Rbm5 sustains leukemia stem cells through a Myc-driven transcriptional circuitry

Abstract

Acute myeloid leukemia (AML) represents a type of malignant hematological disease that is usually caused by the dysregulated developmental program of leukemia stem cells (LSCs). Here, we report that an unappreciated RNA-binding protein, Rbm5, selectively promotes murine leukemogenesis, maintains LSC self-renewal in vivo, and is dispensable for normal hematopoiesis. Rbm5 is highly expressed in LSCs, and its deficiency results in specifically defective LSC function, along with inhibition of self-renewal gene expression and induction of myeloid differentiation. Multi-disciplinary mechanistic investigations further identified Myc as the major and direct transcriptional target of Rbm5 in primary leukemia cells. Moreover, RBM5 not only interacts with MYC but also maintains its protein levels, thereby sustaining the Myc downstream transcriptional network through its proper genome-wide occupancy. Forced expression of Myc sufficiently rescued the Rbm5-depleted LSC defects. Thus, our study demonstrates that Rbm5 regulates the AML LSC program through non-canonical transcriptional mechanisms, providing a strong rationale for targeting Rbm5 therapeutically.

In Brief. Zhang et al. illustrate the role of Rbm5 in sustaining the self-renewal program in leukemia stem. cells (LSCs) primarily through the Myc transcriptional network. Specifically, Rbm5 loss results in a significant decrease in Myc protein levels, thereby disrupting. the Myc downstream transcriptional network in LSCs. Notably, this effect is specific to LSCs, as. normal hematopoietic stem cells (HSCs) do not exhibit such changes upon Rbm5 loss.

Introduction

Acute myeloid leukemia (AML) is a genetically heterogeneous hematological malignancy driven by a variety of complex gene transcription and epigenetic mechanisms [1–8]. Despite recent advances in targeted therapies [9–14], leukemia stem cells (LSCs) remain a key driver of drug resistance and relapse in AML [15–18]. Therefore, there remains a high unmet need to identify more druggable targets for the clearance of LSCs while sparing normal hematopoiesis.

Multiple lines of evidence have demonstrated that mutations and abnormal expression of RNA-binding proteins (RBPs) can promote leukemia development, making them promising druggable targets, such as the well-characterized RBM39 [19–25]. Mechanistically, most RBPs control leukemogenesis by post-transcriptionally regulating RNA splicing, translation, and even non-canonical transcriptional events [26–29]. Although a few RBPs have recently been reported to be crucial for LSC maintenance [30–34], more specific RNA-binding proteins that target LSCs but not hematopoietic stem cells (HSCs) remain to be investigated. RBM5, a member of the RNA-binding motif (RBM) family of proteins, was originally identified as a putative tumor suppressor gene in various solid tumors and functions as a key modulator of apoptosis through the alternative splicing of multiple mRNAs [35]. In contrast, our recent study demonstrated that RBM5 is essential for the fitness of human AML cell lines by directly activating HOXA9, highlighting its context-specific role in blood malignancy [27]. However, the regulatory roles of RBM5 in an in vivo mouse disease model and the underlying mechanisms remain to be explored, especially its functions in LSCs and HSCs.

To mitigate this knowledge gap, we leveraged the power of a newly established conditional knockout mouse model to interrogate the detailed role of Rbm5 in blood malignancy and the normal hematopoietic system. Here, we found that RBM5 loss has a greater effect on LSCs than on normal HSCs. In a classic MLL-AF9 fusion gene-driven mouse AML model, Rbm5 deficiency significantly prolongs mouse survival, reduces the proportion of LSCs in the bone marrow, and inhibits the differentiation and clonogenic capacity of LSCs ex vivo. Furthermore, single-cell RNA sequencing analysis revealed that Rbm5 loss specifically decreased the number and proportion of LSCs and significantly downregulated the expression of the stemness gene Myc. Mechanistically, we demonstrate that RBM5 physically interacts with MYC and its known cofactors, thereby sustaining MYC-dependent activation of the self-renewal transcriptional program through an autoregulatory mechanism. Thus, our study suggests that Rbm5 plays a critical regulatory role in maintaining LSC stemness, with promise for novel therapeutic innovations.

Materials and methods

Mice

Rbm5^fl/fl knock-in mice were generated via gene targeting by GemPharmatech Co., Ltd (as described in Fig. 1A). Vav-iCre transgenic mice were purchased from The Jackson Laboratory. Rbm5-conditional-knock-out (CKO) mice were generated by crossing Rbm5^fl/fl mice with Vav-iCre mice. All the mice were bred on a pure C57BL/6 J genetic background. The genotypes of the mice used in this study have been confirmed by genomic PCR and sequencing prior to use. All genotyping primers were included in Supplementary Table 1. C57BL/6 J (CD45.2) (#N000013) and B6-CD45.1(#T054816) used in this study were purchased from GemPharmatech Co., Ltd., and were randomly assigned to experimental groups.

Fig. 1. Rbm5 promotes murine AML development and progression.

A Construction strategy of Rbm5^fl/fl mice: Loxp sites are located outside the 3rd and 4th exons of Rbm5. After crossing with Vav1-iCre mice, the sequence from the 3rd to the 4th exon of Rbm5 can be deleted. The MLL-AF9-driven murine leukemia model was utilized to isolate c-Kit⁺ hematopoietic stem/progenitor cells from Rbm5^fl/fl and Rbm5^Δ/Δ mice. These cells were then transduced with MLL-AF9 retrovirus to establish primary MLL-AF9 leukemia precursor cells following primary bone marrow transplantation (1° BMT). Subsequently, the primary MLL-AF9 leukemia cells were subjected to either a colony formation assay or secondary bone marrow transplantation (2° BMT). Created in BioRender. Xin, Y. (2026) https://BioRender.com/ny9uzov. B Rbm5 deletion in MLL-AF9-transformed c-Kit⁺ cells reduced colony formation (the left panel shows the first round of plating; the right panel shows the second round of plating). MLL-AF9-transformed c-Kit⁺ cells from Rbm5 ^fl/fl (n = 3) and Rbm5^Δ/Δ (n = 3) mice were used for the colony assay. C Deletion of Rbm5 delays leukemia progression in the MLL-AF9 model. Survival analysis was performed on primary transplants from mice with genotypes Rbm5^fl/fl (n = 9) and Rbm5^Δ/Δ (n = 9). D Rbm5^Δ/Δ leukemic mice have reduced disease burden. Spleen weights of moribund Rbm5^fl/fl and Rbm5^Δ/Δ mice with proper deletion are shown. Results are from Rbm5^fl/fl (n = 7) and Rbm5^Δ/Δ (n = 6) mice. E Deletion of Rbm5 delays leukemia progression in secondary transplant (cells were from bone marrow in Fig. 1C). Survival analysis was performed on secondary transplants from mice with genotypes Rbm5^fl/fl (n = 7) and Rbm5^Δ/Δ (n = 14). F Representative images of cell morphology for the primary bone marrow leukemia cells from the moribund Rbm5^fl/fl and Rbm5^Δ/Δ mice that were achieved by Wright-Giemsa staining after cytospin. Scale bar, 10 μm. G Rbm5 deletion significantly inhibits the multi-round colony-forming ability of leukemia cells derived from the bone marrow of MLL-AF9 leukemic mice. Results are from Rbm5^fl/fl (n = 3) and Rbm5^Δ/Δ (n = 3) mice. H Expression of myeloid markers Cd11b and Gr1 in Rbm5^fl/fl and Rbm5^Δ/Δ MLL-AF9 leukemia cells. MFI: mean fluorescence intensity (n = 3). I The percentage of cell apoptosis measured by the AnnexinV/PI double-positive populations in Rbm5^fl/fl and Rbm5^Δ/Δ MLL-AF9 leukemia cells. J. Representative flow cytometry plot and statistics showing LSC populations between Rbm5^Δ/Δ and Rbm5^fl/fl in the MLL-AF9 leukemia model (n = 6). K In vitro limiting dilution assay (LDA) using MLL-AF9-transduced c-Kit⁺ cells from Rbm5^fl/fl or Rbm5^Δ/Δ mice. L Colony rescue assay that is derived from the empty vector (EV)-transfected Rbm5^fl/fl MLL-AF9 leukemia cells, and Rbm5 overexpressing (Rbm5 OE)-transfected Rbm5^Δ/Δ MLL-AF9 leukemia cells. The Rbm5 protein level was confirmed in each group (left panel), and the colony number in each group after 10 days of growth in methylcellulose medium was quantified (right panel). Data are presented as the mean ± SEM. Statistical significance was determined using Student’s two-tailed unpaired t test (B, D, G, H, I,and J), one-way ANOVA with Tukey’s multiple comparison test (L, K) and log-rank test (C, E) (* P < 0.05; ** P < 0.01; **** P < 0.0001).

Leukemic transplants and BM (Bone marrow) competitive transplantation

For the primary transplant of transformed MLL-AF9 c-Kit⁺ cells in leukemia initiation experiments, approximately 50,000 leukemic cells and 500,000 bone marrow support cells from syngeneic mice were injected into 8-week-old C57BL/6 J female mice that had been lethally irradiated with 900 cGy. In secondary transplants, about 200,000 bone marrow leukemic cells were transplanted into C57BL/6 J mice that had been sub-lethally irradiated with 450 cGy. When mice showed signs of systemic illness, such as immobility, a huddled posture, the inability to eat, ruffled fur, or self-mutilation, the animals were killed immediately by CO₂ anesthesia or cervical dislocation.

For mouse MLL-AF9 cell transplantation, the Cas9-GFP-expressing MLL-AF9 cells were transduced with the Lenti-guide-puro-IRES-GFP plasmid. Transduced MLL-AF9 cells expressing GFP were adjusted to maintain a GFP-positive cell ratio of 30%. These cells were then injected intravenously into 8-week-old C57BL/6 J female mice (n = 7 per group) at a dose of 500,000 cells per mouse, following 450 cGy irradiation. To assess the leukemic burden of the transduced MLL-AF9 cells, bone marrow was harvested from sgRbm5 and sgNT mice 28 days post-transplantation, and GFP-positive leukemia cells were analyzed by flow cytometry.

For the competitive reconstitution analysis, a total number of 5 × 10⁵ donor BM cells from Rbm5^fl/fl or Rbm5^Δ/Δ donor (CD45.2⁺) were mixed with 5 × 10⁵ competitor BM cells (B6-CD45.1), and transplanted into lethally irradiated 8-week-old B6-CD45.1 female recipient mice by retroorbital injection. The chimerism in the peripheral blood was analyzed at 4, 8, 12, and 16 weeks after transplantation by flow cytometry for the congenic markers CD45.1 (Biolegend, #110728) and CD45.2 (Biolegend, #109830).

In vivo and in vitro rescue experiments

For functional rescue studies, MSCV-based retroviral plasmids (10 μg; either empty vector control MSCV-EV-P2A-Puro-PGK-GFP or constructs expressing Rbm5, or Myc with identical backbone) were complexed with 10 μL jetPRIME transfection reagent in the provided buffer and transfected into PLAT-E retroviral packaging cells. Viral supernatants were harvested 48 h post-transfection and used to transduce bulk leukemia cells derived from both Rbm5^fl/fl and Rbm5^Δ/Δ models. Following puromycin selection (1 μg/mL, 72 h), equal numbers of viable transduced cells (1 × 10⁴) were plated in MethoCult methylcellulose medium (STEMCELL Technologies, #M3434) for standard colony-forming unit (CFU) assays. Alternatively, limiting dilution analysis (LDA) was performed by plating serial dilutions of cells. Colonies were quantified after 5 days of culture under standard conditions (37 °C, 5% CO₂). For in vivo validation, 5 × 10⁵ transduced cells were injected intravenously into lethally irradiated recipient mice to assess the leukemogenic potential.

Single-cell RNA sequencing (scRNA-seq) and data analysis

Mouse scRNA-seq analysis was performed using the Chromium Single Cell 30 v2 Reagent Kit and Chromium Controller (10× Genomics) according to the manufacturer’s protocol. Individual libraries were sequenced on the Illumina HiSeq × Ten or Illumina NovaSeq 6000 system with 30,000-40,000 read pairs per cell (Annoroad, China). The scRNA-seq data from mouse AML cells were demultiplexed, aligned to the mouse genome (mm10), and UMI-collapsed using the CellRanger toolkit (v.3.0.2, 10× Genomics). Cells that expressed fewer than 300 genes or over 10% mitochondrial gene content were excluded. Genes expressed in less than three cells per sample were excluded. The Seurat R package v.3.0.2 was used for data normalization, generating uniform manifold approximation and projection plots, and performing differential expression analysis. Wilcoxon rank-sum tests were used to identify differentially expressed genes in each cluster compared with the others.

Software and statistical analysis

Prism software and R were used for data processing, statistical analysis, and result visualization (http://www.graphpad.com). The R language and environment for graphics (https://www.r-project.org) were used in this study for the bioinformatics analysis of the CRISPR screen and RNA-seq data. The R packages used for all analyses described in this manuscript were from the Bioconductor and CRAN. Data were analyzed with an unpaired Student’s t test (two-tailed), one-way ANOVA with Tukey’s multiple comparison test or two-way ANOVA with Sidak’s multiple comparison test accordingly. The overall survival of in different groups was compared using the Kaplan-Meier method with a log-rank test. Sample sizes were chosen according to previously performed power analysis. For the animal experiment, at least a sample size of 3 mice per group was calculated and used; see figure legends for specific sample sizes, noted as “n”. No blinding was done. Gene expression, CFU assay, and proliferation experiments were performed on at least three independent occasions. Chromatin immunoprecipitation was performed in duplicate (2 technical replicates), and immunoblotting experiments were repeated at least three times. No data was excluded from analysis. Statistical significance was set at P < 0.05 (*, P < 0.05; **, P < 0.01; ***, P < 0.001; **** P < 0.0001, and “n.s.” indicates no statistical significance).

Results

Rbm5 promotes murine AML development and progression

To investigate the role of Rbm5 in AML initiation and maintenance, we established a classic human-relevant in vivo model of AML pathogenesis. We utilized the MLL-AF9 retroviral transduction system to induce leukemogenesis in primary bone marrow cells from both wild-type and Rbm5 knockout mice (Fig. 1A; Supplementary Fig. S1A, B). To this end, we first performed a colony formation assay (CFA) using wild-type and Rbm5-deficient c-Kit⁺ cells transduced with the MLL-AF9 retrovirus. We observed a notable reduction in myeloid colony numbers in the Rbm5-deficient groups in primary plating, and a significantly more pronounced reduction in secondary plating (Fig. 1B). Consistently, we observed that MLL-AF9 transduction in wild-type c-Kit⁺ cells resulted in the upregulation of RBM5 expression levels, suggesting its functional requirement in myeloid leukemia (Supplementary Fig. S1C, D). Next, we transplanted the transformed wild-type and Rbm5-deficient c-Kit⁺ cells into half-lethal, irradiated recipient mice for an in vivo study of leukemogenesis. As a result, the host mice transplanted with Rbm5-deficient c-Kit⁺ cells showed a significantly delayed progression of leukemia (average latency, 72 versus 39 days) and a reduced disease burden in spleen compared to controls (Fig. 1C, D). Moreover, the second-round in vivo transplantation of leukemia cells collected from primary sick mice resulted in a progressive delay in leukemogenesis (average latency, 65 versus 36 days) in Rbm5-deficient cells compared to controls (Fig. 1E). Importantly, the Rbm5-deficient leukemia cells from the primary transplanted animals also exhibited a more mature morphology and a significant reduction in plating efficiency for colony formation and cell growth (Fig. 1F, G; Supplementary Fig. S1E, F). Correspondingly, Rbm5 loss led to a considerable increase in the frequency and median fluorescence intensity (MFI) of myeloid markers (CD11b and Gr1), and a substantial increase in apoptosis (Fig. 1H, I; Supplementary Fig. S1G). More strikingly, we validated the role of Rbm5 in the self-renewal of LSCs and found that the leukemic granulocyte-monocyte progenitor (L-GMP) cell population (defined as GFP⁺Lin^-Sca1^- IL-7R^-c-Kit⁺CD34⁺CD16/32⁺) was significantly lower in the Rbm5 KO MLL-AF9 leukemia group than in control WT mice (Fig. 1J). Moreover, the in vitro limiting dilution assay (LDA) showed that the loss of Rbm5 remarkably impaired the stemness of AML cells (Fig. 1K). These effects were specific to Rbm5 loss, as ectopic expression of Rbm5 cDNA fully rescued colony formation in Rbm5-deficient primary leukemia cells (Fig. 1L).To further explore the detailed role of Rbm5 in leukemia maintenance, we utilized the CRISPR/Cas9 approach with two individual Rbm5-specific guide RNAs (Rbm5-sg1 and Rbm5-sg2) to disrupt Rbm5 in primary wild-type MLL-AF9-driven mouse leukemia cells expressing Cas9 protein, followed by functional validations such as competitive proliferation assay (CPA) (Supplementary Fig. S1H). Notably, Rbm5 depletion significantly suppressed the growth of mouse MLL-AF9 leukemia cells, as demonstrated by a notable reduction in GFP⁺ cells compared to the control both in vitro and in vivo (Supplementary Fig. S1I–K). To rule out potential off-target effects of CRISPR/Cas9, ectopic expression of the Rbm5-sg1-resistant Rbm5 cDNA fully rescued the growth defect in Rbm5-depleted cells (Supplementary Fig. S1L, M). Together, these results indicate that Rbm5 promotes leukemogenesis and is critical for the maintenance of mouse primary leukemia-initiating/stem cells in AML.

Rbm5 is dispensable for normal hematopoiesis

As a suitable AML therapeutic target, the impact of gene function is expected to be selective in hematopoietic malignancy. To this end, we comprehensively analyzed the hematopoietic compartment in the bone marrow and peripheral blood from different genotype groups of littermates by crossing the Rbm5^fl/fl strain with the hematopoietic Vav1-Cre strain (Fig. 2A). In general, Rbm5 loss did not affect the expected Mendelian Ratio (Fig. 2B). Immunoblotting successfully confirmed the tissue-specific expression and deletion of Rbm5 in hematopoietic organs, such as the bone marrow and spleen, while remaining unchanged in other tissues and during most stages of organ development (Fig. 2C; Supplementary Fig. S2A, B). We next assessed different hematopoietic cell populations in detail from the bone marrow of both the wild-type and Rbm5 knockout littermates. In Rbm5 hematopoietic-deficient mice, there was no significant difference in the frequency and absolute number of phenotypic HSCs and different hematopoietic progenitors (Fig. 2D–G). The frequency and absolute number of myeloid, B cells, T cells, and early erythroid progenitors were also comparable between the wild-type and Rbm5 knockout mice in both bone marrow and peripheral blood (Fig. 2H; Supplementary Fig. S2C, D). Additionally, the blood cell count, including platelets, red blood cells, and lymphocytes, was normal in Rbm5-deficient mice (Supplementary Fig. S2E). To identify the in vitro functional output of hematopoietic stem progenitor cells (HSPCs), we performed colony assays and observed comparable numbers of myeloid progenitor colonies between wild-type and Rbm5 null c-Kit⁺ cells (Fig. 2I). To further evaluate whether Rbm5 depletion affects the repopulating capacity of HSPCs, we performed in vivo competitive repopulation assays. Total BM cells from the wild-type and Rbm5 knockout mice (CD45.2⁺) were mixed with competitor BM cells (CD45.1⁺) and transplanted into donor WT mice (CD45.1⁺). These chimeric mice were then monitored for 4 months. We found that the percentage of donor-derived total cells (CD45.2⁺) in PB was comparable between the two recipient groups (Fig. 2J). These data suggest that Rbm5 is dispensable for normal mouse hematopoiesis. Nevertheless, elucidating the role of RBM5 in stress hematopoiesis is essential for a comprehensive understanding of its physiological function, which remains to be investigated.

Fig. 2. Rbm5 is dispensable for normal hematopoiesis.

A Crossing Rbm5^fl/fl mice with Vav1-iCre tool mice yields four genotypes of offspring for the following functional assay. Created in BioRender. Xin, Y. (2026) https://BioRender.com/ny9uzov. B Hematopoietic system-specific Rbm5 conditional knockout mice can breed normally. The genotypic frequency of newborn pups (21 days post-birth) generally follows Mendel’s laws of inheritance (n = 89). Successful deletion and recombination of Rbm5 in the Rbm5^Δ/Δ bone marrow were shown in the right panel. PCR by genomic DNA from Rbm5^fl/fl, Rbm5^fl/Δ, and Rbm5^Δ/Δ mice bone marrow cells to validate Rbm5 exon 3–4 deletion (Representative images of n = 3 independent experiments). Primers against Loxp and Vav1-iCre were used for genotype and recombination, respectively. C Protein samples extracted from whole bone marrow (BM) cells, spleen cells, and non-primary hematopoietic organs of Rbm5^fl/fl, Rbm5 ^fl/Δ, and Rbm5^Δ/Δ genotype mice were subjected to Western blot analysis to detect RBM5 expression, with GAPDH used as an internal control. D FACS analysis and gating strategy of BM cells from representative primary mice for evaluating the ratio of stem and progenitor fraction. Hematopoietic Stem and Progenitor Cell (HSPC): Lin^-Sca1⁺Kit⁺; Long Time Hematopoietic Stem Cell (LT-HSC):Lin^-Sca1⁺Kit⁺CD150⁺CD48^-; Short Time Hematopoietic Stem Cell (ST-HSC): Lin^-Sca1⁺Kit⁺CD150⁺CD48⁺; Multipotent Progenitors (MPPs):Lin^-Sca1⁺Kit⁺CD150^-CD48^-. E Bar graph plots of numbers and fractions of HSPC, LT-HSC, ST-HSC, and MPPs in BM of primary 3-month-old Rbm5^fl/fl, and Vav1-Cre; Rbm5^fl/fl / (Rbm5^Δ/Δ). n = 6 independent samples, both males and females. F FACS analysis and gating strategy of BM cells from representative primary mice for evaluating the ratio of progenitor cells. G Bar graph plots of numbers and the fraction of MP, MEP, CMP, and GMP in BM of primary 3-month-old Rbm5^fl/fl, and Vav1-Cre; Rbm5^fl/fl / (Rbm5^Δ/Δ). n = 6 independent samples, both males and females. Myeloid Progenitors (MP): Lin^-Sca1^-Kit⁺; Megakaryocytic-Erythroid Progenitor (MEP): Lin^-Sca1^-Kit⁺CD34^-CD16/32^-; Common Myeloid Progenitors (CMP): Lin^-Sca1^-Kit⁺CD34⁺CD16/32^-; Granulocyte-monocyte progenitor (GMP): Lin^-Sca1^-Kit⁺CD34⁺CD16/32⁺. H The bar graph shows the proportions of various terminally mature cells in the bone marrow of Rbm5^fl/fl (n = 3) and Rbm5^Δ/Δ (n = 3) mice. I Rbm5 is dispensable for the colony-forming ability of normal stem cells. Colony assay was performed with c-Kit⁺ cells from Rbm5^fl/fl (n = 4) and Rbm5^Δ/Δ (n = 4) mice. J Experimental design for competitive bone marrow transplantation. Created in BioRender. Xin, Y. (2026) https://BioRender.com/ny9uzov. Peripheral blood was collected at 4-, 8-, 12-, and 16-week post-transplantation, and the contribution of Rbm5^fl/fl (n = 14) and Rbm5^Δ/Δ (n = 11) donor-derived cells was quantified by flow cytometry. The data was from two independent transplantation experiments. Each dot represents an independent sample. Data are presented as the mean ± SEM. Statistical significance was determined using Student’s two-tailed unpaired t test (E, G,and H) and two-way ANOVA with Sidak’s multiple comparison test (J) (* P < 0.05; “n.s.” indicates no statistical significance).

Rbm5 specifically sustains the activity of LSCs but not the normal HSCs

The functional assay indicates that Rbm5 sustains the maintenance of mouse primary leukemia stem/initiation cells but is dispensable for normal hematopoietic stem and progenitor cells (HSPCs). We sought to explore the underlying molecular mechanism to better understand its role. We hypothesized that RBM5 might control the MLL-fusion induced self-renewal gene signature in LSCs. To identify the gene expression profile in this heterogeneous cell population, we performed high-resolution single-cell RNA-sequencing (scRNA-seq) in wild-type (WT) and Rbm5 knockout (KO) MLL-AF9 primary leukemia cells from the same endpoint after primary transplantation, together with corresponding normal c-Kit⁺ HSPCs (Fig. 3A). We collected scRNA-seq data from a total of 51,517 cells, with more than 2000 genes per cell in each of the four groups (Fig. 3A). Next, the unbiased clustering analysis identified 15 cell populations with distinct gene expression profiles, including the specific CD34⁺ LSCs or HSPCs populations (Fig. 3B; Supplementary Fig. S3A). Notably, a decrease in LSCs was observed in the Rbm5 KO MLL-AF9 leukemia group compared with the control, accompanied by an increase in mature myeloid cell populations (Fig. 3C, D; and Supplementary Fig. S3B). In contrast, the frequency of normal HSPCs and mature cell populations was quite comparable between wild-type and Rbm5 KO c-Kit⁺ cells (Fig. 3D). This observation was further supported by the downregulation of stemness marker genes, such as Cd34, Hoxa9, Flt3, Myc, and Ikzf2, in the Rbm5 KO MLL-AF9 leukemia group compared to the Rbm5 KO c-Kit⁺ cells (Fig. 3E; Supplementary Fig. S3C). We next identified the differentially expressed genes (DEGs) in the Rbm5 KO LSCs, including 157 downregulated and 589 upregulated genes (Fig. 3F; Supplementary Table 2). In contrast, marginal genes showed a significant change in Rbm5 KO c-Kit⁺ cells relative to control (Supplementary Fig. S3D; Supplementary Table 2). Moreover, the enrichR analysis identified known oncogenic TFs, such as AF4 and ENL, among the downregulated genes (Supplementary Fig. S3E; Supplementary Table 3).

Fig. 3. Rbm5 specifically sustains the activity of LSCs but not the normal HSCs.

A Experimental design flowchart of single-cell transcriptome analysis by the 10× genomics platform for the primary MLL-AF9 leukemia cells and the normal c-Kit⁺ cells from Rbm5^fl/fl and Rbm5^Δ/Δ mice. Created in BioRender. Xin, Y. (2026) https://BioRender.com/ny9uzov. UMAP visualization of 51,517 cells from 4 independent samples shown in the left panel, demonstrating cell clustering. B Distribution of the 15 cell subpopulations in the UMAP plot. These 15 cell populations include: Proliferating (cycling)/stem cells; Mature neutrophils; Mast cells; T cells; Erythroid precursors; Plasma cells; Tissue-resident macrophages; neutrophil/granulocytic precursor; Activated neutrophils; Inflammatory neutrophils; CD34⁺ HSPCs or LSCs; Inflammatory monocytes/macrophages; Erythrocytes; Mature B cells; Pre-B cells. C UMAP plots showing the origin of cells from different samples. Red arrows indicate the LSC/HSPC cell populations, while black arrows indicate the mature myeloid cell populations. D Bar graphs displaying the proportions of Cd34⁺ LSCs or HSPCs and Proliferation (cycling)/stem cells in Rbm5^fl/fl and Rbm5^Δ/Δ mice within the MLL-AF9 leukemia model and primary c-Kit⁺ model. E UMAP plots showing the expression of Rbm5, Cd34, c-Myc, and Ikzf2 genes in the MLL-AF9 leukemia model and primary c-Kit⁺ cells, where color intensity represents expression levels, and each dot represents a single cell. F Volcano plot illustrating differentially expressed genes in LSCs populations between Rbm5^Δ/Δ and Rbm5^fl/fl in the MLL-AF9 leukemia model. G Expression levels of RBM5 in LSC and non-LSC populations from the GSE76008 cohort. H Correlation analysis between RBM5 expression levels and the overall survival of AML patients from the TARGET cohort. I Representative flow cytometry analysis and statistical analysis of hCD34 and hCD38 after knocking down RBM5 using two respective shRNAs (shRBM5#1 and shRBM5#2) in four AML patient samples, including three with KMT2A rearrangements (KMT2A-AF9, KMT2A-ENL, and KMT2A-AF17) and one with a KMT2A-PTD mutation. J Representative flow cytometry analysis of CD34 in which normal CD34⁺ cells were treated with lentiviral transduction of shRNA against a control gene (shNT) and shRNAs targeting RBM5. Data are presented as the mean ± SEM. Statistical significance was determined using Student’s two-tailed unpaired t test (G), one-way ANOVA with Tukey’s multiple comparison test (I, J) and log-rank test (H) (** P < 0.01; *** P < 0.001; **** P < 0.0001; “n.s.” indicates no statistical significance). LSC leukemia stem cell, HSPC hematopoietic stem and progenitor cell, HSC hematopoietic stem cell.

We extended our exploration to determine if RBM5 also sustains the maintenance of human primary leukemia stem cells. Indeed, RBM5 expression was higher in functional LSCs than in non-LSCs, as shown in a publicly available dataset (GSE76008) [36] (Fig. 3G). In line with its enrichment in LSCs, high RBM5 expression is associated with poor survival in pediatric AML patients (Fig. 3H). We further validated that suppression of RBM5 significantly reduced the frequency of CD34⁺CD38^- LSCs in four individual AML patients, without affecting the frequency of CD34⁺ cells in two normal donors (Fig. 3I, J, Supplementary Fig. S3F). More importantly, suppression of RBM5 also reduced the frequency of CD34⁺CD38^- LSCs and colony formation in patient-derived xenografts (PDXs) derived from primary cells with a KMT2A-r mutation (Supplementary Fig. S3G, H). Therefore, these data suggest a specific requirement for Rbm5 in sustaining the self-renewal program of LSCs but not in normal HSCs.

Rbm5 loss leads to the suppression of LSC self-renewal genes and activation of myeloid differentiation genes

Although scRNA-seq indicates LSC defects in Rbm5 knockout (KO) MLL-AF9, the number of gene transcripts is relatively lower due to technology limitations, as captured by 3’ UTR reads (10× genomics [37]). To gain genome-wide insights into how Rbm5 regulates the MLL-r induced self-renewal gene program, we performed a genome-wide transcriptome analysis by high-resolution bulk RNA sequencing in wild-type (WT) and Rbm5 knockout (KO) MLL-AF9 primary leukemia cells derived from bone marrow samples (Fig. 4A). As expected, the exons of Rbm5 flanked by two loxP cassettes, were efficiently deleted, as supported by their corresponding RNA-sequencing reads (Supplementary Fig. S4A). Principal component analysis (PCA) was used to compare the whole transcriptome across 14,043 transcripts. Consistently, the WT and KO groups, with three biological replicates, were separated by genotype, suggesting that loss of Rbm5 could result in a notable transcriptome change (Supplementary Fig. S4B). We then performed differential gene expression analysis between Rbm5 knockout (KO) and control MLL-AF9 leukemia cells. Using a stringent cutoff (FDR < 0.05 and Log₂(fold change) >1), 335 increased genes and 302 decreased genes were identified in Rbm5 KO cells relative to the control (Fig. 4A; Supplementary Table 4). Next, we performed Gene Set Enrichment Analysis (GSEA) in Rbm5 KO groups compared to the negative control. As a result, the leukemia stem cell signature genes and Hoxa9 and Meis1 targets were highly enriched in the downregulated genes in Rbm5 KO leukemia cells. In contrast, the myeloid differentiation pathways, such as myeloid cell development and the myeloid CEBPA network, were enriched among the upregulated genes (Fig. 4B; Supplementary Table 5). Consistently, the Kyoto Encyclopedia of Genes and Genome (KEGG) analysis revealed that Wnt and Rap1 signaling were enriched in the downregulated genes, while cytokine-cytokine receptor interaction and osteoclast differentiation were enriched in upregulated genes in Rbm5 KO cells (Fig. 4C; Supplementary Fig. S4C; Supplementary Table 5). Those downregulated genes included many well-known LSC self-renewal genes, and upregulated genes in myeloid differentiation were observed (Fig. 4D). Importantly, we confirmed that the expression levels of LSC signature genes, including Hoxa9, Flt3, Ikzf2, and Cd34, were significantly reduced. In contrast, the myeloid differentiation genes, including S100a8, S100a9, Fpr2, and C3, were significantly upregulated in bone marrow-derived Rbm5 KO primary leukemia cells (Fig. 4E, F; Supplementary Fig. S4D). The expression levels of Hoxa9, Flt3, Ikzf2, and Cd34 were consistently suppressed in spleen-derived Rbm5 KO primary leukemia cells (Supplementary Fig. S4E, F). Thus, these results suggest that loss of Rbm5 not only suppresses the transcriptional activity of self-renewal genes but also activates the myeloid differentiation genes in myeloid leukemia.

Fig. 4. Rbm5 loss leads to the suppression of LSC self-renewal genes and activation of myeloid differentiation genes.

A Flowchart of RNA-sequencing (RNA-seq) analysis performed on MLL-AF9 leukemia cells derived from the bone marrow of 1st generation BMT in Rbm5^fl/fl and Rbm5^Δ/Δ mice (Upper). Created in BioRender. Xin, Y. (2026) https://BioRender.com/ny9uzov. MA plot showing differentially expressed genes between Rbm5^Δ/Δ and Rbm5^fl/fl in MLL-AF9 leukemia cells derived from the bone marrow of 1st generation BMT mice (Down). B Top enriched Gene Set Enrichment Analysis (GSEA) plots for downregulated and upregulated genes in primary MLL-AF9 leukemia cells upon Rbm5 deletion. C KEGG pathway analysis showing that genes that were differentially expressed upon RBM5 knockout, including downregulated genes that are significantly enriched in stem cell-related signaling sets, and upregulated genes in cytokine-cytokine interaction. D Heatmap showing the LSC signature genes that were downregulated and the myeloid CEBPA network genes were upregulated in Rbm5-depleted primary leukemia cells relative to the control. E Real-time qPCR analysis of the representative downregulated mRNA levels of Hoxa9, Flt3, Ikzf2, and Cd34 in primary leukemia cells derived from the bone marrow in the MLL-AF9 model. F Real-time qPCR analysis of the representative upregulated mRNA levels of S100a8, S100a9, Fpr2, and C3 in primary leukemia cells derived from the bone marrow in the MLL-AF9 model. Each dot represents an independent sample that were quantified via the average value from three technical replicates. Data are presented as the mean ± SEM. Statistical significance was determined using Student’s two-tailed unpaired t test (E, F) (*P < 0.05; ** P < 0.01).

RBM5 controls the MYC transcriptional activity via physical interaction, maintenance of its protein, and proper genomic occupancy

Given the significant enrichment of MYC-related downstream targets from the GSEA analysis (Supplementary Table 5) and the well-established role of MYC in leukemogenesis and the maintenance of LSCs [38–42], we proposed that MYC might serve as a novel and major direct target of Rbm5 in leukemia cells. Indeed, several gene signatures relevant to the hallmark of MYC direct targets were significantly enriched among the downregulated genes in Rbm5 knockout cells (Fig. 5A, B). We then sought to validate the expression levels of Rbm5 and Myc in MLL-AF9 primary leukemia cells. Although the endogenous Myc mRNA was mildly changed, the Myc protein was robustly suppressed in Rbm5 knockout MLL-AF9 leukemia cells from both bone marrow and spleen (Fig. 5C, D). To further distinguish the direct regulation of the RBM5-MYC axis from the stress feedback loop, we used the auxin-induced degron (AID) system to acutely target RBM5. We hypothesized that direct regulation of MYC gene transcription would manifest acutely, while regulation of MYC protein levels would not depend on transcriptional activity. To test this, we used previously generated KMT2A-r leukemia SEM cells with bi-allelic endogenous RBM5 N-terminus AID knock-in and stable expression of OsTIR1(F74G). These cells were transduced with lentiviral vectors either constitutively expressing MYC-HaloTag or without, followed by detection of both endogenous and exogenous MYC levels within 24 h of induced RBM5 degradation (Fig. 5E). Notably, after 24 h of treatment with the small molecule 5-Ph-IAA, immunoblotting analysis showed a robust decrease in both endogenous MYC protein and exogenous MYC-HaloTag fusion protein, concurrent with the reduction of the AID-RBM5 fusion protein (Fig. 5F, G). In contrast, MYC mRNA levels remained unchanged within 24 h of 5-Ph-IAA treatment (Fig. 5F, G). Consistently, CRISPR-mediated RBM5 knockout in multiple human leukemia cell lines resulted in a significant decrease in both endogenous MYC mRNA and protein levels, with a more pronounced effect on the protein level (Fig. 5H; Supplementary Fig. S5A). To further confirm whether this effect is specific to leukemia, we examined Myc levels in both wild-type (WT) and Rbm5 knockout (KO) normal c-Kit⁺ HSPCs. Notably, both mRNA and protein levels of Myc expression remained unchanged in Rbm5 KO c-Kit⁺ cells compared to wild-type controls (Supplementary Fig. S5B). Collectively, our data suggest that RBM5 directly controls MYC protein levels as a critical mechanism, in addition to the previously reported RBM5-HOXA9 regulatory axis [27].

Fig. 5. RBM5 controls the MYC transcriptional activity via physical interaction and maintenance of its protein, and proper genomic occupancy.

A GSEA showing that genes reduced in primary MLL-AF9 leukemia cells upon RBM5 deletion are significantly enriched in Myc targets. B Heatmap displaying the expression of Myc target genes in Rbm5^fl/fl and Rbm5^Δ/Δ MLL-AF9 cells. C Real-time -qPCR and Western blot analysis of Myc mRNA and protein levels in leukemia cells derived from the bone marrow of the MLL-AF9 mouse model upon RBM5 deletion. Each dot represents an independent sample. D Real-time -qPCR and Western blot analysis of Myc mRNA and protein levels in leukemia cells derived from the spleen of the MLL-AF9 mouse model upon RBM5 deletion. Each dot represents an independent sample. E Experimental design flowchart for endogenous and exogenous MYC detection after acute RBM5 degradation in the auxin-induced degradation (AID) system. In MLL-rearrangements (MLL-r) SEM leukemia cell clones with homozygous AID knock-in at the RBM5 N-terminus, the constitutive expression of OsTIR1(F74G) enables targeted RBM5 degradation. Upon treatment with 5-Ph-IAA, OsTIR1(F74G) recruits the SCF (Skp1-Cullin-F-box) E3 ubiquitin ligase complex to polyubiquitinate and degrade AID-tagged RBM5 (left panel). The mRNA and protein levels of MYC were further detected in RBM5^AID/AID SEM cells with or without MYC-HaloTag overexpression, following 1, 6, and 24 h of 5-Ph-IAA treatment (right panel). F The endogenous MYC transcript levels were measured by Real-time-qPCR, whereas corresponding protein levels of RBM5, MYC, MAX and GAPDH were determined by Western blot. G The exogenous MYC-HaloTag was analyzed by Real-time-qPCR; the protein levels of RBM5, MYC-HaloTag, endogenous MYC, MAX, and GAPDH were assessed by Western blot. H Real-time qPCR detection and Western blot analysis of MYC mRNA and protein levels in OCIAML2 cells upon RBM5 disruption. I Co-immunoprecipitation using MYC^Tag in OCIAML2-RBM5-MYC^Tag overexpression cells, followed by Western blot detection of proteins interacting with RBM5. J Co-immunoprecipitation using MYC antibody in OCIAML2 cells, followed by Western blot detection of RBM5, MYC, and ACTIN protein levels. K Co-immunoprecipitation using HaloTag antibody beads in RBM5-AID SEM cells that overexpress the HaloTag version of MYC. Immunoprecipitation followed by Western blot detection of RBM5 and MYC (using an mCherry antibody) protein levels was performed before and after RBM5 protein degradation. The protein HDAC2 was used as a positive control, known to interact with MYC. L Genomic heatmaps of the reproducible decreased MYC-HaloTag peaks (6,149 peaks) from two replicates centered on the summit of MYC-HaloTag bound regions in RBM5-AID SEM cells untreated (-5-Ph-IAA) or treated with 1 μM 5-Ph-IAA ( + 5-Ph-IAA) for 24 h. M Genome browser tracks of the MYC loci. Displayed are the Chromatin Immunoprecipitation sequencing (ChIP-seq) profiles of MYC-HaloTag in RBM5-AID SEM cells untreated (-5-Ph-IAA) or treated with 1 μM 5-Ph-IAA ( + 5-Ph-IAA) for 24 h. Each group has two biological replicates. Data are presented as the mean ± SEM. Statistical significance was determined using Student’s two-tailed unpaired t test (C, D and H) and one-way ANOVA with Tukey’s multiple comparison test (F and G) (* P < 0.05; **** P < 0.0001; “n.s.” indicates no statistical significance).

Given that MYC itself and its direct targets were inhibited after RBM5 loss, we hypothesized whether RBM5 could serve as a transcription cofactor for MYC in the leukemia context. To comprehensively integrate the RBM5-related protein network, we first performed immunoprecipitation using tag-fused exogenous RBM5 in the human leukemia cell line OCIAML2, followed by mass spectrometry analysis. Notably, we identified 658 RBM5-interacting proteins relative to an IgG control (Supplementary Fig. S5C; Supplementary Table 6). Consistent with its canonical role in RNA splicing, the majority of proteins identified in the interactome were enriched for RNA splicing function, as determined by protein network and GO enrichment analysis (Supplementary Fig. S5D). Notably, several key transcription factors were identified, including MYC and its known cofactor WDR5 (Supplementary Fig. S5C). The interaction between RBM5 and MYC was further verified in both human leukemia cells and HEK293T cells (Fig. 5I, J; Supplementary Fig. S5E). Such interaction was robustly and specifically disrupted after acute RBM5 degradation in previously generated KMT2A-r SEM leukemia cells (Fig. 5K; Supplementary Fig. S5F). We next asked whether loss of RBM5 alters MYC genome-wide occupancy, either globally or gene-selectively. Because ChIP for endogenous MYC is technically challenging, we instead performed MYC-HaloTag ChIP-seq in RBM5-AID SEM cells that stably express MYC-HaloTag fusion proteins, with or without 5-Ph-IAA treatment for 24 h. Principal-component analysis (PCA) of the resulting ChIP-seq profiles confirmed that untreated and 5-Ph-IAA treatment groups were separated from each other (Supplementary Fig. S6A), indicating that MYC-HaloTag occupancy is dynamically remodeled upon RBM5 loss. Notably, differential-binding analysis revealed 6,149 peaks with significantly reduced MYC-HaloTag signal after RBM5 depletion (Fig. 5L). Specifically, the MYC promoter itself, together with established MYC target loci such as NOP2 and NPM1, exhibited markedly diminished MYC-HaloTag binding occupancy, whereas a subset of other known targets remained unaffected (Fig. 5M; Supplementary Fig. S6B, C). Collectively, these data indicate that RBM5 is essential for stable MYC recruitment to a specific subset of genomic loci, including the MYC locus itself. Thus, our data suggest a particular requirement for RBM5 in MYC regulation in LSCs.

Myc overexpression can fully rescue the defects in Rbm5-deficient LSCs

Given the reduced Myc expression in primary transplanted RBM5-deficient leukemia cells, we hypothesized that Rbm5 may control LSCs through Myc. To test this hypothesis, we overexpressed Myc and the control vector in Rbm5-null primary leukemia cells and validated whether their reintroduction could restore the defects using the colony formation assay and the in vivo transplantation assay (Fig. 6A). Notably, we observed that Myc overexpression resulted in a robust increase in colony numbers and total derived cell numbers compared to the empty vectors (Fig. 6B–D). Importantly, Myc overexpression also restored stemness in Rbm5-deficient LSCs, as determined by in vitro limiting dilution assays (LDAs) (Fig. 6E). Inspired by these observations, Rbm5-depleted MLL-AF9 cells with or without Myc overexpression were transplanted into syngeneic host mice. Subsequently, we observed a significant rescue of the leukemic phenotype, as evidenced by reduced overall survival and increased leukemia burden in peripheral blood (Fig. 6F–H). More importantly, Myc overexpression also restored the leukemic granulocyte-monocyte progenitor (L-GMP) cell population from bone marrow in the Rbm5 KO MLL-AF9 leukemia group (Fig. 6I). Thus, our data suggest that Myc could fully restore the defects of Rbm5-deficient LSCs in vitro and in vivo.

Fig. 6. Myc overexpression can fully rescue the defects in Rbm5-deficient LSCs in vitro and in vivo.

A The workflow to restore Myc expression in primary bone marrow-derived leukemia cells from the Rbm5^Δ/Δ transplanted cohort. The leukemia cells from both Rbm5^fl/fl and Rbm5^Δ/Δ mice transduced with an empty vector (EV) were used as negative controls. Primary leukemia cells were first expanded as colony formation units (CFUs) and then transplanted into recipients to induce AML. Created in BioRender. Xin, Y. (2026) https://BioRender.com/ny9uzov. B Western blotting analysis of Rbm5 and Myc expression in primary bone marrow-derived leukemia cells from the three different groups. C, D Colony rescue assay that is derived from Myc and empty vector-transfected Rbm5^Δ/Δ MLL-AF9 leukemia cells relative to the empty vector-transfected Rbm5^fl/fl MLL-AF9 leukemia cells. The representative colony image is shown in (C), and the colony number and total cell number in each group after 10 days of growth in methylcellulose medium were quantified in (D). E In vitro limiting dilution assay (LDA) rescue assay using the Myc and empty vector-transfected Rbm5^Δ/Δ MLL-AF9 leukemia cells relative to the empty vector-transfected Rbm5^fl/fl MLL-AF9 leukemia cells. F Kaplan-Meier survival curves of recipients were plotted for the Rbm5^fl/fl + EV group (n = 7), the Rbm5^Δ/Δ+ EV group (n = 7), and the Rbm5^Δ/Δ + MYC group (n = 6). G Leukemia burden as indicated by the percentage of GFP⁺ cells in peripheral blood after the injection on day 14, day 28, and day 35. H Quantification of GFP⁺ cells on day 35 and corresponding representative peripheral blood images from each group, obtained by Wright-Giemsa staining after cytospin. Scale bar, 10 μm. I Flow cytometry analysis showing LSC populations using the Myc and empty vector-transfected Rbm5^Δ/Δ MLL-AF9 leukemia cells relative to the empty vector-transfected Rbm5^fl/fl MLL-AF9 leukemia cells. J Model illustrating the role of Rbm5 in sustaining the self-renewal program in leukemia stem cells (LSCs) primarily through the Myc transcriptional network. Specifically, Rbm5 loss results in a significant decrease in Myc protein levels, thereby disrupting the Myc downstream transcriptional network in LSCs. Notably, this effect is specific to LSCs, as normal hematopoietic stem cells (HSCs) do not exhibit such changes upon Rbm5 loss. Created in BioRender. Xin, Y. (2026) https://BioRender.com/ny9uzov. Each dot represents an independent sample. Data are presented as the mean ± SEM. Statistical significance was determined using one-way ANOVA with Tukey’s multiple comparison test (D, E, Hand I), two-way ANOVA with Sidak’s multiple comparison test (G) and log-rank test (F) (* P < 0.05; *** P < 0.001; **** P < 0.0001).

Discussion

Here, using genetic mouse models, we showed that Rbm5 is indispensable for maintaining the LSC program both in vivo and in vitro, but not the normal steady-state hematopoiesis. Mechanistically, we found that Rbm5 controls the transcriptional activity of several key LSC signature genes, including the novel target Myc. Thus, our study demonstrates a differential requirement for Rbm5 in maintaining LSCs in myeloid leukemia, but not in HSCs during normal hematopoiesis, providing a potential therapeutic target for selectively targeting LSCs in AML.

Our study establishes the RBM5–MYC regulatory axis as a novel pathway in myeloid leukemia (Fig. 6J). Moreover, RBM5 not only physically interacted with MYC but also substantially affected its protein steady-state level relative to the transcriptional level. This could be explained by several possible mechanisms. First, the interaction between RBM5 and MYC may antagonize MYC’s interaction with known E3 ubiquitin ligases or co-repressors, thereby protecting it from degradation. Our RBM5 interactome dataset identifies MYC and its known co-activator WDR5, but lacks the well-known E3 ligase FBXW7, which targets the MYC protein [43, 44]. Notably, WDR5 is a known interactor of MYC that could facilitate target recognition and promote tumorigenesis by MYC [45, 46]. Therefore, RBM5 may exert a similar effect by recruiting WDR5, a hypothesis that requires further validation. Second, MYC is a well-established master regulator of ribosome biogenesis and protein synthesis [47]. The reduction in MYC protein level may also result from lower translational efficiency following RBM5 loss. Nonetheless, further studies were required to fully understand the regulatory mechanism underlying the interaction between RBM5 and MYC.

The function of RBM5 in normal hematopoiesis is poorly characterized, although it has been primarily studied in spermatogenesis and neuronal disease [48, 49]. We successfully generated a hematopoietic Rbm5 conditional knockout mouse model and provided a comprehensive analysis of different hematopoietic compartments, suggesting that loss of Rbm5 does not affect normal hematopoiesis in the steady state. This finding was further supported by single-cell RNA-seq analysis, which revealed normal HSC frequency and marginal transcriptome changes in Rbm5 knockout c-Kit⁺ cells. However, we also noticed that the Rbm5 heterozygous mice exhibited significantly higher blood cell counts than wild‑type controls at 17 weeks. This effect may reflect compensatory feedback triggered by the loss of one Rbm5 allele, such as upregulation of alternative genes, though this phenomenon and the underlying mechanism remain to be investigated. The differential requirement of Rbm5 in LSC and HSC is highly correlated with their relative expression levels, as we found that Rbm5 is expressed at much higher levels in MLL-AF9-transformed cells. Nevertheless, we could not completely rule out the potential role of Rbm5 in stress hematopoiesis (e.g., regeneration after myelosuppression, inflammation, or aging), which requires further characterization in other in vivo mouse models.

In our study, Rbm5 is highly expressed in LSCs and plays a critical role in their maintenance. Interestingly, RBM5-AS1 activity, encoded by the RBM5 locus, is critical for the functional activation of colon cancer stem-like cells [50]. Moreover, the RBM5 gene was among the 17 common metastasis signature genes identified across multiple solid tumor types [51]. These results, together with our findings, suggest that RBM5 may be a key player in maintaining the stemness of various cancer stem cell populations. Further studies profiling RBM5’s direct targets in LSC would help distinguish direct from indirect targets in myeloid leukemia. More intriguingly, we observed a significant upregulation of immune-responsive genes in Rbm5-deficient leukemia cells (Supplementary Table 5). The phenomenon likely depends on MYC transcriptional activity, as MYC has been shown to promote immune suppression in both solid tumors and blood cancers [52, 53]. Additionally, a recent study reported that leukemia with mutations in RNA splicing factors, such as SF3B1, can induce mis-splicing-derived neoantigens [54]. Nevertheless, it remains unknown yet how exactly the intrinsic RBM5-MYC oncogenic signaling activates extrinsic crosstalk with tumor-infiltrating immune cells, including CD8⁺ T cells, NK cells, and tumor-associated macrophages. This observation suggests that Rbm5 loss may synergize with immune therapies, such as PD-1/PD-L1 blockade, a possibility warranting further investigation.

In summary, we found that RBM5 sustains the LSCs via the novel target MYC. The identification of dysregulated RBPs in hematological malignancies has led to the rapid development of therapeutic strategies targeting specific RBPs. Overall, we propose that targeting RBM5 for the eradication of LSCs could be a promising new therapeutic target in leukemia.

Author contributions

PX, M-LZ, and C-LL planned the experimental design, analyzed the data, and wrote the manuscript; M-LZ, YX, and YX-OY performed most of the phenotypic validation of Rbm5 mice under the guidance of PX.; M-LZ performed the mouse MLL-AF9 and human leukemia cell growth function validation and molecular validation assay, including RNA-seq, scRNA-seq, ATAC-seq, and ChIP-seq with the help of SF, YX, J-XZ; SF established and validated the MYC expression in the RBM5-AID2 model; and HL helped to provide the primary human samples under the guidance of JC and D-PW; QZ, and YX performed bioinformatics analyses by discussing with PX, B-SX, J-FS, and C-LL; PX supervised the study; and all authors discussed the results and contributed to preparing the manuscript.

Funding

PX discloses support for the research of this work from the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2025ZD0545700), the National Key R&D Program of China (2022YFC2502700), Jiangsu Province National Science and Technology grant (BK20243008), the High-Level Personnel Project of Jiangsu Province (JSSCTD202353), the Interdisciplinary Basic Frontier Innovation Program of Suzhou Medical College of Soochow University (YXY2302018), the Cyrus Tang Foundation (CTJC25002), the Pediatric Hematology & Oncology Key Laboratory of Higher Education Institutions in Jiangsu Province, the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD), and the Collaborative Innovation Center of Hematology. M-LZ discloses support for the research of this work from the Jiangsu Funding Program for Excellent Postdoctoral Talent (2025ZB720), Jiangsu Province National Science and Technology grant (BK20250829), the China Postdoctoral Science Foundation (2025M772841), and the Cyrus Tang Foundation (CTQN25002). C-LL discloses support for the research of this work from the American Cancer Society Scholar Grant (RSG DMC-135487), the V Foundation Scholar Grant (V2021-010), and the American Lebanese Syrian Associated Charities (ALSAC).

Data availability

Source data are provided with this paper. RNA-seq data and ChIP-seq data were deposited in GEO under the accession numbers GSE303605. All other raw data are available from the corresponding author upon reasonable request. Code repositories collected at https://doi.org/10.6084/m9.figshare.c.6186670.

Competing interests

CL is currently an editorial board member at Genome Biology. The other authors declare that they have no other competing interests.

Ethics approval and consent to participate

Human sample acquisition was approved by the Hematological Biobank, Jiangsu Biobank of Clinical Resources. The informed consent forms were approved by the Ethics Committee of Soochow University and The First Affiliated Hospital of Soochow University (ID: 2017169-3, Suzhou, China) in accordance with the Declaration of Helsinki. All mouse procedure protocols utilized in this study were in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) of Soochow University.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41375-026-03064-4.