctrl), sgMYB-RH-WT-#1 and sgMYB-RH-WT-#2 MOLM13 cells. D. DRIP-qPCR analysis of R-loops at the MYB promoter in controls (Ng-Ctrl, sgNT-Ctrl), sgMYB-RH-WT, and RNase-treated (RH-WT) MOLM13 cells. E. CTCF ChIP-qPCR analysis showing CTCF occupancy at the MYB and RUNX1 loci in controls (Ng-Ctrl, sgNT-Ctrl) sgMYB-RH-WT and RNase-treated (RH-WT) MOLM13 cells. F. METTL3 ChIP-qPCR analysis showing METTL3 binding at the MYB locus in Ctrl (Ng-Ctrl) and MYB^RH (sgMYB-RH-WT) MOLM13 cells. G. 3 C-qPCR analysis of the MYB promoter-enhancer interactions in Ctrl (Ng-Ctrl) and MYB^RH (sgMYB-RH-WT) MOLM13 cells. Data are represented as the mean ± SD; two-tailed Student’s t-test; *P < 0.05, **P < 0.01, ***P < 0.001, n = 3 independent experiments

To further confirm which’s arcRNAs involved in CTCF dependent genome organization in specific regions at the TAD boundaries, we performed dCas9-APEX (AP) to target TAD boundary at RUNX1 locus (RUNX1^AP) in MLLr + MOLM13 cells, labeling proximity arcRNAs by biotinylation with APEX-seq technology [51, 52] (Fig. S7H). Our dCas9-APEX2 proximity labeling at the RUNX1 TAD boundary successfully identifies a set of leukemic-associated arcRNAs, including SP1, PTENP-AS, RUNXOR, MALAT1, NEAT1, PVT1, and MEG3 (Fig. S7I). The specific enrichment of two key arcRNAs, MALAT1 and PVT1, at this TAD boundary is quantitatively validated using APEX-qPCR, confirming their significant localization compared to non-targeting controls (Fig. S7J). GO enrichment analysis of these arcRNAs associated genes shows their involvement in chromatin remodeling, chromatin organization, chromatin looping, cell cycle regulation, DNA/RNA helicase, WNT signaling, and PI3K-AKT signaling (Fig. S7K).

Collectively, these findings demonstrate that arcRNA-mediated R-loops at MYB and RUNX1 loci are critical architectural components that modulate the chromatin interactions to regulate the expression of leukemic transcriptional program, which is required for AML cell proliferation.

Discussion

The METTL3-YTHDC1 axis is a transcriptional target and functional effector of MLL in AML

The mixed-lineage leukemia (MLL) gene, a master regulator of hematopoietic differentiation, is frequently rearranged in AML, driving oncogenic transcriptional programs through aberrant histone methylation and enhancer hijacking [6, 7]. Our analysis reveals that METTL3 and MLL exhibit parallel expression dynamics in both normal and malignant hematopoiesis, with high expression in primitive hematopoietic stem cells (HSCs) and MPPs (Fig. S2A). Similar to MLL, METTL3 is integral to maintaining stem cell identity—a role co-opted in AML to preserve LSC self-renewal (Fig. S2B). Mechanistically, we identified a novel regulatory axis in which METTL3 is transcriptionally activated by MLL, supported by MLL ChIP-seq data showing direct promoter binding (Fig. 3B), and the suppression of METTL3 expression and chromatin occupancy upon pharmacological inhibition of the Menin-MLL interaction in MLLr + AML cells (Fig. 3E and G; S3K, S3L). The selective upregulation of METTL3 or YTHDC1 in MLLr + AML, driven by leukemogenic MLL fusions (e.g. MLL-AF9), establishes the METTL3-YTHDC1 axis as a subtype-specific vulnerability (Fig. 3I-K). This finding explains why other AML subtypes, which depend on distinct oncogenic drivers (e.g., FLT3, NPM1 mutations), do not exhibit a comparable dependency, reinforcing its translational potential for MLLr + AML.

The functional importance of this axis is underscored by the impairment of primary LSC self-renewal and engraftment upon METTL3 depletion (Fig. 2B, D-F; Fig. S2L). Crucially, a robust therapeutic window exists, as targeting this axis disrupt LSC function while sparing normal hematopoiesis, an effect potentiated by synergy with conventional chemotherapeutics (Fig. 3C and F, S3G, S3H). Thus, these findings suggest that MLL-mediated activation of METTL3 sustains leukemic self-renewal by coupling transcriptional dysregulation with epitranscriptomic reprogramming, enabling AML cells to maintain stem-like properties.

METTL3 collaborates with architectural proteins to regulate the leukemic epigenome

While prior studies established that METTL3 binds chromatin transcriptional start sites (TSS) of active genes (e.g., CEBPZ) to regulate transcription programs [35, 95]. However, the mechanism by which METTL3 binds to AML genome remains unclear. In our study, we demonstrate that METTL3 co-localizes with CTCF and Cohesin at the promoters and enhancers across the AML genome to regulate the expression of leukemia-associated genes, including MYB, RUNX1, STAT5B, and CDK6 (Fig. 4C–E; Fig. S4C). This spatial co-localization suggests that METTL3 collaborates with architectural proteins (e.g., CTCF, Cohesin complex) to orchestrate chromatin looping and enhancer-promoter communication, thereby amplifying oncogenic transcription.

Recent studies have highlighted the role of CTCF in mammalian genome regulation through homodimerization and interactions with the cohesin complex [16, 96, 97], as well as MYC [17], NPM1C [18], and MAZ [19]. These CTCF-associated factors co-localize with CTCF, stabilizing CTCF-mediated chromatin contacts and enhancing the enhancer-promoter interactions. In our study, we show that METTL3 specifically interacts with CTCF and YTHDC1, thereby regulating cell proliferation and cell cycle progression in MLLr + AML cells (Figs. 1B-E and 2A-D). Consistent with prior reports in MOLM13 cells [33, 36], knockdown of METTL3 or YTHDC1 in MOLM13 cells suppresses proliferation (Fig. S2H, S2I) and reduces colony-forming capacity (Fig. 2B). However, our study now extends these observations by defining a novel mechanistic basis for these effects. We identify a previously uncharacterized role for the METTL3-YTHDC1-arcRNA axis in regulating R-loop homeostasis and 3D chromatin architecture in AML genome. A significant proportion of METTL3 peaks (39.01%) co-localizes with CTCF and R-loop signals (Fig. 4G), suggesting a functional interplay among these factors. Mechanistically, YTHDC1 loss disrupts CTCF-mediated chromatin topology at leukemia-associated gene loci, including RUNX1, MYB, CDK6, and STAT5B (Fig. 5C-D; Fig. S5C). We further demonstrate that CTCF-mediated chromatin TAD structures are required for the activation of METTL3-YTHDC1 axis (Fig. 5A-D). The R-loop-mediated stabilization of TADs at the MYB and RUNX1 loci is a critical driver of the leukemic transcription signature. Some studies showed that RUNX1 is a well-characterized tumor suppressor in non-MLL-r AML, such as FLT3-ITD, where loss-of-function mutations are common and associated with leukemogenesis [98, 99]. However, in MLLr + AML, MLL fusion proteins (e.g., MLL-AF9) directly upregulate RUNX1 expression in leukemogenesis [100–102], and RUNX1 is required for maintaining the leukemic transcriptional program—including activation of HOX genes and stemness factors (e.g., MEIS1) [103]. By sustaining that powerful enhancers (e.g., the + 24 kb enhancer of RUNX1) [104] remain in persistent physical proximity to their target promoters, these stable TADs sustain the high, dysregulated expression of these master oncogenes and their downstream targets, thereby locking cells into a proliferative, undifferentiated state. This sustained chromatin looping facilitates the constitutive activation of MYB and RUNX1, thereby directly promoting AML self-renewal and leukemic progression (Figs. 2A-F and 5J; Fig. S5J). Collectively, METTL3 and YTHDC1 function as CTCF co-factors in enhancer-promoter interactions and 3D genome organization, positioning the METTL3-YTHDC1 axis as a potential therapeutic target for MLLr + AML.

ArcRNA-mediated R-Loops as critical effectors of oncogenic chromatin architecture

As a crucial architectural protein, CTCF mediates genome topology and gene expression, relying on its RNA-interacting domains to facilitate long-range chromatin interactions [20, 21]. Recent studies showed that chromatin-associated RNAs play a critical role in regulating CTCF-dependent genome organization, thereby modulating hematopoiesis and leukemogenesis [105, 106]. Recent study reveals that APOLO lncRNA activates gene transcription by forming R-loops, which sequester the polycomb protein complex (PRC) from promoters in Arabidopsis [67]. Strikingly, HOTTIP lncRNA coordinates with CTCF binding sites (CBSs) to maintain TAD boundary integrity and topology in the AML genome by forming R-loops [11, 22]. Accumulating evidence highlights that the interaction mechanisms between chromatin arcRNAs and CTCF-dependent TADs are critical for transcriptional regulation during hematopoiesis and leukemogenesis [105, 106]. However, how arcRNAs influence CTCF-mediated TAD boundaries is poorly defined. In our study, we demonstrate that arcRNA-mediated R-loops are required for establishing CTCF-mediated TAD boundaries, maintaining TAD integrity, and driving leukemia-associated gene expression and cell proliferation (Fig. 6B-E; Fig. S6A-C). Depletion of R-loops significantly reduces METTL3 binding, CTCF binding and chromatin accessibility at the promoters and enhancers of leukemia-associated genes, including MYB and RUNX1 (Fig. 6B; Fig. S6B). Hi-C data further reveal that loss of R-loops alters TAD structures, decreasing 318 TADs and increasing 256 TADs in the AML genome (Fig. 6C). Notably, arcRNA-mediated R-loop depletion significantly disrupts TAD chromatin structures at RUNX1, MYB, and STAT5B loci, highlighting their role in oncogenic genome organization (Fig. 6E; Fig. S6C). The R-loop-mediated stabilization of TADs at MYB and RUNX1 is a critical driver of leukemic transcription. Notably, understanding the regulation of arcRNA-associated R-loops at CTCF boundaries is crucial to mitigate R-loop-mediated genomic instability. These findings suggest that arcRNA-mediated R-loops broadly influence 3D genome organization and transcriptional regulation.

An integrated model: m⁶A modification couples epitranscriptomic regulation to 3D genome organization

MLL (especially MLL-AF9 fusion) directly occupies the MYB/RUNX1 promoters to initiate basal transcription—this is the “upstream trigger” for oncogene expression, consistent with the previous report [85]. However, our data demonstrate that this initial binding event is necessary but not sufficient for the robust, high-level expression required to sustain the leukemic state. The METTL3-YTHDC1 axis functions as an essential amplifier and stabilizer of this initiated transcription. It does so by establishing a permissive 3D chromatin architecture that encompasses these loci, like MYB, RUNX1. Therefore, we propose that the MLL complex provides the transcriptional initiation signal, while the METTL3-YTHDC1-mediated m⁶A-RNA-chromatin axis provides the structural “scaffold” that locks these domains into an active, open configuration. This resolves the apparent contradiction and positions the METTL3-YTHDC1 axis not as an exclusive regulator, but as a critical co-factor that enables the full oncogenic potential of MLL fusion-driven transcription.

Our central mechanistic model posits that the METTL3-YTHDC1 axis serves as a molecular linker, bridging epitranscriptomic regulation (via m⁶A RNA modification) and 3D genome organization (via chromatin looping and TAD maintenance) in MLLr⁺ AML. This axis is critical for sustaining the leukemic transcriptional program by stabilizing oncogenic chromatin architectures.

At the core of this model is the METTL3-YTHDC1-mediated m⁶A modification of architectural RNAs (arcRNAs)—non-coding RNAs that scaffold chromatin structures—such as MALAT1 (Fig. 6L-P). This modification is not passive; instead, it actively enhances R-loop formation (RNA-DNA hybrid structures) at key genomic loci (Fig. 6M), which in turn reinforces the integrity of CTCF-dependent TADs (Fig. 6B-F; Fig. S6H). TADs are critical for maintaining proximity between enhancers and promoters of leukemogenic genes (e.g., MYB, RUNX1), ensuring their aberrant expression in MLLr⁺ AML. Two key mechanistic steps underpin this process: 1). YTHDC1 as a molecular bridge: YTHDC1 physically interacts with both METTL3 (the core m⁶A methyltransferase) and the chromatin organizers CTCF and Cohesin (Fig. S1G, S1H; Fig. 4O). This tripartite complex stabilizes R-loops at CTCF binding sites (Fig. 4D, E and M; Fig. S4C), creating a platform for sustained chromatin looping. 2). m⁶A-arcRNA reinforcement of chromatin architecture: METTL3-YTHDC1-mediated m⁶A modification of arcRNAs enhances R-loop stability, which directly strengthens CTCF and Cohesin occupancy at TAD boundaries (Fig. 6C-G, N and P; Fig. S6G, S6K). This reinforcement ensures that oncogenic TADs—critical for driving LSC self-renewal and blocking myeloid differentiation—remain intact even in the absence of differentiation signals.

Collectively, our model posits that the epitranscriptomic marking of RNA is a fundamental regulatory layer for 3D genome organization. In MLLr + AML, the METTL3-YTHDC1 axis hijacks this pathway by modifying arcRNAs to function as molecular scaffolds. These m⁶A-modified arcRNAs promote stable R-loop formation, which acts as a structural keystone to lock CTCF/cohesin complexes in place, thereby freezing the 3D genome in an oncogenic configuration. This work bridges the distinct fields of RNA modification and chromatin topology, revealing a previously unexplored mechanism of gene regulation and nominating the disruption of this epitranscriptome-architectome interface as a promising therapeutic strategy for this aggressive leukemia.

Conclusions

In summary, our study elucidates a coherent pathway from genetic lesion to chromatin alteration in MLLr + AML: the oncogenic MLL fusion transcriptionally activates METTL3, and the METTL3-YTHDC1 axis subsequently mediates m⁶A modification of arcRNAs to stabilize R-loops and maintain the activity and integrity of CTCF-dependent oncogenic TADs in the MLLr + AML genome. Collectively, our work bridges epitranscriptomics and 3D genome biology, offering a unified framework to understand how RNA modifications sculpt chromatin architecture in AML. These insights position the METTL3-YTHDC1-CTCF axis as a potential therapeutic target in MLLr + AML.

Abbreviations

AML

Acute myeloid leukemia

TAD

Topologically associating domain

CTCF

CCCTC-binding factor

KMT2A

Lysine methyltransferase 2A

MLL

Mixed-lineage leukemia

MLLr+

MLL-rearranged

CBSs

CTCF binding sites

arcRNAs

Architectural RNAs

HS/PC

Hematopoietic stem/progenitor cell

lncRNA

Long non-coding RNA

m⁶A

N6-methyladenosine

HSC

Hematopoietic stem cell

LSC

Leukemia stem cell

LC-MS

Liquid chromatography - tandem mass spectrometry

GO

Gene ontology

CoIP

Co-immunoprecipitation

TCGA

The Cancer Genome Atlas

pHSCs

Pre-leukemic HSCs

DEGs

Differentially expressed genes

GSEA

Gene Set Enrichment Analysis

GEO

Gene Expression Omnibus

ChIP

chromatin immunoprecipitation

TSS

Transcription start site

TF

Transcription factor

sgRNAs

Small guide RNAs

PDX

Patient-Derived Xenograft

BM

Bone marrow

PB

Peripheral blood

GEO

Gene Expression Omnibus

ChIRP

Chromatin Isolation by RNA Immunoprecipitation

RIP

RNA Immunoprecipitation

ATAC

Assay for Transposase-Accessible Chromatin

Ara-C

Cytarabine

DRIP

DNA: RNA immunoprecipitation

Hi-C

High-throughput chromosome conformation capture

gRNAs

Guide RNAs

IGV

Integrated Genomic Viewer

3C

Chromosome conformation capture

4C

Circular chromosome conformation capture

RIP

RNA Immunoprecipitation

ATAC-seq

Assay for Transposase-Accessible Chromatin using sequencing

METTL3

Methyltransferase-like 3

YTHDC1

YTH domain-containing protein 1

MYB

Myeloblastosis oncogene

MYC

Myelocytomatosis oncogene

RUNX1

Runt-Related Transcription Factor 1

NPM1C

C-terminal mutation of Nucleophosmin 1

MAZ

MYC-associated zinc finger protein

HOXB4

Homeobox B4

STAT1

Signal Transducer and Activator of Transcription 1

MCM

Minichromosome Maintenance

BCL2

B-cell lymphoma 2

METTL14

Methyltransferase-like protein 14

WTAP

Wilms Tumor 1 Associated Protein

DHX9

DExH-box helicase 9

PARP1

Poly (ADP-Ribose) Polymerase 1

DHX15

DEAH-box helicase 15

GAPDH

Glyceraldehyde-3-Phosphate Dehydrogenase

RAD21

Double-strand-break repair protein RAD21 homolog

SMC3

Structural Maintenance of Chromosomes 3

SMC1A

Structural Maintenance of Chromosomes 1A

YTHDF1

YTH domain-containing family protein 1

DHX16

DEAH-box helicase 16

DDX24

DEAD-box helicase 24

CTNNB1

Catenin Beta 1

HOXA9

Homeobox A9

MEIS1

Meis homeobox 1

MALAT1

Metastasis Associated Lung Adenocarcinoma Transcript 1

PVT1

Plasmacytoma Variant Translocation 1

APEX

Ascorbate peroxidase

H3K4me3

Histone H3 lysine 4 trimethylation

H3K27ac

Histone H3 lysine 27 acetylation

ETS1

ETS proto-oncogene 1

STAT5B

Signal transducer and activator of transcription 5B

GSK3B

Glycogen synthase kinase 3 beta

CDK6

Cyclin - dependent kinase 6

CDK1

Cyclin - dependent kinase 1

CCND1

Cyclin D1

IL-3

Interleukin 3

IL-6

Interleukin 6

EPO

Erythropoietin

TPO

Thrombopoietin

SCF

Stem cell factor

Authors’ contributions

R.F., X.Z., R.Z., C.D., H.Y., B.C., H.Q., Y.X., L.W., J.L., and H.L. designed and performed experiments. R.F., C.D., and A.S. established mammalian cell expression vectors. R.Z., C.D., H.Y., A.S. and H.L. performed bioinformatics and statistical analysis. X.Z., R. Zhou, W.Y., and H. Lou provided human patient samples, cell lines and reagents. H.L., X.Z., A.S., R.Z., and R.F. wrote the manuscript and revised the manuscript.

Funding

This study is supported by grants from the National Natural Science Foundation of China (H.L., Grant Numbers 82270193 and 82470156; W.Y., Grant Number 82370151; A. S., Grant Number 32401219), Zhejiang Provincial Natural Science Foundation of China (H.L., Grant Number YXD24H0801), as well as the National Key Research and Development Program of China (W.Y., Grant Numbers 2022YFC2502700 and 2022YFC2502701).

Data availability

All genomics datasets generated in this study can be accessed at GEO database (accession number GSE287825).

Declarations

Ethics approval and consent to participate

This study was conducted in compliance with the declaration of Helsinki and received informed consent from all human subjects. The primary AML patient samples were acquired with approval from the institutional review boards (IRBs) of the First Affiliated Hospital of Zhejiang University School of Medicine and the Guangzhou First People’s Hospital in accordance with the Institutional Animal Care and Use Committee (ICAUR) protocol.

Consent for publication

All authors agree to publication of the article.

Competing interests

The authors declare no competing interests.