RBM25 is required to restrain inflammation via ACLY RNA splicing-dependent metabolism rewiring

Abstract

Spliceosome dysfunction and aberrant RNA splicing underline unresolved inflammation and immunopathogenesis. Here, we revealed the misregulation of mRNA splicing via the spliceosome in the pathogenesis of rheumatoid arthritis (RA). Among them, decreased expression of RNA binding motif protein 25 (RBM25) was identified as a major pathogenic factor in RA patients and experimental arthritis mice through increased proinflammatory mediator production and increased hyperinflammation in macrophages. Multiomics analyses of macrophages from RBM25-deficient mice revealed that the transcriptional enhancement of proinflammatory genes (including Il1b, Il6, and Cxcl10) was coupled with histone 3 lysine 9 acetylation (H3K9ac) and H3K27ac modifications as well as hypoxia inducible factor-1α (HIF-1α) activity. Furthermore, RBM25 directly bound to and mediated the 14^th exon skipping of ATP citrate lyase (Acly) pre-mRNA, resulting in two distinct Acly isoforms, Acly Long (Acly L) and Acly Short (Acly S). In proinflammatory macrophages, Acly L was subjected to protein lactylation on lysine 918/995, whereas Acly S did not, which influenced its affinity for metabolic substrates and subsequent metabolic activity. RBM25 deficiency overwhelmingly increased the expression of the Acly S isoform, enhancing glycolysis and acetyl-CoA production for epigenetic remodeling, macrophage overactivation and tissue inflammatory injury. Finally, macrophage-specific deletion of RBM25 led to inflammaging, including spontaneous arthritis in various joints of mice and inflammation in multiple organs, which could be relieved by pharmacological inhibition of Acly. Overall, targeting the RBM25-Acly splicing axis represents a potential strategy for modulating macrophage responses in autoimmune arthritis and aging-associated inflammation.

Introduction

Macrophages are the major type of innate immune cells that adopt diverse activation states for tissue homeostasis and host defense, but their persistent activation leads to unresolved and chronic inflammation and even multiple inflammatory disorders, such as rheumatoid arthritis (RA) [1]. The enrichment of sublining macrophages in the synovium is the prominent hallmark of active RA, contributing to severe synovial joint dysfunctions, involving inflammatory injury, bone destruction and joint disability [2]. Recently, single-cell RNA-seq (scRNA-seq) and mass cytometry identified the predominant subpopulation of IL1B⁺ monocytes/macrophages in the RA synovium, which are similar to Toll-like receptor (TLR)-activated interleukin-1β (IL-1β)-producing inflammatory cells, as the key inflammatory cell population in the pathogenesis of RA [3]. IL-1β, which is produced mainly by resident and infiltrating macrophages, is a well-accepted harmful cytokine involved in both the initiation and progression of RA, and blocking the biological effects of IL-1β protects bone and cartilage from inflammatory injury [4, 5]. Although the effective role of therapeutic strategies targeting IL-1β has been confirmed in experimental and clinical settings, its endogenous regulatory networks in the host remain largely unknown.

Analyses of the genomic landscape in several inflammatory and autoimmune diseases have revealed several somatic genetic lesions, including frequent mutations in RNA-binding proteins (RBPs) and their associated RNAs [6]. Sequence-specific RBPs bind to pre-mRNAs to modulate alternative splicing (AS). Increasing evidence has demonstrated that genetic mutations or abnormal expression of RBPs that serve as splicing factors promote aberrant splicing, which is the key pathogenic mechanism during the initiation and progression of inflammatory diseases. For instance, diabetes downregulates the expression of the spliceosome regulator serine/arginine-rich splicing factor 7 (SRSF7), which induces a proinflammatory phenotype in the proximal tubule through alternative splicing [7]. In an inflammatory bowel disease (IBD) clinical study, several splicing factors (including HNRPAB, DUSP11, HNRPH3, SF3B14, SFPQ, SLU7, and SFR2IP) were shown to participate in disease progression and might serve as diagnostic markers [8]. In this way, understanding the pathogenic mutations of the spliceosome and splicing factors has considerable therapeutic and pharmacologic value in immune disorders and pathological injuries and needs to be further investigated.

Metabolic reprogramming of inflammatory macrophages is indispensable for their effector diversity and inflammatory disease onset. A comprehensive metabolic map of LPS-activated macrophages revealed the upregulation of glycolytic genes and the downregulation of mitochondrial genes, which were correlated with the expression profiles of altered metabolites [9]. LPS induces the expression and tetramer of pyruvate kinase M2 (PKM2), which attenuates a proinflammatory macrophage phenotype through the inhibition of hypoxia-inducible factor-1α (HIF-1α) [10]. PKM2 and its isoform PKM1 are generated by mutually exclusive PKM pre-mRNA, which is strictly generated by heterogeneous nuclear ribonucleoprotein (hnRNP) protein-mediated alternative splicing [11, 12]. However, whether alternative splicing of metabolic enzymes with immunoregulatory functions occurs remains largely obscure. Therefore, targeting alternative splicing events and upstream RNA splicing regulators has great therapeutic value for treating inflammatory diseases.

RNA binding motif protein 25 (RBM25), termed Snu71 in budding yeast, was identified as a U1 small nuclear ribonucleoprotein that functions in early spliceosome formation [13]. Aberrant expression or activity of RBM25 increases the susceptibility of the host to tumorigenesis, cardiovascular diseases, and Alzheimer’s disease [14–16]. However, there are no reports on the role of RBM25 in the immune system. Through screening RNA splicing regulators in human and mouse arthritis samples, we identified the low expression of RBM25 in macrophages as a master pathogenic factor of RA. Myeloid RBM25 strictly controls the metabolic and epigenetic phenotypes of inflammatory macrophages through alternative RNA splicing. Comprehensive analysis of transcriptome and epigenome data revealed ATP citrate lyase (Acly) as the target transcript of RBM25, whose alternative splicing generates an isoform called Acly S that lacks exon 14 and has increased metabolic activity, contributing to the rewiring of macrophage metabolism for HIF-1α-dependent glycolysis and histone acetylation. In summary, we propose the checkpoint role of the splicing factor RBM25 as well as RBM25-mediated alternative splicing in the metabolic and epigenetic reprogramming of macrophages, which might be a therapeutic target for inflammatory arthritis disease.

Results

Decreased expression of spliceosomal RBM25 in inflammatory and arthritis macrophages

To explore the molecular mechanism underlying RA pathogenesis, we analyzed the gene expression profiles of synovial tissues from healthy controls and RA patients [17] (Fig. S1A). Interestingly, gene set enrichment analysis (GSEA) revealed that several RNA AS pathways were dysregulated in the RA groups (Fig. S1B). In addition, Gene Ontology (GO) functional enrichment analysis of transcriptome data from control and arthritic mice [18] validated the inhibition of several mRNA process pathways, especially those involved in mRNA splicing via the spliceosome (Fig. S1C). In this way, we further analyzed the expression of RNA splicing regulators in both human and mouse arthritis samples to identify the common pathogenic elements in inflamed joints (Fig. 1A, B). Among them, a key component of the U1 spliceosome that mediates pre-mRNA AS events, RBM25, attracted our attention (Fig. 1C). RBM25 expression was significantly suppressed in both arthritis patients and mice (Fig. S1D, E). Intriguingly, RBM25 expression in synovial tissues was negatively correlated with the disease activity score of 28 joints with C-reactive protein (DAS28-CRP), a well-known indicator of RA disease severity (Fig. 1D). To further verify the data from the public database, we collected primary peripheral blood mononuclear cells (PBMCs) from healthy controls and RA patients during different disease periods (stable RA and active RA) and detected a decrease in RBM25 expression in both RA groups and a further decrease in the active RA groups (Fig. 1E). There was also a negative correlation between the expression of RBM25 and that of IL1B, the main pathogenic cytokine involved in RA disease progression (Fig. 1F). Immunohistochemistry (IHC) confirmed the decreased expression of RBM25 in synovial tissues of RA patients (Fig. 1G). Next, we used scRNA-seq to understand the distribution of RBM25 in different cell types in synovial tissues during arthritis [19] (Fig. S1F, G). Although Rbm25 is widely expressed in both hematopoietic and nonhematopoietic cells, it is relatively highly expressed in macrophages and fibroblasts (Fig. S1H). In contrast, other RNA-binding proteins also presented cell type-specific expression patterns. For example, Rbm38 was restricted to monocytes but not macrophages, whereas Ddx24 was highly expressed in adaptive immune cells (Fig. S1H).

Fig. 1. Decreased expression of spliceosomal RBM25 in inflammatory macrophages promotes arthritis pathogenesis.

Volcano plot of the gene expression data for RNA splicing in synovial tissues from RA patients and healthy people from the GEO dataset GSE55457 (A) and from experimental arthritis and control mice from the GEO dataset GSE22971 (B). C Venn diagram representing the common downregulated genes from A and B. D Correlation analysis of RBM25 mRNA levels in synovial tissues from RA patients with the DAS28-CRP score. E Q‒PCR analysis of RBM25 mRNA expression in primary PBMCs from healthy (n = 10), stable RA (n = 16) and active RA (n = 12) patients. F Correlation analysis of RBM25 with IL1B mRNA levels in PBMCs from RA patients (n = 28) in E. G IHC staining of RBM25 in synovial tissues from RA patients and disease controls. Scale bar: 200 μm. Clinical scores (H) and changes in paw thickness (I) of WT and RBM25 cKO CIA model mice after the 2nd immunization (n = 8 per group). H&E, Safranin O, CD68 and Ly6G staining (J) and histological scores (K) of knee joints from WT and RBM25 cKO mice with or without CIA (n = 5 per group). Scale bar: 500 μm. Micro-CT images (L) and bone mineral density and bone volume (M) of the front paws of the mice described in J (n = 5 per group). N ELISA analysis of the levels of the proinflammatory cytokines IL-1β, IL-6, IL-17A and TNF-α in the sera of H mice (n = 8 per group). O Clinical scores of WT, RBM25 cKO and RBM25 cKO+Il1r1 double-KO CIA model mice after the 2nd immunization (n = 6 per group). H&E and safranin O staining (P) and histological scores (Q) of knee joints from mice in the O group (n = 5 per group). Scale bar: 500 μm. *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001. Simple linear regression (D, F), two-way ANOVA (H, I and O), one-way ANOVA (E), and unpaired two-tailed Student’s t test (K, M, N and Q)

Therefore, we speculated that the proinflammatory environment in arthritic tissues results in decreased RBM25 expression, which might have immunomodulatory effects on the pathogenesis of autoimmunity and other immune disorders. We observed changes in the expression of RBM25 in bone marrow-derived macrophages (BMDMs) in response to various inflammatory stimuli, including lipopolysaccharide (LPS), Pam3CSK4, high mobility group box1 protein (HMGB1), poly(I:C) and CpG ODN. Both the mRNA and protein expression of RBM25 decreased after treatment with either stimulus (Fig. S2A, B). In addition, the decrease in RBM25 expression was time dependent in BMDMs treated with LPS or Pam3CSK4 (Fig. S2C, D). To further explore the mechanism underlying the decreased expression of RBM25 in macrophages stimulated with LPS, we analyzed chromatin accessibility, histone modification and RNA Pol II activity at the Rbm25 gene locus in BMDMs [20–22] and reported that LPS treatment led to a reduction in chromatin accessibility and histone H3 lysine 36 trimethylation (H3K36me3) as well as removal of elongating RNA polymerase II (RNA Pol II S2P) (Fig. S2E). GRO-seq [21] revealed that LPS treatment inhibited the synthesis of nascent Rbm25 RNA in macrophages (Fig. S2E). These results indicate that RBM25 expression is inhibited by inflammatory stimuli owing to transcriptional inhibition, especially impaired transcription elongation, providing a potential explanation for the decreased expression of RBM25 in PBMCs or lesion tissues under inflammatory circumstances in patients with autoimmune arthritis.

Macrophage RBM25 is a negative regulator of experimental arthritis pathogenesis

To investigate the role of RBM25 deficiency in the progression of arthritis, we generated Rbm25^fl/fl mice and then constructed macrophage-specific RBM25 knockout mice (Rbm25^fl/flLyz2-cre⁺, hereafter referred to as cKO mice) by crossing with Lyz2-cre mice (Fig. S3A, B). The knockout efficiency of RBM25 was validated at both the mRNA and protein levels in primary BMDMs from cKO mice compared with those from littermate wild-type controls (Rbm25^fl/fl, hereafter referred to as WT mice) (Fig. S3C, D). RBM25 deficiency in macrophages did not influence immune cell development in the spleen or thymus (Fig. S3E–J). RBM25 also had no effect on the development of peritoneal macrophages (PMs) or the in vitro M-CSF-induced differentiation of BMDMs (Fig. S3K, L). Taken together, these findings indicate that RBM25 cKO mice display normal immune cell differentiation under stable conditions.

A collagen-induced arthritis (CIA) model, which is a widely acknowledged animal inflammation model with pathological characteristics similar to those of human RA, was used to assess the biological function of RBM25 in vivo [23]. Although C57BL/6 mice are an experimental mouse strain with low CIA susceptibility [24], RBM25 cKO mice (C57BL/6 background) develop more severe inflammatory arthritis, with higher clinical scores, greater paw thickness and deterioration of joint swelling of the front and hind paws (Fig. 1H, I and Fig. S4A, B). Histological analyses of knee joints from RBM25 cKO mice revealed aggravated joint inflammation with increased synovial hyperplasia and infiltration of inflammatory cells, mainly macrophages and granulocytes (Fig. 1J, K). Safranin O staining also revealed more cartilage damage due to RBM25 deficiency (Fig. 1J). Three-dimensional microcomputed tomography (micro-CT) imaging revealed decreased bone density and bone volume in the finger phalanges and wrist joints of the RBM25 cKO mice, indicating pathogenic arthritis-associated bone erosion and destruction (Fig. 1L, M).

Next, we investigated the pathogenic changes in the immune system caused by RBM25 deficiency in CIA model. Compared with WT mice, RBM25 cKO mice presented more severe splenomegaly and lymphadenopathy, with increased populations of a variety of inflammatory immune cells (Fig. S4C–F). The percentages of splenic T cells with a memory phenotype were markedly increased in the RBM25 cKO mice (Fig. S4G, H). The levels of inflammatory cytokines, including interleukin-1β (IL-1β), IL-6, tumor necrosis factor-α (TNF-α), IL-17a, IFN-γ, CXCL10 and RANKL, were increased in the sera of RBM25 cKO mice (Figs. 1N and S4I). We detected a decrease in regulatory T cells in both the spleen and popliteal lymph nodes (pLNs) (Fig. S4J, K), indicating the important role of RBM25 in regulating the inflammatory cytokine network in inflammatory joint disease.

Upon characterization of RBM25 function in myeloid macrophages, we next validated these findings at the single-cell level in both human RA patients and mouse arthritis model. We analyzed the scRNA-seq profile of PBMCs from RA patients [25] and found that RBM25^hi cells barely overlapped with their highly inflammatory counterparts and displayed minimal inflammatory features, as quantified by the inflammatory score and exemplified by the expression of representative proarthritic genes (Fig. S5A-C). A similar phenomenon was also found in Rbm25^hi macrophages from mouse arthritis tissue [19], which reprensented the inhibited inflammatory features (Fig. S5D–F). These single-cell data indicate that RBM25 marks a population of hypoinflammatory monocytes/macrophages in arthritis settings and that the decreased expression of RBM25 in RA patients might be an important pathogenic factor for chronic and unresolved inflammation in joint tissues.

Considering that RBM25 was negatively correlated with the expression of inflammatory mediators, especially the pro-arthritis cytokine IL-1β, in both human and mouse arthritis settings, we hypothesized that excessive IL-1β production and activity in RBM25 cKO mice resulted in worsened experimental arthritis pathology. Double-knockout mice deficient in Rbm25 and Il1r1 (interleukin 1 receptor type 1) were generated by crossing Rbm25^fl/fl Lyz2-cre⁺ mice with Il1r1-KO mice. Deletion of Il1r1 in Rbm25 cKO mice significantly ameliorated arthritis progression, as shown by reduced arthritis clinical scores (Fig. 1O). The knees of the double knockout mice presented little inflammatory cell infiltration, attenuated synovial hyperplasia and bone damage (Fig. 1P, Q). These data demonstrate that hyperinflammation with excessive IL-1β production in RBM25-deficient mice is detrimental to arthritis pathogenesis.

Macrophage-specific deficiency of RBM25 enhances IL-1β production in vitro and in vivo

Transcriptome profiling via RNA sequencing (RNA-seq) of BMDMs from RBM25 WT and cKO mice was performed to further investigate the regulatory role of RBM25 in macrophage effector function. RBM25 deficiency led to global alterations in gene expression, and the number of differentially expressed genes (DEGs) increased with LPS stimulation (Fig. 2A). GO enrichment analyses revealed that DEGs were most enriched in immune-associated pathways, including immune system processes, inflammatory responses, granulocyte chemotaxis, etc. (Fig. 2B). Intriguingly, as the stimulation time increased, the number of DEGs enriched in immune system processes increased, and the mRNA levels of many genes were regulated by RBM25 deficiency under both untreated and LPS-treated conditions (Fig. 2C). Among the genes whose expression was most significantly increased were cytokines (Il1b, Il6, Il12a and Il23a) and chemokines (Cxcl10, Tnfsf9 and Ccl3), as well as lectins and integrins, including sialic acid binding Ig-like lectin 1 (Siglec1), integrin, and alpha 4 (Itga4) (Fig. 2D), while the genes whose expression was most downregulated were immunosuppressive molecules, such as arginase 1 (Arg1), Cd300e and ectonucleotide pyrophosphatase/phosphodiesterase 2 (Enpp2) (Fig. 2D). Next, we classified these DEGs into six clusters on the basis of their different gene expression patterns in response to RBM25 deficiency or the LPS-triggered macrophage phenotype (Fig. S6A). The results revealed that RBM25 deficiency promoted the expression of cluster genes induced by LPS, promoted immune activation, and inhibited the expression of genes that mediate immunosuppressive effects (Fig. S6B). The RBM25-influenced DEGs were involved mainly in cytokine‒cytokine receptor interactions that are responsible for multiple inflammatory and autoimmune diseases (Fig. S6C). Q-PCR and ELISA confirmed the transcriptome data showing that RBM25 specifically inhibited the production of LPS-induced proinflammatory mediators, including Il1b, Il6 and other pro-arthritic cytokines, without directly affecting TNF-α transcription (Fig. 2E, F and Fig. S6D). The induction of these proinflammatory genes by Poly(I:C) or Pam3SCK4 was also enhanced in RBM25 cKO BMDMs (Fig. S6E, F). A similar phenomenon was also observed in RBM25-deficient alveolar macrophages (AMs) treated with LPS (Fig. 2G), which are tissue-resident macrophages responsible for host defense and lung inflammation.

Fig. 2. Macrophage-specific deficiency of RBM25 enhances IL-1β production in vitro and in vivo.

Number of DEGs (A), GO analysis of DEGs (B), Venn diagram of overlapping DEGs in GO: 0002367 (C) and heatmap of upregulated genes (D) in WT and RBM25 cKO BMDMs stimulated with LPS for the indicated times. The log2FC values were calculated as Log2 FPKM + 1 (cKO/WT). Q‒PCR analysis of Il1b, Il6 and Tnf mRNA levels (E) and ELISA analysis of pro-IL-1β protein levels in cell lysates and IL-6 and TNF-α levels in media (F) from WT and cKO BMDMs stimulated with LPS for the indicated times (n = 3 biological replicates). ND, not detected. G Q‒PCR analysis of Il1b, Il6 and Tnf mRNA levels in primary WT and RBM25 cKO alveolar macrophages stimulated with LPS for the indicated times (n = 3 biological replicates). ELISA analysis of IL-1β, IL-6 and TNF-α levels in sera (H) and H&E staining of lung tissues (I) from WT and RBM25 cKO mice 8 h after i.p. injection of LPS (n = 8 for the LPS groups, n = 3 for the PBS groups). Scale bar: 1 mm; 500 μm. ELISA analysis of IL-1β, IL-6 and TNF-α protein levels (J), FACS analysis of CD11b⁺Gr-1⁺ cells (K) and absolute numbers of white blood cells (WBCs) and granulocytes (L) in peritoneal lavage fluid from WT and RBM25 cKO mice 8 h after i.p. injection with MSU (n = 8 per group). **P < 0.01; ***P < 0.001. NS, not significant. Unpaired two-tailed Student’s t test (E–H, J, L)

NLRP3 inflammasome activation is indispensable for IL-1β production and release [26]. Therefore, we determined the mRNA expression of several components of the NLRP3 inflammasome and found that RBM25 selectively suppressed Nlrp3 and Il1b expression at the priming step (Fig. S7A). ELISA assays revealed that IL-1β release in culture medium, but not that of TNF-α, was increased in LPS-primed RBM25-deficient BMDMs upon stimulation with ATP, nigericin or monosodium urate (MSU) (Fig. S7B, C). Immunoblotting confirmed the inhibitory role of RBM25 in the inducible expression of NLRP3 and pro-IL-1β (Fig. S7D). RBM25 also markedly promoted NLRP3 inflammasome activation, including the cleavage of caspase-1 and pro-IL-1β, as well as ASC oligomerization (Fig. S7E, F), which was largely dependent on inducible NLRP3 protein expression and activity. We next utilized another two in vivo acute inflammation models to verify the immunosuppressive function of RBM25, both of which have been proven to be closely related to macrophage function and proinflammatory cytokine secretion, especially NLRP3-dependent IL-1β production [27, 28]. In the LPS-induced sepsis model, the serum levels of IL-1β and IL-6 were significantly increased in the RBM25 cKO mice, whereas the production of TNF-α was hardly affected (Fig. 2H), with more severe pathological damage and more inflammatory cell infiltration in the lung tissues of the RBM25 cKO mice than in those of their WT littermates (Fig. 2I). Furthermore, intraperitoneal (i.p.) injection of MSU in RBM25-cKO mice led to increased levels of IL-1β and IL-6 in the peritoneal cavity fluid (Fig. 2J). The numbers of recruited white blood cells (WBCs) and granulocytes were also greatly elevated in the peritoneal cavity (Fig. 2K, L). These in vitro and in vivo data indicate that macrophage-specific ablation of RBM25 promotes inflammatory macrophage function and pathogenic IL-1β production.

RBM25 controls metabolic rewiring during proinflammatory macrophage activation

To understand how RBM25 specifically controls proinflammatory macrophage activation, we examined the canonical signaling pathway and transcription factor activation, which are indispensable for proinflammatory gene induction. However, RBM25 did not directly influence canonical TLR-triggered NF-κB or MAPK signal activation (Fig. S8A–C). RBM25 had no effect on either NF-κB or AP-1 transcriptional activity, as shown by dual-luciferase reporter gene assays (Fig. S8D, E). These data indicate that RBM25 inhibits the activation of inflammatory macrophages without enhancing TLR signaling cascades. A recent large-scale RBP analysis revealed that the transcription factor activity of RBM25 was essential for Yinyang1 (Yy1)-mediated chromatin binding, DNA looping, and transcription, which broadened the biological function of RBM25 in gene expression [29]. However, we did not find an interaction between RBM25 and Yy1 in macrophages with or without LPS stimulation (Fig. S8F). Although a previous study reported the critical role of Yy1 in the positive regulation of IL-6 production in rheumatoid arthritis [30], RBM25 deficiency had no influence on the recruitment of Yy1 to the Il6 gene locus (Fig. S8G). Silencing Yy1 in macrophages did not abrogate the immunostimulatory effects mediated by RBM25 deficiency (Fig. S8H, I). These results indicate that RBM25 controls inflammatory macrophage activation in a Yy-1-independent manner.

Recent work revealed the importance of metabolic remodeling in the switch from naïve macrophages to proinflammatory phenotypes [31]. In particular, the metabolism of classically activated macrophages is most notably characterized by increased dependence on glycolysis and altered mitochondrial respiration [32]. Therefore, metabolic profiling of WT and RBM25 cKO macrophages upon LPS treatment revealed alterations in central carbon metabolism, and the top 20 differentially expressed metabolites are shown (Fig. 3A). Deletion of RBM25 led to an accumulation of isocitrate and succinate but a decrease in pyruvate, which was coupled with metabolic adaptation from the disrupted TCA cycle to aerobic glycolysis in proinflammatory macrophages (Fig. 3B–D). In addition, the increase in lactate, which might be the product of glycolysis during macrophage inflammation, indicated an increased dependence on aerobic glycolysis in RBM25-deficient macrophages to enhance macrophage-mediated inflammatory responses (Fig. 3C). Next, the extracellular acidification rate (ECAR) and oxygen consumption rate (OCR) of WT and RBM25-cKO BMDMs were measured to determine the metabolic levels of glycolysis and oxidative phosphorylation (OXPHS), respectively. RBM25 deficiency resulted in significant increases in the basal and maximal ECAR under both resting and inflammatory activation conditions (Fig. 3E). Compared with WT macrophages, RBM25 cKO macrophages adopted increased amounts of glycolysis to meet the energy demand for proinflammatory effector functions (Fig. 3F). In contrast, the basal and maximal OCR levels were greatly impaired by RBM25 deficiency (Fig. 3G). Interestingly, we found that, compared with WT cells, RBM25 deficiency promoted the phosphorylation of AKT and mTOR, indicating that the activation of AKT‒mTOR signaling controls the metabolic phenotype of inflammatory macrophages (Fig. 3H). We then performed transmission electron microscopy (TEM) to examine the mitochondrial morphology of WT and cKO macrophages. RBM25 deficiency led to increased mitochondrial damage, including irregular mitochondrial swelling with fractured and fuzzy cristae (Fig. 3I). Furthermore, pharmacological inhibition of glycolysis by 2-deoxyglucose (2-DG) abrogated the increase in inflammatory stimulus-induced Il1b and Il6 transcription mediated by RBM25 deficiency (Fig. 3J). Moreover, we treated RBM25 WT and cKO BMDMs with dimethyl malonate (DMM), an inhibitor of succinate oxidation, to investigate the effect of succinate accumulation on RBM25 function. DMM treatment abrogated the promoting effect of RBM25 deficiency on HIF-1α and IL-1β production (Fig. 3K, L). These data reveal the indispensable role of RBM25 in controlling glycolysis rewiring and succinate accumulation for inflammatory macrophage metabolic adaptation.

Fig. 3. RBM25 controls metabolic rewiring during proinflammatory macrophage activation.

Heatmap representing the expression profile of the top 20 glucose metabolism intermediates (A), relative metabolite changes in glucose metabolism pathways (B) and concentrations of the indicated metabolites that were upregulated (C) or downregulated (D) in WT and RBM25 cKO BMDMs stimulated with LPS. ECAR analysis (E), ratio of the basal ECAR to the basal OCR (F) and OCR analysis (G) of WT and cKO BMDMs stimulated with or without LPS for 6 h. H Immunoblot analysis of the indicated proteins in WT and cKO BMDMs stimulated with LPS for the indicated times. I TEM images of mitochondria in resting or inflammatory (treated with LPS) WT and cKO BMDMs. J Q‒PCR analysis of Il1b and Il6 mRNA levels in WT and cKO BMDMs treated with LPS or LPS+2-DG (n = 3 biological replicates). Immunoblot analysis of HIF-1α and pro-IL-1β protein levels (K) and Q-PCR analysis of Il1b mRNA levels (L) in WT and cKO BMDMs treated with LPS or LPS+DMM. *P < 0.05; **P < 0.01; ****P < 0.0001. Unpaired two-tailed Student’s t test (C, D, G); one-way ANOVA (J, L)

RBM25-mediated transcription suppression depends on the glycolysis‒epigenetics axis

In addition to providing energy for macrophage effector function, a variety of intermediate metabolites of glycolysis coordinate with gene expression by modulating specific histone modifications [33]. Following the establishment of RBM25 in controlling glycolysis metabolism and specific gene expression, we next sought to explore the epigenetic mechanisms underlying proinflammatory gene transcription. Cleavage Under Targets and Tagmentation (CUT&Tag) assays were performed in inflammatory WT and cKO macrophages to analyze global histone 3 (H3) acetylation on lysines K9 and K27, which are associated with chromatin accessibility and transcription. We observed increased H3K9ac and H3K27ac in RBM25 cKO macrophages compared with WT macrophages (Figs. 4A and S9A). RBM25 deficiency altered histone acetylation in transcriptionally active regions with recognition motifs for important innate immunity-associated transcription factors, including interferon regulatory factor (IRF), PU.1, AP-1 and NF-κB, most of which were enriched in regions with elevated H3K9ac and H3K27ac in RBM25 cKO macrophages (Figs. 4B and S9B). To define the regulatory elements at different gene loci in LPS-stimulated macrophages, we aligned our H3K9ac and H3K27ac CUT&Tag sequencing data with published ATAC-seq datasets and detected significantly increased histone acetylation around the chromatin-accessible regions of the Il1b, Il6, Siglec1 and Cxcl10 genes (Fig. 4C). However, the histone acetylation levels of several housekeeping genes, such as Actb, Pgk1, Hprt and B2m, were not obviously different between the two groups (Fig. S9C). We further confirmed the increase in H3 acetylation at the Il1b, Il6 and Cxcl10 promoter regions via chromatin immunoprecipitation (ChIP)-qPCR assays with antibodies against acetylated H3K9 (Fig. 4D) and H3K27 (Fig. 4E) residues. These results indicate that specific enhancement of active histone acetylation greatly contributes to the increased gene expression of these proinflammatory mediators in RBM25 cKO macrophages.

Fig. 4. RBM25-mediated transcription suppression depends on the glycolysis-epigenetics axis.

Transcription start site (TSS)-centered heatmaps (upper) and profile (down) of H3K9ac and H3K27ac (A), information on the top transcription factors enriched in UP peaks ranked by P value (B) and IGV analysis of H3K9ac and H3K27ac signals at the indicated gene loci (C) via CUT&Tag analysis of WT and cKO BMDMs stimulated with LPS. ChIP‒qPCR analysis of H3K9ac (D) and H3K27ac (E) modifications at the gene loci of Il1b, Il6 and Cxcl10 in WT and cKO BMDMs stimulated with or without LPS (n = 3 biological replicates). F Immunoblot analysis of nuclear HIF-1α protein levels in WT and cKO BMDMs stimulated with LPS (upper) or Pam3CSK4 (lower) for the indicated times. G Dual-luciferase reporter analysis of HIF-1α activity in WT and cKO BMDMs 48 h after transfection with the HRE-luciferase reporter plasmid (n = 4 biological replicates). H ChIP‒qPCR analysis of HIF-1α enrichment at the gene loci of Il1b, Hk2 and Slc2a1 in WT and cKO BMDMs stimulated with or without LPS (n = 3 biological replicates). I Q-PCR analysis of Il1b mRNA levels in WT and cKO BMDMs pretreated with DMSO or 2-ME and then stimulated with LPS (n = 3 biological replicates). *P < 0.05; **P < 0.01; ***P < 0.001. Unpaired two-tailed Student’s t test (D–E, G–I)

HIF-1α is widely accepted as the core transcription factor in glycolytic metabolism for the inflammatory capacity of macrophages [34]. Since our above experiments revealed that total HIF-1α expression was increased due to RBM25 deficiency (Fig. 3H), we sought to examine the participation of HIF-1α in RBM25-mediated coordination of glycolysis with gene transcription. Immunoblotting confirmed the increased accumulation of HIF-1α in the nuclear fraction of RBM25 cKO macrophages upon stimulation (Fig. 4F). We next performed hypoxia response element (HRE)-driven luciferase reporter experiments to measure HIF-1α activity in response to inflammatory stimulation. Although no obvious alteration in HRE activity was detected in untreated cells, RBM25 deficiency increased HIF-1α activity after LPS treatment (Fig. 4G). The ChIP‒qPCR results revealed increased enrichment of HIF-1α at the gene promoter regions of both proinflammatory genes and core glycolytic genes (Fig. 4H). Inhibition of HIF-1α activity with 2-methoxyestradiol (2-ME) significantly blocked the immunostimulatory effect owing to RBM25 deficiency (Fig. 4I). These results highlight the importance of histone acetylation and HIF-1α activity in the RBM25-mediated transcriptional suppression of inflammatory macrophages.

Identification of Acly pre-mRNA as direct splicing RNA target of RBM25

Considering that RBM25 controls alternative splicing, especially pre-mRNA binding and early spliceosome formation [35], we sought to identify the functional target transcripts of RBM25 involved in metabolic reprogramming in macrophages. We subsequently analyzed the AS events from the transcriptome-sequencing via rMATS (http://rnaseq-mats.sourceforge.net/index.html), a typical software for identifying the difference in AS events between groups, which divides AS events into five main categories: skipped exon (SE), alternative 5’ splice site (A5SS), alternative 3’ splice site (A3SS), mutually exclusive exon (MXE) and retained intron (RI) (Fig. 5A). We then quantified the differences in these five AS events in WT and RBM25 cKO macrophages during different periods of LPS treatment (Fig. 5B). Among them, RBM25-controlled SE accounted for the majority of all AS events, whereas RBM25 had little effect on other kinds of AS events, as displayed by the volcano plot representing the delta percent-splice-in (PSI) (Figs. 5C and S10A–D). Further analyses were performed to characterize the regulated cassette exons by dividing all SE events into three groups (PSI-UP: FDR < 0.05 & IncLevelDifference > 0.05; PSI-DOWN: FDR < 0.05 & IncLevelDifference < −0.05; not significant, NS: the others). We found the greatest change in the Splicing Maxent score of the PSI-UP group, in terms of 3’ SS sites, compared with that of the NS group (Fig. S10E). No difference in the Maxent score at the 5’SS sites was found between the NS group and the DOWN group or UP group (Fig. S10E). Moreover, the UP group presented greater exon length, and the DOWN group presented lower exon length, as expected (Fig. S10F). Finally, we detected a greater GC content (ratio) of the cassette exons regulated by RBM25 (Fig. S10G), suggesting that the AS pattern of RBM25 might be determined or largely influenced by the guanine (G) or cytosine (C) content of potential skipping exons. To identify the core elements that are directly bound and controlled by RBM25 in macrophages, we performed naïve RNA immunoprecipitation sequencing (RIP-seq) using an antibody against RBM25, and the motif recognized by RBM25 was determined to be GCGGGA, which is similar to the well-accepted RNA-binding motif described previously [35, 36] (Fig. S10H, I). The functional annotation of RBM25-bound transcripts revealed enrichment in various metabolic pathways, including central carbon metabolism and the TCA cycle (Fig. S10J). In this way, we validated the hypothesis that RBM25 might license the metabolic phenotype during proinflammatory macrophage activation at the posttranscriptional level. A comprehensive analysis combining RIP-seq assays with SE scores by rMATS was performed, and the positive effect of RBM25 on the exon skipping of its target pre-mRNAs was confirmed (Fig. 5D). Among these SE transcripts with direct RBM25-binding signals, several cellular metabolic processes ranked first in the top 10 GO functional categories (Fig. 5E). Intriguingly, we noted a prominent decrease in the expression of 14^th exon included Acly variant in RBM25 cKO macrophages (Fig. 5F and Table S1). Acly is the primary enzyme responsible for the synthesis of acetyl-CoA (ac-CoA) and oxaloacetate (OAA) from citrate, which has been proven necessary for oxidized LDL cholesterol- or TLR-mediated macrophage activation [37, 38]. Semiquantitative PCR (Semi-PCR) revealed two isoforms of Acly mRNA in RBM25 WT macrophages, one containing exon 14 (Acly L) and the other not (Acly S) (Fig. 5G, H). However, in RBM25 cKO macrophages, Acly S occupied a dominant position. Interestingly, there was a moderate decrease in Acly L mRNA during the late period (after 24 h) after LPS stimulation, which might be attributed to decreased RBM25 expression (Fig. 5H). To confirm the AS of Acly pre-mRNA, we constructed an Acly exon 14 minigene splicing reporter system as previously reported [39], in which exon 14 skipping retained the intact open reading frame (ORF) of red fluorescent protein (RFP), whereas exon 14 inclusion disrupted this ORF and subsequent RFP production. As shown in Fig. 5I, the RFP signal was impaired by RBM25 overexpression in contrast to the control GFP signal, verifying the role of RBM25 in controlling Acly 14^th exon skipping.

Fig. 5. Identification of Acly pre-mRNA as the direct splicing RNA target of RBM25.

A Schematic diagram of the 5 main AS events determined by rMATS. B Numbers of different AS events (cKO vs WT) in BMDMs stimulated with LPS for the indicated times. C Volcano plot of different SE events in WT and cKO BMDMs. D Combined analysis of SE events and RIP-seq data acquired with an antibody against RBM25. E Top 10 GO functional categories of different SE events overlapping with RIP-seq peaks in BMDMs stimulated with LPS. F Exon expression and PSI analysis of Acly exon 14. G Schematic diagram of Acly isoforms encoded by two separate transcripts with or without exon 14. H Semi-PCR of Acly exon 14 in WT and cKO BMDMs stimulated with LPS for the indicated times, with Actb used as the control. I Splicing minigene reporter system for detecting Acly exon 14 expression. The fluorescence signal in the RFP channel represented the exon 14 splicing efficiency in HEK293T cells transfected with the Mock or RBM25-Flag plasmid, and the fluorescence signal in the GFP channel served as the transfection control. J RIP‒qPCR analysis of RBM25 enrichment on Acly RNA in BMDMs stimulated with or without LPS (n = 3 biological replicates). K Schematic diagram of RBM25 WT and two mutant plasmids. L RIP‒qPCR analysis of the enrichment of Flag-RBM25 WT or mutants on Acly RNA as in J (n = 3 biological replicates). ChIP‒qPCR analysis of H3K9ac enrichment at the promoter of the Il1b gene (M) and Q‒PCR analysis of Il1b mRNA levels (N) in WT and RBM25 cKO BMDMs pretreated with different concentrations of SB204990 followed by stimulation with LPS (n = 3 biological replicates). **P < 0.01; ***P < 0.001. Unpaired two-tailed Student’s t test (J) and one-way ANOVA (L–N) were used

Murine RBM25 is an 857 amino acid protein that consists of two RNA-binding regions: the N-terminal RRM (RNA-recognition motif) and the C-terminal PWI domain [35]. First, RIP‒qPCR confirmed the direct binding between RBM25 and Acly pre-mRNA (Fig. 5J). To map the domain of RBM25 for binding Acly transcripst, we generated a full-length RBM25-expressing plasmid (Rbm25-FL) and two truncated plasmids lacking the N-terminal RRM domain (Rbm25-ΔRRM) or lacking the PWI domain (Rbm25-ΔPWI) (Fig. 5K). We reconstituted full-length or truncated RBM25 in RBM25 cKO BMDMs to determine the functional domain. RIP‒qPCR assays revealed that, compared with Rbm25-FL or Rbm25-ΔPWI, Rbm25-ΔRRM lost the ability to bind Acly pre-RNA (Fig. 5L). N6-methyladenosine (m6A) modification reportedly influences RNA metabolism, including RNA AS [40]. However, knocking down either the m6A writer METTL3 or the eraser ALKBH5, two main m6A modifiers in macrophage biology [41, 42], barely affected the skipped exon of Acly (Fig. S10K, L), indicating that this AS event is independent of RNA m6A modification. Finally, we treated RBM25 cKO macrophages with the Acly-specific inhibitor SB204990, which effectively suppressed Acly activity and the level of its metabolic product ac-CoA [43]. SB204990 dose-dependently decreased both the histone H3K9ac level and the mRNA expression of Il1b (Fig. 5M, N). In this part, we identified Acly as the direct target of RBM25 in the regulation of proinflammatory activation and effector functions.

The acquisition of Acly short isoform by 14th exon skipping supports immunostimulatory macrophage function

To verify the biological differences between Acly L and Acly S in terms of macrophage activation, three separate siRNAs targeting exon 12, exon 14 and exon 17 of Acly were designed and transfected into macrophages. Compared with Ctrl siRNA, siRNA specifically targeting exon 14 (Acly E14 siRNA) effectively inhibited Acly L expression and promoted Acly S expression, whereas both Acly L and Acly S were silenced in E12 or E17 siRNA-transfected macrophages (Fig. 6A). Using this model, we tested the effects of the Acly exon 14 variants on the macrophage transcriptome. E14 siRNA transfection resulted in alterations in global gene expression in proinflammatory macrophages triggered by LPS, and the DEGs were enriched in several important pathways associated with immune system activation and inflammatory diseases (Fig. 6B, C). We noted the upregulation of proinflammatory genes, such as Il1b and Siglec1, after Acly E14 siRNA transfection, which phenocopied the phenomenon in RBM25 cKO macrophages (Fig. 6D). Q‒PCR confirmed the transcriptome sequencing results showing that Acly E14 siRNA significantly promoted LPS-induced Il1b and Il6 transcription but not Tnf transcription (Fig. 6E). As a control, E17 siRNA, which silenced total Acly expression, inhibited Il1b and Il6 transcription (Fig. 6E), which was the same conclusion reported previously [37, 38]. In addition, E14 siRNA selectively accelerated NLRP3 and pro-IL-1β protein synthesis at the transcriptional level during inflammasome priming stage (Fig. 6F, G). Next, we examined the metabolic status via the ECAR and OCR and found that E14 siRNA significantly promoted glycolysis, as indicated by the energetic and glycolytic phenotypes of inflammatory macrophages (Fig. 6H, I). In contrast, decreases in the basal OCR and maximal OCR were observed in macrophages transfected with Acly E14 siRNA (Fig. 6J). The TEM results revealed increased mitochondrial fission and damage in Acly E14 siRNA-transfected macrophages (Fig. 6K). In addition, the Acly S upregulation mediated by Acly E14 siRNA transfection in macrophages increased the production of acetyl-CoA in response to LPS treatment (Fig. 6L). E14 siRNA transfection led to greater enrichment of H3K9ac and H3K27ac at the gene promoters of Il1b and Il6, respectively (Fig. 6M, N). HIF-1α expression and activity were also increased in Acly E14 siRNA-transfected macrophages, as indicated by immunoblotting and HRE-luciferase activity determination (Fig. 6G, O). Collectively, these results demonstrate that RBM25 deficiency promotes macrophage metabolic activation via the increased generation of alternative spliced Acly S.

Fig. 6. The acquisition of Aclyl short isoform by 14^th exon skipping supports immunostimulatory macrophage function.

A Semi-PCR analysis of Acly and Actb mRNA levels (upper) and immunoblot analysis of Acly, H3Ac and β-actin protein levels (lower) in BMDMs 48 h after transfection with the indicated siRNAs. Number of DEGs (E14 siRNA vs Ctrl siRNA) (B), functional annotation of DEGs (C) and heatmap of DEGs (D) in BMDMs transfected with the indicated siRNAs and then stimulated with LPS. Q-PCR analysis of the indicated proinflammatory cytokine (E) and Nlrp3 (F) mRNA levels and immunoblot analysis of HIF-1α, NLRP3 and pro-IL-1β protein levels (G) in BMDMs transfected with the indicated siRNAs, followed by stimulation with LPS for the indicated times (n = 3 biological replicates). ECAR analysis (H), ratio of the basal ECAR to the basal OCR (I) and OCR analysis (J) of BMDMs transfected with control siRNA or Acly E14 siRNA and then stimulated with or without LPS. K TEM images of mitochondria in BMDMs transfected with control siRNA or Acly E14 siRNA and then stimulated with or without LPS. Acetyl-CoA abundance (L), ChIP‒qPCR analysis of H3K9ac (M) and H3K27ac (N) modifications at the gene loci of Il1b and Il6, and dual-luciferase reporter analysis of HIF-1α activity (O) in BMDMs transfected with control siRNA, Acly E14 siRNA or Acly E17 siRNA followed by LPS stimulation or not (n = 3–4 biological replicates). *P < 0.05; **P < 0.01; ***P < 0.001, ****P < 0.0001. Unpaired two-tailed Student’s t test (E–F, J) and one-way ANOVA (L–O) were used

Acly activity is influenced by protein lactylation at K918/995 through P300

Despite the established catalytic process of Acly, which transforms citrate and ATP into ac-CoA and OAA for cellular metabolism, no evidence has revealed the biological importance of 14^th exon skipping of Acly in its metabolic activity. Nontargeted metabolomics by LC‒MS was performed and confirmed the alteration in Acly metabolic function in proinflammatory macrophages, which transformed more citrate into downstream metabolites (Fig. S11A‒C). DMM treatment also abrogated the immunoregulatory effects of Acly E14 siRNA on HIF-1α and pro-IL-1β expression, suggesting the importance of glycolysis rewiring and succinate accumulation in macrophages with 14^th exon skipping of Acly (Fig. S11D). The metabolic activity of Acly has been linked with  protein posttranslational modifications (PTMs), such as Ser455 phosphorylation [38, 44]. However, Ser455 phosphorylation was unchanged in RBM25 cKO macrophages when a commercial antibody specific for Acly S455 phosphorylation was used (Fig. S11E). A previous study revealed that the tetramer of Acly was essential for substrate binding and catalysis [45, 46], and we hypothesized that the interaction of Acly S and Acly L had a regulatory effect on the stability of the Acly forming tetramer complex. We tested the interaction of Acly S and Acly L through cotransfection of HEK293T cells with Acly L or Acly S with a Flag or HA tag, according to a previously reported method [46]. However, there was no difference in the immunoprecipitation of Acly L with Acly S, and vice versa, indicating that the alteration in enzyme activity was not due to multimerization (Fig. S11F).

To further characterize the biological differences between the Acly two isoforms, we generated an Acly KO iBMDM cell line via two separate sgRNA-mediated CRISPR/Cas9 method (Fig. S12A) and confirmed the effective knockout (Fig. S12B). Knockout inhibited proinflammatory cytokine transcription, HIF-1α expression, AKT‒mTOR pathway activation and cellular glycolysis, as reported previously (Fig. S12C–E). We reconstituted Acly KO iBMDMs with lentivirus-Acly L or lentivirus-Acly S with a Flag-Tag and stimulated them with LPS (Fig. S12F). We found that Acly S reconstitution in iBMDMs resulted in increased glycolysis, ac-CoA production and proinflammatory gene transcription, which was consistent with our previous findings from siRNA assays (Fig. S12G–I).

PTM is closely influenced by cellular metabolism and is also important for determining the activity of metabolic enzymes; therefore, we detected Acly PTMs, such as acetylation, succinylation and lactylation, via immunoblot assays. Interestingly, we found that Acly L was subjected to protein lactylation, whereas Acly S was not (Fig. 7A). The protein lactylation of Acly L was reversibly catalyzed by treatment with lactic acid (LA) or lactate dehydrogenase (LDHA) inhibitors (Fig. 7B, C). Notably, inflammatory stimuli induced the protein lactylation of Acly L in a time-dependent manner but not that of Acly S in macrophages (Fig. 7D). Next, we investigated the role of protein lactylation in Acly activity and immunoregulatory effects. LA treatment of macrophages expressing Acly L effectively led to a large decrease in enzyme activity and inflammatory gene transcription, while it barely affected these aspects of macrophages expressing Acly S (Fig. 7E, F). These data suggest that the two Acly isoforms are distinguished by protein lactylation in proinflammatory macrophages, which might influence their metabolic and immunoregulatory activities. To search for lactyltransferases for Acly lactylation, we expressed each of the most previously reported lactyltransferases [47, 48] and found that only the E1A-binding protein P300 facilitated Acly protein lactylation (Fig. 7G). Importantly, we mapped the domains responsible for the interaction between P300 and Acly via protein‒protein docking via the HDOCK server [49, 50] and PLIP [51] (Fig. 7H). Intriguingly, Acly was predicted to interact with P300 in a manner dependent on the hydrogen bond between Arg-483 of Acly and Met-1490 of P300; the former was located in the 14^th exon-encoding protein domain (Fig. S13A). Acly L could directly interact with P300, whereas Acly S, which lacks the 14^th exon due to RBM25-mediated exon skipping, could not (Fig. 7I). This defect in physical interaction might account for the difference in protein lactylation between Acly L and Acly S in inflammatory macrophages and HEK293T cells (Fig. 7D, J). K918/995 were further identified as lysine sites subjected to protein lactylation modification via LC‒MS/MS technology (Figs. 7K and S13B), and mutation of these lysine sites abrogated the effect of the lactyltransferase P300 (Fig. 7L). In addition, mutation of K918/995 to alanine also abrogated the inhibitory effects of LA on Acly enzyme activity and on macrophage-derived inflammatory mediators (Fig. 7M, N). These data prove that protein lactylation is a major important factor for regulating Acly activity and explain why the two Acly isoforms have different metabolic and immune functions in proinflammatory macrophages.

Fig. 7. Aclyl activity is influenced by protein lactylation at K918/995 through P300.

A Immunoblot analysis of the indicated PTMs of Acly-L and Acly-S in Acly KO iBMDMs rescued with Acly-L, Acly-S or Mock construct, and then stimulated with LPS for 8 h (la, lactylation; ace, acetylation; suc, succinylation). B Immunoblot analysis of the lactylation of Acly L in Acly KO iBMDMs rescued with the Acly L construct, followed by treatment with the indicated doses of lactic acid (LA). C Immunoblot analysis of the lactylation of Acly L in Acly KO iBMDMs rescued with the Acly L construct and then treated with 5 mM LA and pretreated with FX-11, GSK2837808A or DMSO. D Immunoblot analysis of the lactylation of Acly L and Acly S in Acly KO iBMDMs rescued with the Acly L or Acly S construct and then stimulated with LPS for the indicated times. Aclyl enzyme activity (E) and Q‒PCR analysis of Il1b, Il6 and Cxcl10 mRNA levels (F) in LPS-stimulated Acly KO iBMDMs rescued with the Acly L or Acly S construct and then treated with 5 mM LA. G Immunoblot analysis of the lactylation of Acly L in HEK293T cells overexpressing the indicated individual HATs. H Molecular docking analysis of the protein interaction of Acly L with P300. I Co-IP analysis of Flag-tagged Acly L or Acly S with HA-tagged P300 in HEK293T cells overexpressing the indicated plasmids. J Immunoblot analysis of the lactylation of Acly L or Acly S in HEK293T cells overexpressing the indicated plasmids. K Identification of modified lysine residues K918 (upper) and K995 (down) on Acly L in iBMDMs via IP combined with mass spectrometry. Major peaks on the spectrum were assumed to be generated by internal cleavage. L Immunoblot analysis of the lactylation of WT Acly L and mutants in HEK293T cells overexpressing the indicated plasmids. Acly enzyme activity (M) and Q‒PCR analysis of Il1b, Il6 and Cxcl10 mRNA levels (N) in LPS-stimulated Acly KO iBMDMs rescued with Acly L (WT) or the Acly L mutant (K918/995 A), followed by treatment with 5 mM LA. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Unpaired two-tailed Student’s t test (E, F, M and N)

Next, we determined the difference between Acly L and Acly S in terms of their different affinities for the metabolic substrates citrate and ATP. Through a molecular docking model, the energy of the Acly substrate interaction was calculated via BIOVIA Discovery Studio software. The results showed that the affinity of Acly S for ATP was greater than that of Acly L, while there was no difference in the affinity of Acly S for citrate (Fig. S13C). To verify the predicted results of the molecular docking experiments, surface plasmon resonance (SPR) assays were performed to determine the binding affinities of the two Acly isoforms to their metabolic substrates, ATP or citrate, with purified Acly L and Acly S, respectively. The results revealed that the K_D of Acly S for ATP was 21.12 μM, whereas the K_D of Acly L for ATP was 32.91 μM, suggesting a stronger binding capacity of Acly S for ATP (Fig. S13D, E). However, no obvious difference was found in the binding capacity of the Acly isoforms for citrates (Fig. S13F, G). These results also provide a possible reason for the difference in enzyme activity between the two Acly isoforms. However, the relationship between protein lactylation of Acly and metabolic substrate affinity is not clear and remains to be fully elucidated.

Pharmacological inhibition of Acly ameliorates macrophage-mediated arthritis and aging-related inflammation

Our above data demonstrate that RBM25 deficiency triggers misregulation of exon skipping and the metabolic activity of Acly, facilitating glycolytic reprogramming in macrophages and resulting in abnormal activation of inflammatory responses. Therefore, we wanted to verify whether Acly inhibitors could alleviate RBM25-associated inflammatory diseases, such as arthritis. First, we confirmed the inhibitory effects of SB204990 on both Acly isoforms. Enzyme activity assays demonstrated that high concentrations of SB204490 effectively reduced Acly activity in proinflammatory macrophages (Fig. S14A). Next, a CIA model was generated to determine the in vivo effect of SB204990. SB2014990 was i.p. administered to WT and RBM25 cKO mice after the second immunization in the CIA model, with MCC950, an inhibitor of the NLRP3 complex, used as a positive control (Fig. S14B). As indicated by the clinical scores of the mice with arthritis, SB204990 effectively alleviated joint inflammation in the mice (Fig. S14C). SB204990 attenuated lymph node enlargement and splenomegaly of CIA model mice, indicating that pharmacological inhibition of Acly had an effective therapeutic effect on arthritis immunopathology (Fig. S14C, D). H&E histological analysis further revealed the effectiveness of SB204990 as a potent therapeutic strategy for suppressing joint inflammation, including synovial hyperplasia, inflammatory cell infiltration, and bone destruction, in RBM25 cKO mice (Fig. S14E, F).

Recent works have focused on age-related changes in the immune system, known as inflammaging, and increased secretion of proinflammatory mediators, which are the major results of aging-associated chronic inflammation with high circulating levels of IL-1β and IL-6 in joints and other organs [52, 53]. Considering the role of RBM25 as a checkpoint of inflammatory macrophages in microbe- or arthritic-induced inflammatory responses, we explored the influence of RBM25 deficiency on age-related inflammation in vivo. As expected, in the natural aging model, RBM25 cKO mice spontaneously developed arthritis and even multiple-organ inflammation, as indicated by H&E histological analysis (Fig. 8A). With increasing age, RBM25 deficiency resulted in more severe and extensive signs of inflammatory cell infiltration in the above tissues or organs (Fig. 8A). Immunofluorescence assays revealed that CD4⁺ T cells and Ly6G⁺ neutrophils accounted for the majority of infiltrating immune cells in the RBM25 cKO aging mice (Fig. 8B). In addition, a high proportion of NLRP3⁺ cells in these organs subjected to immunopathology injuries was observed, suggesting that intensive activation of the NLRP3 inflammasome might be an important cause of aging pathogenesis (Fig. 8B). The levels of autoantibodies against double-stranded DNA (anti-dsDNA) and nuclear antigens (ANAs) were significantly greater in the RBM25cKO mice than in the control mice, especially during the 18-month-old period (Fig. 8C, D). In accordance with the severe inflammation in the lung tissues, the mRNA levels of Il1b, Il6 and Nlrp3 were dramatically elevated in the RBM25cKO mice (Fig. 8E). The serum levels of IL-1β and IL-6 were also highly induced by aging in cKO mice compared with WT mice (Fig. 8F). Furthermore, we investigated whether the aggravated inflammatory phenotype in RBM25 cKO mice was mediated by changes in Acly activity. We treated these aging mice with the Acly inhibitor SB204990 and detected an effective decrease in the mRNA levels of proinflammatory genes in lung tissues and in the serum levels of IL-1β and IL-6 (Fig. 8G, H). These data indicate that RBM25 may be a negative regulator of inflammation in an Acly dependent manner.

Fig. 8. RBM25 cKO mice develop aging-associated inflammation in multiple organs and can be ameliorated by Acly inhibitor.

A H&E staining of the wrist joint, ankle joint, colon, lung, liver and kidney from WT and RBM25 cKO mice at the indicated ages. B IF analysis and statistics of CD45⁺, CD4⁺, CD8⁺, Ly6G⁺ and NLRP3⁺ cells in the lungs, liver and kidneys of aging RBM25 cKO mice. ELISA analysis of anti-dsDNA (C) and ANA (D) concentrations in sera from WT and RBM25 cKO mice at the indicated ages (n = 8 per group). E Q‒PCR analysis of Il1b, Il6 and Nlrp3 mRNA levels in lung tissues from WT and RBM25 cKO mice at 18 months of age (n = 8 per group). F ELISA analysis of the IL-1β and IL-6 concentrations in the sera of 18-month-old WT and RBM25 cKO mice (n = 8 per group). G Q‒PCR analysis of Il1b, Il6 and Nlrp3 mRNA levels in lung tissues from 18-month-old RBM25 cKO mice treated with vehicle or SB204990 (10 mg/kg body weight) for 2 weeks (n = 5 per group). H ELISA analysis of the IL-1β and IL-6 concentrations in the sera of the mice described in G (n = 5 per group). *P < 0.05; **P < 0.01; ***P < 0.001. Unpaired two-tailed Student’s t test (C–H)

To further investigate the biological importance of Acly SEs in inflammatory responses in vivo, we generated antisense oligonucleotides (ASOs) targeting the Acly 14^th exon, which were also named splicing-switching antisense oligonucleotides (SSOs) [54] (Fig. S15A). SSOs have been proven to regulate exon skipping, intron retention and other types of pre-mRNA AS in vivo and in vitro [54, 55]. Acly E14 ASO transfection effectively promoted 14^th exon skipping in BMDMs to upregulate Acly S production (Fig. S15B). Transcriptome sequencing assays revealed the upregulation of several macrophage-derived proinflammatory genes and other genes related to immune system activation and the downregulation of immunosuppressive membrane receptors, including Cd300e and Cxcr4 (Fig. S15C). Functional enrichment analysis also revealed that many pathways involved in inflammatory responses and cytokine signaling pathways, as well as genes associated with RA pathogenesis, were upregulated in cells transfected with Acly E14 ASO, which phenocopied the effects of RBM25 deficiency (Fig. S15D). In vivo experiments involving i.v. injection of Acly E14 ASO or Ctrl ASO were performed, which revealed that mice injected with Acly E14 ASO presented increased inflammatory damage to their lung tissues, with bleeding and inflammatory cell infiltration (Fig. S15E, F). The serum levels of IL-1β and IL-6 were also significantly elevated in the mice treated with Acly E14 ASO, indicating a stronger inflammatory response in the systemic inflammation model (Fig. S15G). Overall, we demonstrated that the Acly E14 ASO can mimic the predominant expression of Acly S mediated by RBM25 deficiency to enhance the inflammatory response in macrophages, which may be an unfavorable factor for several inflammatory and autoimmune diseases, such as RA (Fig. S16).

Discussion

Previous studies have demonstrated that RBM25 is a master regulator of the function of U1 spliceosome ribonuclear protein, whose expression is strictly controlled to determine the physiological function of host cells and affect the pathogenic progress of multiple diseases. Angiotensin II and hypoxia conditions during heart failure induce remarkable expression of RBM25 and thereby RBM25-mediated upregulation of the expression of the sodium voltage-gated channel alpha subunit 5 (SCN5A) variants E28C and E28D, resulting in decreased Na⁺ current and acquired cardiac diseases [56]. In addition, RBM25 was identified as a novel tumor suppressor via in vivo short hairpin RNA (shRNA) screening in a mouse acute myeloid leukemia (AML) model through the modulation of alternative splicing of MYC inhibitor bridging integrator 1 (BIN1), which was closely correlated with MYC activity and poor AML outcomes [57]. These studies show that RBM25 upregulation leads to cardiac diseases or tumors; however, whether decreased expression of RBM25 in specific immune cells results in immune disorders and diseases remains unknown. Through combined analysis of RNA-seq and scRNA-seq data from RA patients and arthritis model mice, we revealed the notably decreased RBM25 expression in inflammatory macrophages. Recently, another work investigating the involvement of AS in the toxic mechanism induced by 2,3,7,8-tetrachlorodibenzo-p-dioxin also revealed the aberrant expression of genes within the spliceosome pathway in various host cells, especially RBM25 in THP-1 monocytes [58]. Our in vitro experiments also demonstrated that new synthesis of RBM25 nascent RNA was suppressed by multiple inflammatory stimuli, including LPS and nucleic acid. This might be attributed to defects in the deposition of histone H3K36me3 and chromatin accessibility. Alterations in H3K36me3 levels in the gene body are well known to reflect productive RNA Pol II elongation of gene transcripts, which is dynamically controlled by the histone methyltransferase SETD2 and the demethylase KDM4 family [59, 60]. The epigenetic mechanism underlying how histone modifications and chromatin modifiers, especially H3K36me3 modifications regulated by SETD2/KDM4, affect the transcriptional regulatory process of RBM25 during infection or other immune disorders might be worth exploring in further studies.

Our comprehensive analysis of RIP-seq and transcriptome sequencing data revealed the direct interaction of RBM25 with Acly transcripts, which encode key metabolic enzymes involved in glucose metabolism, cholesterol and fatty acid synthesis, and the inflammatory cascade [61]. Several separate studies have proven the beneficial effects of targeting Acly in shaping protective immunity against multiple inflammatory diseases, such as atherosclerosis and metabolic dysfunction-associated steatohepatitis [62, 63]. Recent studies reported the indispensable role of Acly in metabolic reprogramming in activated macrophages and T cells, both of which require the metabolic switch from oxidative phosphorylation to aerobic glycolysis for their effector functions [64, 65]. Acly dependent acetyl-CoA has been shown to support the histone acetylation of the Il1b and Il6 gene loci in macrophages [37, 66] and the Il17a and Il17f gene loci in Th17 cells [64]. Nevertheless, despite increasing inflammatory responses in vitro, Acly deficiency in macrophages does not affect acute or chronic inflammatory responses in vivo, especially in experimental autoimmune encephalomyelitis (EAE), an experimental model of multiple sclerosis [67]. Therefore, therapeutic targeting of Acly in macrophages might be beneficial in some, but not all, inflammatory diseases. With the acceptance that the senescence-associated secretory phenotype (SASP) is the representative pathogenic factor in aging and aging-associated inflammatory diseases, including RA and Alzheimer’s disease, targeting RBM25-mediated alternative splicing or pharmacological inhibition of Acly might be a promising therapeutic target for treating inflammatory diseases, as effective senotherapy blocks the aging phenotype through SASP inhibition [68]. In this work, treatment with the Acly inhibitor SB204990 effectively ameliorated the inflammatory response and tissue damage in both autoimmune arthritis and aging-associated inflammation. Further studies should explore the relationship between RBM25 and Acly pre-mRNA splicing in prospective long-term treatments for infectious or  other inflammatory disorders.

This work identified Acly as the candidate transcript regulated by RBM25-dependent alternative splicing. In addition to Acly, several other metabolic enzymes that participate in the citrate cycle, e.g., isocitrate dehydrogenase 3 (Idh3) and oxoglutarate dehydrogenase (Ogdh), were also identified as RBM25-binding RNAs in our data. The transcriptome sequencing results also revealed that skipping exons in Idh3 or Ogdh was influenced by RBM25. However, considering the well-known modulatory role of these two enzymes in macrophage metabolism and activation [69, 70], the physiological functions of these metabolic enzymes in RBM25-mediated alternative splicing and immunoregulatory effects remain to be further investigated. Alternative splicing of metabolic enzymes has been reported to regulate the host metabolic phenotype in tumors and other mammalian cells. For example, there are two main isoforms of PKM resulting from mutually exclusive PKM pre-mRNA, reflecting the inclusion of either exon 9 (PKM1) or exon 10 (PKM2): the former promotes oxidative phosphorylation, whereas the latter promotes aerobic glycolysis, particularly in tumors and other proliferating cells [11, 12]. The selective generation of two transcripts has been reported to involve posttranscriptional processing steps involving hnRNP proteins such as hnRNP I, hnRNP A1 and hnRNP A2B1. In proliferating tumor cells, the oncogenic transcription factor c-Myc upregulates the expression of these three hnRNP proteins and promotes selective PKM2 generation for high glucose consumption and energy requirements [71]. However, little is known about the alternative splicing of Acly transcripts. Our data demonstrate that the 14^th exon skipping mediated by RBM25 deficiency generates the Acly S isoform, which reprograms macrophage metabolism for citrate-derived acetyl-CoA production and glycolysis, leading to increased histone acetylation and HIF-1α activity. Nevertheless, the mechanism underlying the difference in metabolic activity between Acly L and Acly S remains unknown. Protein lactylation is a newly identified type of posttranslational modification regulated by glycolysis metabolism and lactate production and is involved in the regulation of macrophage function and fate [72, 73]. In our work, we revealed that the Acly L isoform was subjected to protein lactylation upon LPS treatment or lactic acid accumulation, repressing its metabolic activity and immunostimulatory functions, which reflected the phenomenon that lactate-lactylation is involved in metabolic reprogramming and immunosuppression, which is consistent with what has been described previously [74]. Through molecular docking analysis, we found that the domain encoded by the 14^th exon was necessary to physically interact with the lactyltransferase P300, which also functions as a flexible bridge linking the ATP-Grasp domain and CoA-binding domain [45, 75]. Mutation of lysine residues 918/995 notably inhibited P300-catalyzed protein lactylation but fostered Acly activity in macrophage metabolism and epigenetic remodeling. Thus, we provide a preliminary explanation for the inhibition of enzyme activity by P300-mediated Acly lactylation, which is strictly regulated by RBM25-dependent spliceosome activity.

ASOs were developed as potent activators of disease-specific alternative exon splicing. For example, ASOs that switch CACNA1C exon utilization from 8A to 8 represent a potential therapeutic strategy for treating Timothy syndrome [76]. ASOs for PKM isoform switching have been shown to be effective in the treatment of hepatocellular carcinoma through the modulation of pyruvate kinase activity and alteration of glucose metabolism [77]. In this way, the specific ASOs for Acly 14^th exon switching remain to be explored for potential splicing therapy for RBM25-associated immune disorders. Moreover, Li et al. designed guide RNA-directed, deactivated Cas CRISPR enzymes fused to splicing effectors and identified dCasRx-RBM25 as a potent regulator of multiplexed activation and repression of exons, which might also shed light on splicing-directed therapeutic applications [78]. Finally, the role of the splicing checkpoint of RBM25 in Acly pre-mRNA splicing highlights its notable value in research and clinical intervention for various human inflammatory, autoimmune, and aging diseases.

Methods

Research with human participants

PBMCs from 38 subjects (28 RA patients and 10 healthy controls) were collected from Changhai Hospital (Shanghai, China), as described previously [79]. All the RA patients met the 2010 criteria of the American College of Rheumatology and the European Union League Against Rheumatism. Active RA was defined as having a 28-joint disease activity score (DAS28) of 2.6 or higher. For healthy controls, subjects with severe cardiovascular diseases, liver and kidney dysfunction, malignant tumors and other autoimmune or musculoskeletal disorders were excluded. All samples were collected with written informed consent from the patients and healthy controls. This study was approved by the Medical Ethics Committee of Naval Medical University (Shanghai, China).

Mice

Rbm25^fl/fl mice on a C57BL/6 background were generated via the CRISPR/Cas system (Cyagen Biosciences, China). Exons 3-4 were selected as the conditional knockout region. Deletion of this region resulted in the loss of function of the Rbm25 gene. Rbm25^fl/fl mice were then crossed with B6.129P2-Lyz2^tm1(cre)Ifo/J mice (Jackson Laboratory, #004781) to construct macrophage-conditional RBM25 knockout mice (Rbm25^fl/flLyz2-cre⁺). IL-1 receptor KO mice (B6.129S7-Il1r1^tm1Imx/J mice, #003245) were obtained from the Jackson Laboratory and crossed with RBM25 cKO mice, as described in the manuscript. WT C57BL/6 mice were purchased from Shanghai SIPPR-BK Laboratory Animal Company (Shanghai, China). The mice were bred under specific pathogen-free conditions. All animal experiments were performed in accordance with the National Institutes of Health’s Guide for the Care and Use of Laboratory Animals with the approval of the Medical Ethics Committee of Naval Medical University (Shanghai, China).

Collagen-induced arthritis model

Eight-week-old mice were used to establish CIA models as previously described [79, 80]. Bovine type II collagen emulsified in complete Freund’s adjuvant (CFA) was injected into the mouse tail base on day 0 for the first immunization. A booster immunization was then administered on day 21. We evaluated the severity of redness and swelling in the wrist and paw on a scale of 0–4. The maximum score is 16. All the mice were scored in a blinded manner. Joints were fixed in paraformaldehyde for further H&E and Safranin O staining. Sera were collected for cytokine measurement via ELISA. Spleen and pLNs were harvested for analyzing immune cell clustering via a fluorescence-activated cell sorter.

Acute inflammation model

For the LPS-induced sepsis model [81], 8-week-old mice were i.p. injected with LPS (10 mg/kg body weight, Sigma). The sera and lung tissues were harvested for ELISA and H&E analysis 8 h after LPS injection. For the MSU-induced peritonitis model [82], 8-week-old mice were i.p. injected with MSU (1 mg per mouse in 500 μL of saline, InvivoGen). Peritoneal cavity fluid was collected for FACS analysis or ELISA after ultrafiltration and was concentrated by centrifugal filter units (Millipore).

Primary macrophage culture and stimulation

Mouse PMs and BMDMs were prepared as previously described [83, 84]. AMs were pooled from bronchoalveolar lavage fluid (2 mM EDTA in PBS) and purified via adherence for 2 h in RPMI-1640 (Gibco) supplemented with 10% FBS (Gibco) and 1% Pen/Strep (MedChemExpress LLC) at 37 °C and 5% CO₂. The macrophages were stimulated with LPS (100 ng/ml, Sigma) for 4 h unless otherwise specified. For NLRP3 inflammasome activation, macrophages were primed with LPS (100 ng/ml) and then stimulated with ATP (5 mM, Invitrogen), nigericin (10 μM, Invitrogen) or MSU (200 μg/ml, Invitrogen).

Cell lines and CRISPR/Cas9-mediated knockout

The human embryonic kidney cell line HEK293T was obtained from the American Type Culture Collection (ATCC) and cultured following the manufacturer’s instructions. The immortalized BMDM (iBMDM) cell line was a kind gift from Prof. Kai Zhao (Xiangya Hospital, Central South University, China), which was described previously [82]. To generate Acly KO iBMDM cells, CRISPR/Cas9 sgRNAs targeting the mouse Acly gene were designed and inserted into the lentiviral vector U6-sgRNA-SFFV-Cas9-FLAG-P2A-EGFP. The inserted sequences in the vectors were as follows: CON359: 5’-CGC TTCCGCGGCCCGTTCAA-3’; Acly sgRNA1: 5’-AGAACCGGTTCAAGTACGCC-3’; and Acly sgRNA2: 5’-TTCCTCGACGTTTGATTAAC-3’. The knockout efficiency was confirmed by immunoblotting and Q-PCR analysis.

siRNA and ASO transfection

siRNAs targeting Rbm25 (siGENOME SMART pool siRNA, M-057045-01) and nontargeting siRNAs (siGENOME nontargeting siRNA Pool #1) were purchased from Dharmacon. The siRNAs targeting Yy1 (sc-36864) and the control siRNA (sc-37007) were purchased from Santa Cruz Biotechnology. The siRNAs targeting the 12th exon, 14th exon and 17th exon of mouse Acly were designed and synthesized by GenePharma (China) and are as follows: 12th exon-sense: 5’-CAGCUCACACUGCCAACUUTT-3’; 14th exon-sense: 5’-AUUCAGUCCCAAGUCCAAGTT-3’; 17th exon-sense: 5’-UCACACGGAAGCUCAUCAATT-3’. Negative control (NC)-sense: 5’-UUCUCC GAACGUGUCACGUTT-3’. The macrophages were transfected with the siRNAs via Lipofectamine RNAiMAX (Thermo Fisher Scientific) according to the manufacturer’s instructions. The ASOs targeting mouse Acly (esASO-m-Acly) and the control ASO were designed and synthesized by RIBOBIO (China), and the targeting sequence was 5’-GTCCCAAGTCCAAGATCC-3’. ASOs were delivered into macrophages via riboFECT CP Reagent (RIBOBIO, China) or delivered in vivo in saline (15 mg/kg mice) through i.v. injection following the manufacturer’s instructions.

Flow cytometry and antibodies

Single-cell suspensions were prepared from the indicated tissues or peritoneal cavity and subjected to FACS using a SONY ID7000. FACS data were analyzed with FlowJo software. After incubation with anti-CD16/CD32 to block nonspecific antibody binding, the cells were stained with fluorochrome-labeled antibodies. Intracellular FOXP3 staining was performed using a FOXP3/Transcription Factor Staining Buffer Set (Invitrogen) without cell stimulation. The anti-mouse CD45-BV605 (2D1), anti-mouse CD4-BV421 (GK1.5), anti-mouse CD8-BV510 (53-6.7), anti-CD19-APC (6D5), anti-mouse CD44-APC (IM7), anti-mouse CD62L-PE (MEL-14), anti-mouse CD11c-APC (N418), and anti-mouse Ly6G/Ly6C-PE (Gr-1) (RB6-8C5) antibodies were from Biolegend; the anti-FOXP3-PE (FJK-16s) antibody was from eBioscience.

Immunoblot and coimmunoprecipitation assays

Immunoblot and coimmunoprecipitation (co-IP) analyses were performed as described previously [85]. Briefly, total proteins from cells were extracted with cell lysis buffer (Cell Signaling Technology) supplemented with a protease inhibitor cocktail and PMSF (Merck). The protein concentration was measured with a BCA protein assay kit (Thermo Fisher Scientific) and subjected to immunoblot assays via a Bio-Rad electrophoresis system.

Protein lactylation modification mass spectrum

Acly KO iBMDMs were infected with lentivirus-Flag-Acly L or lentivirus-Flag-Acly S and stimulated with LPS for protein extraction. SDS‒PAGE was performed, followed by Coomassie blue staining, and the bands were excised and subjected to detection. A mass spectrometer (LTQ Orbitrap Elite, Thermo Fisher Scientific) was used to identify lactylation of Acly isoforms by PTM Bio (China), as described previously [86].

RNA extraction and quantitative RT‒PCR analysis

Total RNA was extracted from cells via TRIzol reagent (Thermo Fisher Scientific) according to the manufacturer’s instructions. Q-PCR assays were performed using a LightCycler 480 (Roche) and SYBR RT‒PCR kit (Takara). The primers used in the quantitative RT‒PCR assays are listed in Table S2.

micro-CT analysis

The front paws and wrist joints were harvested and scanned via micro-CT (Skyscan1176, Bruker) with the following parameter settings: source voltage, 50 kV; source current, 450 μA; AI 0.5 mm filter; pixel size, 9 μm; rotation step, 0.4 degrees. The obtained images were reconstructed with NRecon software via the following parameter settings: ring artifact correction, 8%; smoothing, 2%; and beam hardening correction, 30%. A refined volume of interest was generated 0.45 mm above or under the vertebral/growth plate of the distal vertebral body/femur or proximal tibia and 1.8 mm in height. The region of interest (ROI) for trabecular bone within the above volume was manually defined. The constant threshold was set as 80--255; then, the bone volume per tissue volume (BV/TV), trabecular number (Tb. N) and trabecular thickness (Tb. Th) were analyzed via the program CTAn (Bruker microCT, Kontich, Belgium).

scRNA-seq analysis

Mouse CIA synovial tissue datasets [19] and RA patient PBMC datasets [25] were acquired from the GEO database under accession No. GSE192504 and GSE159117. All gene‒barcode matrices were loaded as Seurat objects in R software (v.4.1.1) via the Seurat package [87]. For quality control, filtering was conducted with criteria for CIA synovial cells (gene number between 200 and 7000, UMI count > 1000 and mitochondrial gene percentage < 0.05) and RA PBMCs (gene number between 200 and 6000, UMI count > 1000 and mitochondrial gene percentage < 0.1). Data from each sample were normalized via the NormalizeData() function, and variable features were identified via FindVariableFeatures() with 2000 genes, setting “vst” (variance stabilizing transformation) as the selection method. On the basis of the results from the analysis via JackStraw and elbow plots, clustering was conducted via the FindNeighbors() and FindClusters() functions via 20 principal component analysis components, and the resolution parameters were set to cluster at a resolution of 0.3 in CIA synovial tissues and 1.125 in RA PBMCs. The resulting clusters were visualized in a 2D UMAP representation and annotated with certain cell types. The subset function was subsequently used for further analysis of the cell types of interest.

For CIA synovial tissue data analysis, the visualization of transcripts highly enriched in specific cell clusters that provided an overview of annotated cell phenotypes via the DotPlot() and VlnPlot() functions in Seurat was used to visualize selective gene expression. Four hundred ninety-three macrophages from CIA mice and 1290 monocytes from RA patients were extracted for downstream analysis, with the top 25% of cells defined as RBM25^high cells on the basis of the normalized expression level of RBM25 in each cell in each set of scRNA-seq data. The inflammatory score was calculated from the mean expression of seventeen selected inflammatory genes, namely, Il1a, Il1b, Il6, Il12a, Il23a, Tnf, Ccl3, Ccl4, Cxcl13, Tlr4, Dusp1, Map3k7, Stat1, Hif1a, Acod1, Lyz1, and Nlrp3, which are widely used to profile the characteristics of macrophages or monocytes in diseases, especially RA [88]. On the basis of the inflammatory score, infla^high cells were identified as the top 25% of inflammatory cells from CIA synovial tissues and RA PBMCs. The common cells in the Venn diagram were renamed to compare two distinct RBM25^high and infla^high populations in each dataset.

RNA sequencing and alternative splicing analysis

Total RNA was isolated via an RNeasy mini kit (Qiagen, Germany). Strand-specific libraries were prepared via the TruSeq Stranded Total RNA Sample Preparation Kit (Illumina, USA) following the manufacturer’s instructions. Library construction was performed following the steps of mRNA isolation, fragmentation, first-strand cDNA synthesis, second-strand cDNA synthesis, end repair, adding A at the 3’ end, connecting connectors, and enrichment, using the VAHTS Universal V6 RNA-seq Library Prep Kit for Illumina (Vazyme), AHTS mRNA capture beads (Vazyme), VAHTS DNA Clean Beads (Vazyme), the Qubit dsDNA HS Assay Kit (Invitrogen) and the Agilent High Sensitivity DNA Kit (Agilent). Alternative splicing analysis was performed via rMATS software (http://rnaseq-mats.sourceforge.net/index.html).

ChIP and RIP

For the ChIP assays, 1 × 10⁶–10⁷ BMDMs were stimulated and crosslinked with 1% (v/v) methanol-free formaldehyde for 10 min, terminated with glycine solution, and then subjected to ChIP assays via a ChIP assay kit (Millipore) according to the manufacturer’s instructions. Purified DNA was detected by Q-PCR and then normalized to the input DNA for each sample. RIP-seq was performed via a Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Millipore) according to the product instructions. BMDMs (5 × 10⁶–1 × 10⁷ cells) were harvested and resuspended in RIP lysis buffer. RBM25-binding RNA was immunoprecipitated with an antibody against RBM25 (Bethyl) and then purified for use. The primers used for amplification and quantification via ChIP‒qPCR or RIP‒qPCR are listed in Table S3.

Plasmid constructs and transfection

The mouse Rbm25 gene was obtained from mouse macrophages via RT‒PCR and subsequently cloned and inserted into the pcDNA 3.1 vector, as described previously [85]. pECMV-3×Flag and pEnCMV-3×HA plasmids expressing mouse Acly L or Acly S were purchased from Miaoling Biology (Wuhan, China). Truncated mutants and point mutants were generated via PCR-based amplification, and the construct encoding WT Acly was used as the template. Each construct was confirmed by sequencing. The corresponding primers are listed in Table S4. The TIP60, MYST1, GCN5, PCAF, P300 and CBP plasmids were kindly provided by Prof. Di Wang (Zhejiang University School of Medicine).

CUT&Tag assay and data analysis

CUT&Tag was performed via a Hyperactive Universal CUT&Tag assay kit for Illumina (Vazyme, China) according to the product instructions. Briefly, BMDMs were stimulated with LPS and then harvested after washing with PBS. The cells (1 × 10⁵) were bound to concanavalin A-coated magnetic beads and then incubated with antibodies against H3K9ac or H3K27ac at 4 °C overnight. The next day, the samples were incubated with secondary antibodies and then incubated with Hyperactive pA/G-Transposon for 1 h. The DNA was treated with TTBL at 37 °C for 1 h and extracted, followed by purification for library amplification. The next-generation sequencing technique was applied to sequence the DNA. Genome mapping was performed via Bowtie (version: 0.12.8), and the clean reads were compared with the mm10 reference genome. MACS (version 2.2.7.1) was applied to call peaks. The bam file and R package DiffBind were used to analyze the differences between the peaks of the two specified samples. The default parameters were set as a fold change ≥ 2 or ≤ 1/2 and a P value < 0.05, and the differences between the obtained peaks were annotated. The annotation program was the same as the above peak annotation program and included different peak location information, annotation information, etc. Motif analysis was conducted via the “HOMER” tool. The result of the HOMER motif was the motif information predicted on the basis of the existing database within the peak.

Luciferase reporter assay

HEK293T cells or macrophages were cotransfected with a mixture of the indicated luciferase reporter plasmids, the pRL-TK-Renukka-luciferase plasmid, or the indicated plasmids. After 24–36 h, luciferase activities were measured via the Dual Luciferase Reporter Assay System (Promega) according to the manufacturer’s instructions. The data were normalized for transfection efficiency by dividing firefly luciferase activity by Renilla luciferase activity.

SPR assay

The SPR assay was performed as previously described [89]. Briefly, a Biacore T200 system (GE Healthcare, Uppsala, Sweden) was used, with recombinant Acly L and Acly S proteins immobilized on CM5 sensor chips via an amine coupling procedure. Binding specificity was assessed in HBS-EP+ at a constant flow rate of 30 mL/min. ATP and citrate (2, 4, 8, 16, 32, 64, and 128 μM) were used as ligands, and HBS-EP+ was used as a negative control to confirm the suitability of the SPR system. The response signals were recorded and analyzed.

Metabolic assays

BMDMs were prepared and uniformly plated into XF96 plates (102601-100, Agilent) and stimulated with or without LPS. A Seahorse XF96 analyzer (Agilent) and Wave 2.6.1 software (Agilent) were used to conduct the experiments. The OCR was measured both basally and in response to sequential additions of 1.5 mM oligomycin, 2.0 µM FCCP and 0.5 µM rotenone plus antimycin A in XF medium (DMEM containing 10 mM glucose, 2 mM glutamine, and 1 mM sodium pyruvate). For ECAR measurement, the following chemicals were added sequentially: 25 mM glucose, 1.5 mM oligomycin, and 50 mM 2-DG after the baseline measurements. After these measurements, the Seahorse XF data were normalized to the total protein for each well.

Metabolite profiling analysis

BMDMs (1 × 10⁷ cells) were used to perform a metabolomics assay at Shanghai Biotree Biotech Co. (China). Briefly, for each sample, 500 μl of precooled MeOH/H2O (3/1, v/v) was added. The samples were then vortexed for 30 sec. After being precooled on dry ice, the samples were frozen and thawed three times in liquid nitrogen. The samples were then vortexed for 30 sec and sonicated for 15 min in an ice–water bath. Next, as a subsequent step, the samples were incubated at −40 °C for 1 h and centrifuged at 12000 rpm for 15 min at 4 °C. Afterwards, an aliquot of 400 μl of clear supernatant was collected and dried by spinning. As a reconstitution solution, 200 ml of ultrapure water was mixed with the residue. The reconstituted samples were vortexed before passing through the filter of the centrifuge tube, after which they were transferred to inserts in injection vials for HPLC-MS/MS analysis.

Statistical analyses

The statistical significance of differences between the two groups was determined by unpaired Student’s t test. For comparisons of more than two groups, one- or two-way ANOVA was performed. P values of less than 0.05 were considered statistically significant. The statistical tests were justified as appropriate according to the assessment of normality and variance of the distribution of the data. No randomization or exclusion of data points was used.

Author contributions

Conceptualization: XL, ZZ, and YZ. Methodology: YZ, YG, YW, YJ, YX, BC, SX, and FM; investigation: YZ, YG, YW, YJ, YX, XW, ZW, YD, HC, BR, WH, and BC; visualization: YZ, YG, YJ, and YX; resources: SX, FM, and XR; funding acquisition: XL and ZZ; supervision: XL and ZZ; writing–original draft: YZ, XL, and ZZ; writing–review and editing: YZ, XL, and ZZ.

Funding

This work was supported by the National Key Research & Development Program of China (2023YFC2307302, 2019YFA0801502, 2023YFC2307001, 2023YFC2307002), the National Natural Science Foundation of China (82071790, 82070415, 82271797, 82341065, 82371825, 32400727, 82201955), the program of Shanghai outstanding academic leader in public health subject (GWVI-11.2-XD29), and the experimental animal program sponsored by the Science and Technology Commission of Shanghai Municipality (23141902300).

Data availability

RNA sequencing, CUT&Tag sequencing and RIP sequencing raw data were deposited in the NCBI Gene Expression Omnibus database under the accession numbers GSE240157, GSE240158, GSE272304, GSE240159, and GSE240188, respectively.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41423-024-01212-3.