Macrophage trained immunity requires a mitochondrial glutamine-restriction checkpoint for antitumor protection in mice Authors: Ya Chun Yu, Yulseung Sung, Seonghun Lim, Jeong Ah Kwon, Min Ju Lee, Hyeong Yun Kim, Sunghoon Kim, Kuglae Kim, Do Sik Min, Hee Chan Yoo, Jung Min Han Published: 2026 ABSTRACT AbstractTrained immunity enables innate immune cells to acquire memory-like responses, offering a strategy to enhance antitumor immunity. However, the metabolic‒epigenetic mechanisms underlying this process remain poorly defined. Here, we show that lipopolysaccharide-induced macrophage training is encoded by a mitochondrial metabolic checkpoint. Integrated transcriptomic, metabolomic, and epigenomic profiling reveals that TLR4-NF-κB signaling represses SLC1A5_var, a mitochondrial glutamine transporter, limiting glutaminolysis and reducing α-ketoglutarate availability. This metabolic restriction limits removal of the activating histone mark histone H3 lysine 4 trimethylation by KDM5B, thereby maintaining inflammatory gene accessibility. Functionally, pharmacological inhibition or myeloid-specific knockdown of SLC1A5_var potentiates macrophage training and improves tumor control in murine cancer models, whereas enforced SLC1A5_var expression or α-ketoglutarate supplementation abrogates these effects. These findings define an SLC1A5_var-α-ketoglutarate-KDM5B metabolic-epigenetic axis that programs macrophage trained immunity and illustrate how targeted metabolic restriction can be leveraged to enhance innate immune responses against cancer.Subject terms: Metabolomics, Innate immunity, Tumour immunology, Monocytes and macrophages Introduction Despite major advances in cancer immunotherapy, the immunosuppressive tumor microenvironment (TME) continues to limit durable clinical responses1–3. Myeloid cells have a central role in this limitation, as they can either support immune surveillance or promote tumor progression4. This duality has motivated efforts to reprogram macrophages toward stable and potent antitumor effectors capable of resisting tumor-derived inhibitory cues5–7. Trained immunity offers a conceptual and mechanistic framework for such reprogramming. Through epigenetic and metabolic reprogramming, innate immune cells can acquire memory-like responsiveness that enhances host defense8–11. Although originally described in infection, trained immunity has recently been implicated in antitumor responses12–16, particularly following β-glucan or BCG vaccination, which enhances myeloid-dependent tumor suppression17–21. These observations raise the possibility that manipulating innate immune memory could improve the durability of antitumor immunity. Beyond β-glucan and BCG vaccination, recent studies indicate that TLR agonists, including lipopolysaccharide (LPS), can also induce innate immune memory22–25. Whether such LPS-driven training confers antitumor effects, however, remains unclear. Notably, this concept is reminiscent of Coley’s early use of bacterial components to elicit host antitumor responses, in which LPS was considered the major active component26–29. Despite these historical observations, whether LPS can elicit trained immunity in myeloid cells with functional antitumor consequences, and the molecular mechanism underlying this process, remains unsolved. Metabolic rewiring is fundamental to trained immunity, with prior work emphasizing enhanced glycolysis and the accumulation of immunoregulatory metabolites such as fumarate, itaconate, and lactate30–33, that remodel chromatin and enhance epigenetic memory formation34–36. However, the contribution of other metabolic circuits, particularly amino acid metabolism, to innate immune memory remain poorly defined. Glutamine metabolism is of particular relevance due to its dual role in bioenergetics and epigenetic control37,38. Mitochondrial glutaminolysis couples anaplerosis with epigenetic regulation through α-ketoglutarate (α-KG), a cofactor for chromatin-modifying enzymes39–42. Thus, alterations in mitochondrial glutamine metabolic flux may profoundly influence the establishment and persistence of trained immunity. We previously identified a transcriptional variant of SLC1A5, named SLC1A5_var, arising from an alternative transcription start site, which localizes to mitochondria and mediates glutamine import to support glutaminolysis43. Whether SLC1A5_var plays a role in innate immune memory has not been explored. Here, we aimed to elucidate how LPS-driven trained immunity contributes to antitumor immunity and to define the underlying metabolic–epigenetic mechanisms. We show that TLR4-NF-κB signaling represses SLC1A5_var transcription, thereby limiting mitochondrial glutaminolysis and reducing α-KG availability. This restriction constrains KDM5B-mediated H3K4me3 demethylation, stabilizing inflammatory chromatin states and conferring durable antitumor function. Moreover, myeloid-specific SLC1A5_var knockdown or pharmacological inhibition amplifies trained immunity and enhances tumor suppression. Together, our findings identify SLC1A5_var as a metabolic checkpoint linking glutamine metabolism to epigenetic memory, providing a mechanistic foundation for targeting metabolic restriction to reprogram innate immunity in cancer. Results Prior exposure to LPS induces macrophage-trained immunity against cancer To determine whether LPS-induced trained immunity contributes to antitumor defense and to determine the duration of this effect, we first evaluated LPS administration prior to subcutaneous (s.c.) tumor xenograft challenge. Wild-type (WT) C57BL/6 mice received a single intraperitoneal (i.p.) injection of either endotoxin-free saline (“naïve”) or low-dose LPS (50 μg kg−1; hereafter “LPS-trained”) 7, 14, or 21 days prior to s.c. inoculation with MC38 colon carcinoma cells (Fig. 1a). LPS-trained mice exhibited significantly reduced tumor volume (Fig. 1b) and weight (Fig. 1c) compared with the control naïve group, indicating that LPS-induced training exerts tumor-suppressive effects. However, mice in the D+21 group showed markedly greater tumor growth than mice in the D+7 and D+14 groups, indicating that LPS-induced antitumor immunity persists for several weeks, with maximal efficacy observed within two weeks. To exclude residual systemic inflammation as a confounding factor, serum cytokine levels of mice were measured at multiple time points after low-dose LPS injection (Supplementary Fig. S1a). LPS triggered acute but transient increases in TNF-α, IL-1β, and IL-6 levels, which returned to baseline levels within 24 h. These results indicate that the antitumor effect of LPS training is not attributable to residual inflammation from initial exposure to low-dose LPS. Taking these results into consideration, we adopted a 1-week LPS training protocol for all subsequent in vivo experiments (Supplementary Fig. 1b). LPS-trained macrophages exhibit enhanced antitumor activity. a–c C57BL/6 mice intraperitoneally (i.p.) injected with PBS or LPS, 7, 14, or 21 days before subcutaneous (s.c.) MC38 inoculation (a). Tumor volume (b) and tumor weight (c) (n = 6 mice). One-way ANOVA test. Exact p-values (p < 0.0001): b 4 × 10−8, 2 × 10−7, 1 × 10−6; c 4 × 10−8, 3 × 10−7, 5 × 10−5. d Frequencies of monocytes, neutrophils and macrophages in tumor-infiltrating myeloid population (n = 6). See Supplementary Fig. 1e for representative plots. e Ratio of tumor-infiltrated CD86+/CD206+ macrophages (p = 1 × 10−5) (n = 6). f Tumor-infiltrated macrophages intracellularly stained for indicated cytokines (p = 9 × 10−6) (n = 6). LPS training in Rag1−/− mice (n = 5) (g), tumor volume (h) and tumor weight (i). Exact p-values (p < 0.0001): h 3 × 10−5; i 8 × 10−6. Macrophage depletion with clodronate (Clo) liposome in C57BL/6 mice (n = 5) (j), tumor volume (k) and tumor weight (l). m–p Ex vivo restimulation experiments. MΦ, macrophage (m). IL-6 secretion in spleen (n), lymph nodes (o) (p = 3 × 10−5), macrophages and bone marrow-derived macrophages (BMDMs) (p) (p = 6 × 10−5) (n = 6 mice). q Phagocytosis of MC38 by in vivo-trained splenic macrophages (n = 6 mice). r–t in vitro LPS training protocol (r). IL-6 (s) and IL-1β (t) secretion (n = 6 biological replicates). Exact p-values (p < 0.0001): s 2 × 10−7, 3 × 10−7, 1 × 10−8; t 2 × 10−7, 1 × 10−7, 6 × 10−7. u LEGENDplex analysis showing mean absolute cytokine levels and log2 fold change (Log2FC) (n = 4 biological replicates). v Phagocytosis of MC38 by BMDMs (n = 6 biological replicates) (p = 7 ×10−5). w–y BMDMs were adoptively transferred with MC38 into the recipient (w). Tumor volume (x) and tumor weight (y) (n = 7 mice). The data are presented as the mean ± SEM. For (n–p, s–t), box plots show median (center), 25th and 75th percentiles (box boundaries), and whiskers extending to the minima to maxima. Significance was assessed using a two-tailed Student’s t-test unless otherwise specified. All scale bars, 1 cm. Source data are provided as a Source Data file. (a, g, j, m, r, and w) were created with BioRender (Created in BioRender. https://BioRender.com/n400kl0). As tumor-infiltrating immune cells are key determinants of antitumor responses, we next tried to examine how LPS training reshapes the tumor microenvironment (TME). MC38 cells were inoculated into naïve and LPS-trained mice, and the reduced tumor burden in trained mice was again confirmed (Supplementary Fig. 1c, d). Flow cytometric analysis of tumor-infiltrating leukocytes revealed increased infiltration of monocytes and macrophages (Fig. 1d and Supplementary Fig. 1e), with elevated ratios of CD86+ antitumor to CD206+ protumor macrophages (Fig. 1e and Supplementary Fig. 1f). Notably, macrophages within LPS-trained tumors displayed markedly increased expression of IL-6, IL-1β, and TNF-α (Fig. 1f). In addition, CD8+ T-cell infiltration was elevated (Supplementary Fig. 1g), and cytotoxic T lymphocytes (CTLs) expressed higher levels of IFN-γ and granzyme B (Supplementary Fig. 1h, i). Collectively, these results indicate that LPS training reshapes the tumor microenvironment toward a proinflammatory, antitumor state. Given that the emergency myelopoiesis is a well-established hallmark of central trained immunity44–46, we next examined whether LPS-induced training may also elicit similar hematopoietic adaptations in the bone marrow (Supplementary Fig. 1j). LPS-trained mice showed expansion of Lin−Sca-1+c-kit+ (LSK) hematopoietic progenitor stem cells (HPSCs) (Supplementary Fig. 1k). We also observed expansion in multipotent progenitors (MPPs), which was primarily attributable to an increase in the size of the myeloid-based MPP3 subset, whereas the lymphoid-biased MPP4 and erythroid/megakaryocytic-biased MPP2 subsets remained unchanged (Supplementary Fig. 1l, m). Consistently, the number of CD11b+ myeloid cells was elevated in the spleen and lymph nodes of LPS-trained mice (Supplementary Fig. 1n, o), and this increase was accompanied by an increased number of inflammatory monocytes in the spleen (Supplementary Fig. 1p). These findings indicate that LPS training triggers systemic myelopoietic remodeling characteristic of innate immune training, linking the observed antitumor effects to innate immune reprogramming. To determine whether the effects of LPS training were independent of adaptive immune memory, we injected Rag1−/− mice that lack functional B and T cells with LPS, followed by inoculation with MC38 cells (Fig. 1, g). Tumor suppression was preserved in LPS-trained Rag1−/− mice (Fig. 1h, i), indicating that the antitumor effects of LPS-induced training can occur independently of adaptive immunity. To further identify the key effector population, macrophages were depleted using clodronate liposomes (Supplementary Fig. 1q, r), which abrogated the training effect and restored tumor growth to control levels (Fig. 1j–l). Given that clodronate liposomes primarily deplete macrophages while also affecting other phagocytic populations, these results implicate phagocytic cells in general, with macrophages likely representing a major contributing population. To further confirm that LPS-trained macrophages acquire durable innate immune memory, we performed ex vivo rechallenge assays. LPS-trained macrophages isolated from spleen, lymph nodes, and bone marrow-derived macrophages (BMDMs) secreted higher levels of IL-6 upon coculture with MC38 cells (Fig. 1m–p). Also, splenic macrophages from LPS-trained mice exhibited enhanced phagocytosis of CFSE-labeled MC38 cells (Fig. 1q). We also reassessed duration-dependent training in an independent experiment using ex vivo rechallenge assays. Consistent with in vivo xenograft results (Fig. 1a), splenic macrophages of LPS-trained mice demonstrated increased cytokine secretion and phagocytosis compared to control mice (Supplementary Fig. 1s, t). However, splenic macrophages from the D+21 group displayed diminished responses relative to those from the D+7 and D+14 groups (Supplementary Fig. 1s, t). These results indicate that LPS elicits a systemic increase in trained macrophages and that the functional potency of trained macrophages correlates with the persistence of innate immune memory. To evaluate whether LPS-induced training could be recapitulated in vitro, BMDMs were exposed to low-dose LPS (2.5 ng ml−1) overnight, washed, and allowed to rest for up to six days prior to functional assays (Fig. 1r). In response to restimulation by MC38 cells, Lewis lung carcinoma (LLC) cells, or the TLR2 agonist Pam3CSK4 (P3C), LPS-trained BMDMs exhibited enhanced effector functions, including increased secretion of cytokines including IL-1β and IL-6 relative to that in naïve controls (Fig. 1s–u), as well as increased phagocytosis of cancer cells (Fig. 1v and Supplementary Fig. 1u). These findings indicate that the heightened responsiveness of LPS-trained macrophages is not limited to a single cancer cell line and that trained immunity can also be elicited by pattern-associated molecular patterns (PAMPs). Together, these results demonstrate that macrophage training is cell-intrinsic and can be reproduced in vitro. To determine whether LPS-trained macrophages are functionally sufficient to mediate tumor suppression, we performed adoptive transfer experiments. Naïve or in vitro-trained BMDMs were co-inoculated with MC38 cells into untreated recipient mice (Fig. 1w). Mice that received trained macrophages exhibited significantly reduced tumor growth (Fig. 1x,y). Notably, tumors from trained BMDM recipients displayed an increased CD86+/CD206+ macrophage ratio and enhanced CD8+ T-cell infiltration, and also the expansion of IFN-γ+ and granzyme B+ CTLs (Supplementary Fig. 1v–z). Together, these findings identify macrophages as key effectors that mediate LPS-induced trained immunity and underscore their central role in durable innate immune memory with antitumor potential. Transcriptomic and metabolomic profiling corroborates the suppression of glutaminolysis and glycolytic activation during trained immunity As metabolic reprogramming is closely linked to the induction of trained innate immunity47, we next aimed to investigate how LPS training alters macrophage metabolism by performing RNA-sequencing (RNA-seq) of naïve versus trained BMDMs. Gene set enrichment analysis (GSEA) of Gene Ontology (GO) terms using differentially expressed genes (DEGs) revealed a sustained shift in metabolic pathways, characterized by decreased glutamine/glutamate metabolism and tricarboxylic acid (TCA) cycle activity, and increased glycolysis-related pathways (Fig. 2a, b and Supplementary Fig. 2a). Accordingly, we observed suppressed expression of glutaminolysis-associated genes (Gls, Glud1, and Gpt2) and induced expression of glycolysis-related genes (Hk2, Pkm, and Gpi1) in LPS-trained BMDMs (Fig. 2a). Notably, chromatin remodeling pathways were also enriched (Fig. 2b), with upregulation of regulators such as Ep300 and Hdac4 (Fig. 2a), implying that metabolic changes may be accompanied by epigenetic reprogramming—another hallmark feature of trained immunity. LPS training suppresses mitochondrial glutaminolysis in macrophages. a Heatmap depicting differential gene expression determined by RNA-sequencing in trained versus naïve bone marrow-derived macrophages (BMDMs) (n = 3). b Gene set enrichment analysis (GSEA) comparing trained and naïve BMDMs. Gene Ontology (GO) terms are shown. FDR, false discovery rate; NES, normalized enrichment score. c Immunoblots of indicated proteins. Mean band intensities relative to naïve control normalized to Actin are shown (n = 3 independent biological replicates). d Principal component analysis (PCA) of normalized metabolite data of naïve and trained BMDMs (n = 3). e, f Heatmap depicting significantly altered metabolites (p < 0.05) from LC‒MS analysis (n = 3 independent biological replicates). Glu, glutamate; PPP, pentose phosphate pathway (e). Metabolite pathway enrichment analysis using MetaboAnalyst 6.0 (f). g Energy map of the basal oxygen consumption rate (OCR) vs. basal extracellular acidification rate (ECAR), and the maximal OCR vs. maximal ECAR (n = 6 independent biological replicates). h Relative contributions of mitochondrial ATP (mitoATP) and glycolytic ATP (glycoATP) production (n = 3 independent biological replicates). i Schematic illustration of atom transitions using uniformly labeled 13C-glutamine (U-13C-Gln) or 15N-glutamine (U-15N-Gln) to infer the intracellular metabolic flux of glutamine. j Abundance of U-13C-Gln-derived glutamate (m + 5), α-ketoglutarate (α-KG) (m + 5), succinate (m + 5) and aspartate (m + 4) (n = 3 biological replicates). k Abundance of U-15N-Gln-derived glutamine (m + 2), glutamate (m + 1) (p = 7.5 × 10−7), aspartate (m + 1) (p = 4.7 × 10−5) and alanine (m + 1) (n = 3 biological replicates). l, m Splenic macrophages were sorted from naïve and LPS-trained mice. Immunoblot analysis (l) and Real-time PCR (qPCR) analysis (m) of SLC1A5_var expression (n = 4 mice per group). MΦ, macrophage. n, o Experimental scheme (n). Tritium-labeled glutamine (3H-Gln) uptake analysis using mitochondria isolated from naïve and LPS-trained BMDMs (n = 5 independent biological replicates) (p = 2.5 × 10−5) (o). For (o), box plots show median (center), 25th and 75th percentiles (box boundaries), and whiskers extending to the minima to maxima. The data are presented as the mean ± SEM. Significance was assessed using a two-tailed Student’s t-test. Source data are provided as a Source Data file. n is created with BioRender (Created in BioRender. https://BioRender.com/n400kl0). The reduction in glutamine metabolism—an underexplored feature in the context of trained immunity—prompted further investigation. Validation at the protein level confirmed decreased expression of glutaminolysis-related enzymes, but elevated expression of glycolysis-related enzymes in trained macrophages, supporting a genuine metabolic shift (Fig. 2c). To complement these findings, we performed nontargeted metabolomic profiling of intracellular metabolites. Principal component analysis (PCA) revealed clear separation of metabolic profiles between naïve and LPS-trained macrophages (Fig. 2d). Metabolite-level analysis further demonstrated reduced α-KG, succinate, and fumarate and increased glucose, pyruvate, lactate, glyceraldehyde-3-phosphate (G3P), and ribose-5-phosphate (R5P) levels in trained macrophages (Fig. 2e). Pathway enrichment analysis using MetaboAnalyst 6.0 highlighted alterations in glutamine metabolism, TCA cycle, glycolysis and the pentose phosphate pathway (PPP) activity, in agreement with the RNA-seq data (Fig. 2f). Collectively, these multi-omics data consistently point to a marked suppression of glutamine catabolism as a characteristic metabolic feature of LPS-trained macrophages, accompanied by enhanced glycolytic activity. To assess the functional consequences of these metabolic shifts, we performed Seahorse real-time metabolic analysis. LPS-trained BMDMs exhibited a pronounced shift toward glycolysis with a reduced reliance on aerobic respiration (Fig. 2g). More specifically, there was a substantial reduction in basal and maximal respiration and a moderate increase in basal glycolysis and glycolytic capacity (Supplementary Fig. 2b, c). This reduction in mitochondrial respiration was also observed in splenic macrophages isolated from naïve and in vivo-trained mice (Supplementary Fig. 2d, e). Consistently, analysis of ATP production rates revealed a reversal in energetic dominance, with the level of glycolysis-derived ATP (gylcoATP) surpassing that of OXPHOS-derived ATP (mitoATP), indicating compensatory reliance on glycolysis (Fig. 2h). Reduced OXPHOS activity in LPS-trained BMDMs was further confirmed by TMRE staining, which showed decreased mitochondrial membrane potential (Supplementary Fig. 2f), and by MitoSOX assay, which revealed diminished mitochondrial superoxide generation, consistent with reduced electron transfer chain (ETC) activity (Supplementary Fig. 2g). Despite of decreased mitochondrial ROS production, total cellular ROS levels showed a modest increase with a minor decrease in the GSH/GSSG ratio were observed (Supplementary Fig. 2h, i). These findings indicate that LPS-trained macrophages undergo metabolic rewiring that lowers mitochondrial respiration. These results prompted us to directly trace nutrient utilization to confirm metabolic reprogramming in LPS-trained macrophages. Given the observed decrease in glutamine metabolism and mitochondrial respiration in LPS-trained macrophages, we traced the fate of glutamine-derived carbon and nitrogen by liquid chromatography‒mass spectrometry (LC‒MS) using uniformly labeled isotopes. U-¹³C-glutamine tracing revealed markedly reduced glutaminolysis and incorporation of carbon into TCA intermediates, reflected by decreased labeling of M + 5 glutamate and α-KG and reduced labeling of M + 4 isotopomers of succinate and aspartate (Fig. 2i, j). Complementary U-¹⁵N-glutamine tracing showed accumulation of glutamine with reduced downstream glutamate levels, together with reduced levels of nitrogen-derived alanine and aspartate, indicating suppressed transamination (Fig. 2k). These results demonstrate impaired glutaminolysis and attenuated mitochondrial anaplerosis, the replenishment of TCA intermediates from glutamine, in LPS-trained macrophages. Given the reduced glutamine metabolism and mitochondrial activity, together with enhanced glycolysis in LPS-trained macrophages, we next directly traced glucose utilization to confirm this metabolic compensation. U-¹³C-glucose tracing revealed increased incorporation of glucose-derived carbon into glycolytic intermediates (G6P, F-1,6-BP), and lactate (Supplementary Fig. 2j), and PPP intermediates such as ribulose-5-phosphate and phosphoribosyl pyrophosphate (PRPP) were increased in trained macrophages (Supplementary Fig. 2k). Consistently, intracellular glucose levels were elevated in trained macrophages (Supplementary Fig. 2l), indicating enhanced glucose uptake. Importantly, PPP activation not only promotes nucleotide biosynthesis but also induces the generation of reducing cofactors critical for redox homeostasis. Indeed, LPS-trained macrophages displayed elevated NADPH levels, consistent with the increase in PPP flux and enhancement of cytoprotective metabolic adaptation (Supplementary Fig. 2m). Together, these multi-omics analyses demonstrate that LPS-trained macrophages undergo coordinated metabolic rewiring, characterized by suppressed glutaminolysis and mitochondrial catabolism and compensatory increase in glycolysis and PPP activity. We next investigated the mechanism underlying the decrease in glutamine metabolism in LPS-trained macrophages. Analysis of transcriptomic and protein expression profiles revealed the suppression of genes involved in glutamine transport and utilization (Fig. 2a–c). Notably, the expression of the mitochondrial glutamine transporter SLC1A5_var was markedly downregulated in LPS-trained macrophages, a finding recapitulated in splenic macrophages from in vivo-trained mice (Fig. 2l,m) as well as in vitro-trained BMDMs and mouse macrophage cell line Raw264.7 and PMA-differentiated human monocytic cell line THP-1 (Supplementary Fig. 2n). In our previous studies, we identified SLC1A5_var as a novel mitochondrial glutamine transporter with a critical role in cancer metabolism37,43, although its function in innate immunity has not been explored. Because SLC1A5_var regulates glutamine supply into mitochondria—the primary site for glutamine-to-glutamate conversion—we aimed to determine whether the reduction in glutaminolysis in trained macrophages is correlated with impaired mitochondrial glutamine import. A ³H-glutamine uptake assay was performed using isolated mitochondria, and it was found that mitochondrial glutamine import was significantly decreased in LPS-trained BMDMs (Fig. 2n, o). These findings indicate that the metabolic suppression observed in trained macrophages could originate at the level of substrate entry, with LPS training reducing SLC1A5_var-mediated mitochondrial glutamine import and thereby limiting glutaminolysis. NF-κB-mediated repression of SLC1A5_var drives metabolic reprogramming in LPS-trained macrophages To elucidate the mechanism underlying suppression of glutamine metabolism in LPS-trained macrophages, we next investigated the upstream regulatory mechanism by which LPS suppresses SLC1A5_var expression. Given that TLR4 mediates LPS signaling in macrophages, we compared BMDMs isolated from TLR4 WT and TLR4LPS−D mice, which are unresponsive to LPS due to a mutation in TLR448,49, to determine whether SLC1A5_var repression is directly dependent on TLR4 activation. A time-course experiment revealed a progressive reduction in SLC1A5_var expression and mitochondrial glutamine uptake following LPS exposure (Supplementary Fig. 3a). In contrast, TLR4LPS−D BMDMs showed no changes in SLC1A5_var expression or glutamine uptake and failed to mount trained responses upon MC38 restimulation, as reflected by a lack of changes in cytokine production (Supplementary Fig. 3a–d). Similar results were observed when TLR4 signaling was pharmacologically blocked with TAK-242 (Supplementary Fig. 3e, f), which also prevented the LPS-training-induced reduction in the OCR and increase in the ECAR (Supplementary Fig. 3g, h). Together, these findings demonstrate that TLR4 signaling is required for LPS-training-dependent suppression of glutaminolysis through transcriptional repression of SLC1A5_var. To investigate the upstream molecular mechanism driving SLC1A5_var suppression downstream of TLR4 signaling, we performed in silico motif analysis (ALGGEN-PROMO.v8.3), which revealed a putative NF-κB binding sequence upstream of the SLC1A5_var promoter (Fig. 3a). Given that NF-κB is a key downstream effector of TLR4 signaling upon LPS stimulation in macrophages, we next examined whether NF-κB directly regulates SLC1A5_var expression. To test the functional relevance of this binding sequence, we performed dual-luciferase reporter assays using pGL4.1 empty vector (EV), pGL4.1 containing the promoter region with NF-κB binding site (pVar), or a shorter construct lacking NF-κB binding site (pVar_s). While LPS exposure suppressed luciferase activity in pVar-transfected cells, the luciferase activity of pVar_s-transfected cells remained unchanged, indicating the NF-κB-dependent transcriptional control of SLC1A5_var (Fig. 3a). Consistently, LPS treatment or overexpression of the NF-κB subunits p50 and p65 downregulated SLC1A5_var protein and mRNA expression (Fig. 3b and Supplementary Fig. 3i), and reduced luciferase reporter activity (Fig. 3c), whereas overexpression of STAT1 had no effect, indicating that NF-κB activation mediates the transcriptional repression of SLC1A5_var in response to LPS. Mutation or deletion of the NF-κB binding site abolished this repression upon LPS stimulation or p50 overexpression (Fig. 3d). Furthermore, pretreatment with the NF-κB inhibitor JSH-23 rescued SLC1A5_var expression following LPS training (Fig. 3e, f). Chromatin immunoprecipitation (ChIP) assays further confirmed the enrichment of p50 and p65 at the SLC1A5_var promoter after LPS stimulation, which was reversed by JSH-23 pretreatment (Fig. 3g and Supplementary Fig. 3j). Together, these results demonstrate that NF-κB directly represses SLC1A5_var transcription in trained macrophages. NF-κB activation by LPS training suppresses SLC1A5_var-mediated mitochondrial glutamine influx. a Dual-luciferase activity in MC38 cells transfected with the indicated pGL4.1 firefly luciferase vectors (n = 3 independent biological replicates). Exact p-values (p < 0.0001): 2 × 10−5. RE, responsive element. b Immunoblotting analysis of bone marrow-derived macrophages (BMDMs) transfected with indicated expression vectors or empty vector (EV) (n = 3 independent biological replicates). c Dual-luciferase assay of pVar-transfected cells stimulated with LPS or transfected with indicated expression vectors (n = 4 independent biological replicates). One-way ANOVA test. d Dual-luciferase assay with pVar plasmids with mutated or deleted NF-κB-RE (n = 3). qPCR analysis (e) and immunoblot analysis (f) of SLC1A5_var expression in BMDMs pretreated with JSH-23 (n = 3 independent biological replicates). g Chromatin immunoprecipitation (ChIP) analysis of Slc1a5_var promoter in pretreated BMDMs using anti-p65 or anti-p50 antibodies (n = 4 biological replicates). Exact p-values (p < 0.0001): 9 × 10−5, 3 × 10−5, 5 × 10−5. h 3H-glutamine uptake assay using mitochondria isolated from indicated BMDMs (n = 6 biological replicates). Exact p-value (p < 0.0001): 2 × 10−5. Fluctuations in oxygen consumption rate (OCR) (i), basal and maximal respiration (j) of BMDMs (n = 3 independent biological replicates). Rot/AA, rotenone and antimycin A. k IL-1β secretion upon MC38 rechallenge of indicated BMDMs (n = 5 biological replicates). iMQT_020 was added during the resting phase; one-way ANOVA test. Exact p-values (p < 0.0001): 3 × 10−8, 3 × 10−11, 1 × 10−7. l–o BMDMs were transfected with the indicated plasmid two days after LPS training. 3H-glutamine uptake analysis in mitochondria of indicated BMDMs (n = 3 biological replicates) (l), exact p-values (p < 0.0001): 5 × 10−6, 7 × 10−5. Seahorse analysis showing OCR (m), basal and maximal respiration (n) (n = 3 biological replicates). IL-1β secretion upon MC38 rechallenge in indicated BMDMs (n = 5 biological replicates) (o), exact p-values (p < 0.0001): 2 × 10−5, one-way ANOVA test. For (k, o), box plots show median (center), 25th and 75th percentiles (box boundaries), and whiskers extending to the minima to maxima. The data are presented as the mean ± SEM. Significance was assessed using a two-tailed Student’s t-test unless otherwise specified. Source data are provided as a Source Data file. To test whether the NF-κB-dependent suppression of SLC1A5_var is functionally required for macrophage training, we examined the effect of JSH-23 treatment prior to LPS exposure in our in vitro BMDM training models. Suppressed NF-κB signaling by JSH-23 pretreatment rescued mitochondrial glutamine uptake (Fig. 3h) and hence the mitochondrial respiration following LPS training (Fig. 3i and Supplementary Fig. 3k), and this blunted trained immune responses, as evidenced by reduced IL-1β secretion upon MC38 cell restimulation (Fig. 3k). Interestingly, pharmacological inhibition of SLC1A5_var with iMQT_020, a selective SLC1A5_var inhibitor described in our previous study50, potentiated LPS training effect and restored training effect even under JSH-23 pretreatment (Fig. 3k). Given previous reports implicating type I interferon and JAK/STAT signaling in trained immunity25,51, we assessed whether this pathway contributes to trained immunity in our system. However, pretreatment with the JAK1/2 inhibitor ruxolitinib during the priming phase did not alter SLC1A5_var downregulation or trained cytokine responses upon secondary stimulation (Supplementary Fig. 3l–n). Collectively, these findings indicate that NF-κB signaling is a key driver of LPS-induced metabolic and functional reprogramming. Finally, overexpression of SLC1A5_var during the resting phase restored mitochondrial glutamine import and respiratory capacity and blunted the LPS-training effect on IL-1β secretion upon MC38 cell restimulation, phenocopying the effects of TLR4 or NF-κB inhibition (Fig. 3l–o). Thus, the NF-κB-driven repression of SLC1A5_var functions as a central metabolic checkpoint enabling LPS-trained macrophage immunity. We next examined whether the LPS-training-driven NF-κB-SLC1A5_var axis is unique to TLR4 signaling or shared across other TLR pathways. Using the same training-restimulation protocol, we tested P3C (TLR2 agonist), poly(I:C) (TLR3 agonist), and R848 (TLR7/8 agonist) (Supplementary Fig. 3o). P3C- and Poly(I:C)-trained BMDMs displayed increased proinflammatory cytokine production upon MC38 cell restimulation and were associated with reduced SLC1A5_var expression (Supplementary Fig. 3p, q), partially resembling the phenotype of LPS-trained macrophages. In contrast, R848-pretreated BMDMs failed to mount a trained response and did not show SLC1A5_var downregulation. ChIP-qPCR analysis further showed that p65 and p50 were recruited to the Tnfa promoter upon stimulation with all tested TLR agonists. In contrast, recruitment of p65 and p50 to the Slc1a5_var promoter was observed following LPS, P3C and Poly(I:C) stimulation, whereas no significant recruitment was detected following R848 treatment (Supplementary Fig. 3r). These findings suggest that NF-κB recruitment to the Slc1a5_var locus is stimulus-dependent, and that the NF-κB-SLC1A5_var axis may be engaged in a subset of TLR-driven training contexts beyond LPS. SLC1A5_var overexpression attenuates macrophage training against cancer Having established the functional relevance of SLC1A5_var repression in trained macrophage responses, we investigated whether SLC1A5_var overexpression interferes with the establishment of trained immunity. To this end, we generated SLC1A5_var transgenic (Tg) mice (Supplementary Fig. 4a), and the elevated expression of SLC1A5_var was confirmed in BMDMs from the SLC1A5_var Tg mice. Because glutamine metabolism is critical for immune cell development and differentiation52–54, to determine whether SLC1A5_var overexpression affects macrophage differentiation, we examined in vitro differentiation of Tg BMDMs, which was comparable to WT controls (Supplementary Fig. 4b). Additionally, expression of LPS receptor TLR4 (Supplementary Fig. 4c) and downstream NF-κB signaling activity (Supplementary Fig. 4d) were similar between WT and Tg BMDMs. These findings indicate that SLC1A5_var overexpression does not affect baseline macrophage differentiation or activation by LPS, allowing us to directly assess the role of SLC1A5_var in metabolic reprogramming during trained immunity. Subsequently, BMDMs from WT and SLC1A5_var Tg mice were subjected to LPS training in vitro (Fig. 4a). Cytokine gene expression and secretion were markedly increased in LPS-trained WT BMDMs compared with naïve controls upon restimulation by MC38 or P3C, whereas this response was attenuated in Tg BMDMs (Supplementary Fig. 4e–h). Consistently, profiling of cytokines in the culture supernatants following restimulation with MC38 cells revealed robust secretion of CXCL1, TNF-α, IL-1β, IL-6, and IL-1α by WT trained BMDMs, whereas Tg BMDMs exhibited diminished cytokine production despite undergoing identical training (Fig. 4b). Similarly, LPS training enhanced the phagocytic capacity of WT BMDMs for tumor cells, but this effect was blunted in Tg BMDMs (Fig. 4c). Together, these findings indicate that SLC1A5_var overexpression impairs the functional acquisition of macrophage-trained immunity against cancer in vitro. SLC1A5_var-driven glutaminolysis impairs macrophage training against cancer. a Experimental scheme for in vitro training of C57BL/6 wild-type (WT) and SLC1A5_var transgenic (Tg) bone marrow-derived macrophages (BMDMs). b LEGENDplex analysis of WT and Tg BMDMs rechallenged with MC38 cells. The mean absolute cytokine levels (top heatmap) and log2 fold change (Log2FC) in the means of LPS-trained Tg BMDMs versus trained WT BMDMs (bottom heatmap) (n = 4 biological replicates). c Phagocytosis of CFSE-labeled MC38 cells by naïve and trained BMDMs of the indicated genotypes (n = 4 independent biological replicates). Exact p-values (p < 0.0001: 5 × 10−5). d–f Naïve and LPS-trained WT and Tg BMDMs were adoptively transferred to recipient WT mice with MC38 cells at a ratio of 1:3 on day 6 after training (d). Tumor volume over 21 days (e) and tumor weight at the end of the experiment (f) (n = 7 mice per group). s.c., subcutaneous. g Ratio of CD86+ and CD206+ macrophages within tumor-infiltrated macrophage cells. h, i Frequency of CD45+CD3+CD8+ T cells in total digested tumors (h), and quantitative analysis of IFNg+ cytotoxic T lymphocytes (CTLs) (i) (n = 7). j Relative migration of CD8+ T cells cocultured with different groups of BMDMs and MC38 cells (n = 4 independent biological replicates). Exact p-values (p < 0.0001): 2 × 10−5, 1 × 10−5, 1 × 10−5. k–n Experimental scheme (k). WT and Tg BMDMs were trained with LPS, and vehicle (veh) or iMQT_020 (5 μM) was supplemented during the resting phase until the 2nd stimulation with MC38 cells. 3H-glutamine uptake by mitochondria isolated from the indicated group (n = 5 biological replicates) (l). Measurement of IL-6 secretion (m) and phagocytosis of CFSE-labeled MC38 cells (n) by the indicated groups of BMDMs (n = 6 independent biological replicates), one-way ANOVA test. Exact p-values (p < 0.0001): l 1 × 10−9, 2 × 10−11, 4 × 10−10; m 9 × 10−8, 2 × 10−9, 1 × 10−5, 3 × 10−5; n 2 × 10−5, 4 × 10−6. The data are presented as the mean ± SEM. Significance was assessed using a two-tailed Student’s t-test. Source data are provided as a Source Data file. (a, d, j, and k) are created with BioRender (Created in BioRender. https://BioRender.com/n400kl0). To assess whether these effects of SLC1A5_var overexpression translated into impaired antitumor activity in vivo, LPS-trained WT and Tg BMDMs were adoptively transferred into recipient mice with MC38 cells (Fig. 4d). As observed previously, compared with naïve WT controls, trained WT macrophages markedly suppressed tumor growth (Fig. 4e, f). In contrast, trained Tg BMDMs exhibited only a modest improvement over naïve Tg BMDMs in terms of tumor growth suppression, and their antitumor activity was substantially weaker than that of trained WT macrophages (Fig. 4e, f). Notably, although less pronounced than under trained conditions, naïve Tg BMDMs showed relatively lower cytokine production and phagocytosis against cancer cells in vitro (Fig. 4b, c) and were less effective at suppressing tumor growth in vivo compared with WT BMDMs (Fig. 4d–f). GSEA of GO terms using DEGs between naïve Tg and WT BMDMs revealed altered pathways involved in the regulation of immune responses (Supplementary Fig. 4i, j). These findings indicate that the expression and activity of SLC1A5_var is closely associated with the regulation of innate immune memory and responses. Consistent with earlier findings, tumor immune profiling revealed that the adoptive transfer of trained WT BMDMs increased the CD86+/CD206+ macrophage ratio within tumor-infiltrating populations, whereas SLC1A5_var Tg macrophages failed to induce a comparable shift (Fig. 4g and Supplementary Fig. 4k, l). Moreover, only trained WT BMDMs promoted T-cell recruitment (Fig. 4h) and the expansion of IFN-γ+ CTLs, and these effects were abolished in SLC1A5_var Tg (Fig. 4i). In support of this, in vitro migration assays demonstrated that trained WT macrophages facilitated CD8⁺ T-cell recruitment, whereas SLC1A5_var Tg BMDMs did not (Fig. 4j). Together, these results indicate SLC1A5_var overexpression could diminish LPS-training-induced macrophage-dependent TME remodeling. Next, we tested whether pharmacological inhibition of SLC1A5_var could restore LPS-induced training using iMQT_02050. After overnight LPS exposure and washout, BMDMs were treated with iMQT_020 during the resting phase until day 6 (Fig. 4k). Pharmacological inhibition of SLC1A5_var reduced mitochondrial glutamine transport in both WT and Tg BMDMs (Fig. 4l) and rescued the LPS-induced training phenotype in Tg BMDMs, as demonstrated by increased cytokine productions (Fig. 4m and Supplementary Fig. 4m) and enhanced phagocytosis of cancer cells (Fig. 4n). Notably, P3C and poly(I:C) induced macrophage training was similarly attenuated in Tg BMDMs, whereas iMQT_020 treatment enhanced training responses (Supplementary Fig. 4n, o), indicating that SLC1A5_var dependent regulatory axis may also be engaged under these training conditions. Together, these results suggest that the downregulation of SLC1A5_var is required for optimal metabolic reprogramming and the acquisition of antitumor effector functions during LPS-trained immunity. Glutamine metabolism regulates the epigenetic landscape in trained macrophages Given the established interplay between cellular metabolism and epigenetic rewiring in trained immunity, we hypothesized that diminished mitochondrial glutamine import and glutaminolysis may alter the epigenetic landscape governing innate immune memory. To test this hypothesis, we examined whether SLC1A5_var overexpression affects epigenetic remodeling in macrophages. We assessed global chromatin accessibility using the assay for transposase-accessible chromatin with sequencing (ATAC-seq) in naïve and LPS-trained BMDMs from WT and Tg mice. A total of 1,302 differentially accessible regions (DARs) were identified between trained WT and Tg BMDMs (Fig. 5a and Supplementary Fig. 5a). Of these, 782 regions (60.06%) were more accessible in WT cells; these regions were mapped to genes involved in inflammatory responses (e.g., Cxcl1, IL6, and Ifngas1) and epigenetic regulation (e.g., Zfp932 and Baz1b) (Fig. 5b and Supplementary Fig. 5b). Conversely, 520 regions (39.94%) displayed greater accessibility in Tg macrophages; these regions were associated with immune suppression (e.g., Socs1) and metabolic regulation (e.g., Mid1 and Sod3). GO enrichment analysis confirmed that the genes corresponding to the DARs in trained WT BMDMs compared with naïve WT BMDMs were enriched in pathways related to chemokine and cytokine production, regulation of inflammatory responses, and T-cell migration (Supplementary Fig. 5c, d), whereas the genes corresponding to the DARs in trained Tg BMDMs compared with trained WT BMDMs were negatively associated with these pathways; this indicates that SLC1A5_var overexpression constrains chromatin accessibility at immune-related loci (Fig. 5c, d). Notably, hypergeometric optimization of motif enrichment (HOMER) motif analysis revealed greater enrichment of AP-1 family transcription factor motifs, including c-FOS and JunB, in trained WT BMDMs than in SLC1A5_var Tg BMDMs (Supplementary Fig. 5e, f). As AP-1 has been implicated in chromatin remodeling through the recruitment of histone-modifying enzymes55–57, these data indicate defective epigenetic programming in SLC1A5_var Tg macrophages. Glutaminolysis reshapes the epigenetic landscape through histone modification. a Volcano plot showing differentially accessible genes in WT and SLC1A5_var transgenic (Tg) BMDMs identified by assay for ATAC-sequencing (ATAC-seq). Statistics were calculated using edgeR with Benjamini–Hochberg correction. b IGV snapshots showing ATAC-seq tracks at the Cxcl1 locus. Region with a significant decrease (WT trained vs. Tg trained, p = 0.007) is highlighted. c, d Pathway enrichment analysis in Tg versus WT trained BMDMs (c). Representative gene set enrichment analysis (GSEA) plot. FDR, false discovery rate; NES, normalized enrichment score. Representative plots of intracellular staining for H3K4me1, H3K4me2 and H3K27ac (e), quantitative analysis of mean fluorescence intensity (MFI) (f) (n = 5 independent biological replicates). Exact p-values (p < 0.0001): f 2 × 10−7, 8 × 10−7, 9 × 10−6. g, h H3K4me3 ChIP-sequencing analysis results. Clustering heatmap showing differential peaks (g). IGV snapshots showing the Il1α and Il1β loci (h). Regions with significant decreases in peaks (WT trained vs Tg trained, p = 0.017, 0.02) are highlighted. i, j Pathway enrichment analysis of ChIP-seq differential peaks in Tg versus WT trained BMDMs (i). GSEA plot for GO pathways (j). k, l LPS-trained BMDMs were transfected with empty vector (EV) or SLC1A5_var, GLS1, or GLUD1 expression vectors and challenged with MC38 cells. Analysis of IL-1β and IL-6 secretion (n = 5 biological replicates) (k) and H3K4me3 enrichment at their promoter regions by ChIP‒qPCR (n = 3 biological replicates) (l), one-way ANOVA test. Exact p-values (p < 0.0001): k 1 × 10−8, 1 × 10−7, 7 × 10−12, 1 × 10−10, 2 × 10−6, 1 × 10−6; l 2 × 10−7, 3 × 10−6, 1 × 10−5. m, n LPS-stimulated BMDMs were treated with iMQT_020, CB839, or R162 during the resting phase and challenged with MC38 cells. Analysis of IL-1β and IL-6 secretion (n = 5) (m) and H3K4me3 enrichment at their promoter regions by ChIP‒qPCR (n = 3) (n), one-way ANOVA test. Exact p-values (p < 0.0001): m 1 × 10−6, 8 × 10−6, 9 × 10−6, 1 × 10−9, 3 × 10−7. For (f, k, m), box plots show the median (center), 25th and 75th percentiles (box boundaries), and whiskers extending to the minima to maxima. Significance was assessed using a two-tailed Student’s t-test unless otherwise specified. The data are presented as the mean ± SEM. Source data are provided as a Source Data file. We next examined histone modifications central to trained immunity to further elucidate the role of SLC1A5_var and glutaminolysis in the epigenetic regulation of trained macrophages. Global enrichment analysis demonstrated that LPS training increased the levels of active chromatin marks, including H3K4me3 and H3K27ac (Fig. 5e, f and Supplementary Fig. 5g). Notably, SLC1A5_var overexpression preferentially reduced H3K4me3, but did not affect H3K27ac and H3K4me1 levels, suggesting a specific link between SLC1A5_var-mediated regulation and H3K4me3 (Fig. 5e, f). Consistently, ChIP‒qPCR confirmed that LPS-training-mediated H3K4me3 enrichment at loci of proinflammatory cytokines was reduced in Tg BMDMs, whereas H3K27ac and H3K4me1 enrichment were comparable between genotypes (Supplementary Fig. 5h–j). To obtain a genome-wide view, we performed ChIP-seq profiling, which revealed distinct H3K4me3 peak patterns between LPS-trained and naïve WT BMDMs that were diminished in SLC1A5_var Tg BMDMs (Fig. 5g and Supplementary Fig. 5k, l). Trained WT BMDMs exhibited higher H3K4me3 levels at cytokine loci (e.g., Il1α, Il1β and Il6), whereas H3K4me3 levels were reduced in trained Tg BMDMs (Fig. 5h and Supplementary Fig. 5m). GO analysis of the differential H3K4me3 peaks between trained WT and trained Tg BMDMs revealed enrichment of pathways related to inflammatory responses, cytokine production, and leukocyte migration (Fig. 5i, j). As these epigenetic changes suggested enhanced migratory potential in trained WT macrophages, we next assessed their tumor-directed motility. In line with ChIP-seq analysis, in vitro transwell migration assays revealed enhanced migration of trained WT BMDMs toward MC38 cells, whereas trained Tg BMDMs exhibited reduced migration (Supplementary Fig. 5n). Notably, GO analysis of differential H3K4me3 peaks between naïve and trained WT BMDMs revealed enrichment of genes associated with glucose and carbohydrate metabolism, suggesting an epigenetic basis for elevated glycolysis in macrophages (Supplementary Fig. 5o). To directly test the link between mitochondrial glutaminolysis and H3K4me3 regulation during LPS-induced macrophage training, we examined the effects of SLC1A5_var, GLS1 and GLUD1 overexpression during the resting phase following initial LPS exposure (Supplementary Fig. 5p). These enzymes act sequentially in glutamine catabolism—SLC1A5_var mediates mitochondrial glutamine import, GLS1 converts glutamine to glutamate, and GLUD1 oxidizes glutamate to α-KG—constituting the core pathway of mitochondrial glutaminolysis. Overexpression of these enzymes reduced IL-1β and IL-6 production and decreased H3K4me3 enrichment at their promoters (Fig. 5k, l), recapitulating the phenotype observed in SLC1A5_var Tg BMDMs. Notably, GLS1 or GLUD1 overexpression exerted a significantly weaker inhibitory effect than SLC1A5_var overexpression, possibly suggesting that SLC1A5_var may serve as an upstream rate-limiting step in mitochondrial glutaminolysis. Conversely, pharmacological inhibition of SLC1A5_var, GLS1 and GLUD1 with iMQT_020,50 CB839,58 and R162,59 respectively, increased cytokine production and H3K4me3 at cytokine promoters (Fig. 5m, n). Together, these findings support a model in which SLC1A5_var couples mitochondrial glutaminolysis with the regulation of H3K4me3, thereby shaping the epigenetic landscape that underlies trained macrophage immunity. SLC1A5_var establishes an α-KG–KDM5B metabolic–epigenetic checkpoint in trained macrophages Having established that SLC1A5_var downregulation restricts glutaminolysis while promoting H3K4me3, we next sought to define the molecular mechanism underlying this effect. Among the metabolites downstream of glutaminolysis, our metabolomics analysis revealed a significant reduction in α-KG levels (Fig. 2j). Because α-KG serves as an essential substrate for histone demethylases that remove H3K4me3 marks, including KDM5 family members60,61, we hypothesized that reduced α-KG availability preserves H3K4me3 and thereby sustains trained immunity. Consistent with previous observations, LPS training progressively reduced intracellular glutamate and α-KG levels (Supplementary Fig. 6a), whereas SLC1A5_var overexpression restored α-KG levels (Supplementary Fig. 6b). To test the functional consequences of these changes, we treated cells with the cell-permeable α-KG analog dimethyl-α-KG (DM-α-KG) following LPS exposure. DM-α-KG treatment restored mitochondrial respiration in a dose-dependent manner, which indicates recovery of TCA flux (Supplementary Fig. 6c, d). In contrast, despite the similar reduction in intracellular levels of succinate and fumarate, treatment with succinate or fumarate analogs had a minimal effect on mitochondrial respiration (Supplementary Fig. 6e–h), underscoring the importance of glutaminolysis and α-KG supply in oxidative phosphorylation. Moreover, DM-α-KG treatment simultaneously abolished the trained phenotype, as reflected by reduced cytokine secretion in both of the trained WT and Tg BMDMs (Supplementary Fig. 6i, j). Importantly, enhanced training responses induced by pharmacological inhibition of glutaminolysis were accompanied by reduced intracellular α-KG levels (Fig. 6a), and concomitant DM-α-KG supplementation restored H3K4me3 demethylation, thereby counteracting the effects of glutaminolysis inhibitors on cytokine expression during trained immunity (Fig. 6b, c and Supplementary Fig. 6k, l). Reduced α-KG–KDM5 activity sustains H3K4me3 and epigenetic memory in LPS-trained macrophages. a–c Intracellular α-ketoglutarate (α-KG) levels in bone marrow-derived macrophages (BMDMs) treated with indicated inhibitors (n = 5 biological replicates) (a). IL-6 secretion (n = 5 biological replicates) (b) and H3K4me3 enrichment at the promoter of Il6 by ChIP‒qPCR (n = 3 biological replicates) (c), one-way ANOVA test. Exact p-values (p < 0.0001): b 2 × 10−7, 3 × 10−6, 7 × 10−6, 2 × 10−6, 2 × 10−8, 4 × 10−7, 2 × 10−5. d–f LPS-trained BMDMs were transfected with empty vector (EV) or indicated expression vectors. Intracellular α-KG levels (n = 5 biological replicates) (d). IL-6 secretion (n = 5) (e) and H3K4me3 enrichment by ChIP‒qPCR (n = 3 biological replicates) (f). One-way ANOVA test. Exact p-values (p < 0.0001): e 2 × 10−5, 4 × 10−6, 5 × 10−7, 2 × 10−7, 3 × 10−6, 2 × 10−6; f 4 × 10−6, 1 × 10−4. g Schematic summary of mitochondrial glutaminolysis and α-KG-dependent histone demethylation. h Mean fluorescence intensity (MFI) of H3K4me3 levels in BMDMs transfected with scramble (si-Con) or specific siRNAs (n = 5 biological replicates). Exact p (p < 0.0001): 1 × 10−8, 3 × 10−7, 5 × 10−6. i IL-6 secretion in BMDMs challenged with MC38 cells (n = 6 biological replicates), one-way ANOVA test. Exact p-values (p < 0.0001): 6 × 10−13, 4 × 10−7. IL-6 secretion (n = 6 biological replicates) (j), and H3K4me3 enrichment at Il6 and Il1b promoters (k) BMDMs with indicated treatment (n = 4 biological replicates). One-way ANOVA test. Exact p-values (p < 0.0001): (j) 1 × 10−6, 2 × 10−6, 2 × 10−8, 8 × 10−6, 6 × 10−6; k 2 × 10−9, 1 × 10−5, 4 × 10−7, 6 × 10−9, 2 × 10−5, 7 × 10−7, 4 × 10−8, 4 ×10−10. l–n BMDMs from C57BL/6 donor with indicated treatment were adoptively transferred to recipient mice (n = 7) (l). Tumor volume (m) and tumor weight (n). Scale bar, 1 cm. One-way ANOVA test. o Ratio of CD86+/CD206+ tumor-associated macrophages (TAM) in harvested tumors from (l–n) (n = 7), one-way ANOVA test. For (b, e, i, j), box plots show median (center), 25th and 75th percentiles (box boundaries), and whiskers extending to the minima to maxima. The data are presented as the mean ± SEM. Significance was assessed using a two-tailed Student’s t-test. Source data are provided as a Source Data file. (l) is created with BioRender (Created in BioRender. https://BioRender.com/n400kl0). We next examined whether altered α-KG availability directly influenced histone methylation and trained responses. Overexpression of mitochondrial glutaminolysis-related enzymes blunted training responses, concomitant with the restoration of α-KG levels (Fig. 6d–f and Supplementary Fig. 6m, n). Conversely, inhibition of SLC1A5_var with iMQT_020 decreased α-KG levels and increased H3K4me3 and cytokine production (Fig. 6d–f and Supplementary Fig. 6m, n). Together, these findings identify SLC1A5_var as an upstream metabolic checkpoint that regulates glutamine-derived α-KG availability, thereby coupling mitochondrial metabolism with the epigenetic programming of innate immune memory (Fig.6g). Importantly, inhibition of KDM5 activity—a family of α-KG-dependent histone demethylases that remove methyl group from H3K4me3—with pan-KDM5 inhibitor CPI-455 counteracted the effects of DM-α-KG and prevented loss of H3K4me3 (Supplementary Fig. 7a). As different KDM5 isoforms have been implicated in the epigenetic regulation of distinct gene sets62, we next knocked down KDM5A, KDM5B, and KDM5C to identify another key regulator in epigenetic memory (Supplementary Fig. 7b). While KDM5A and KDM5C knockdown also increased global H3K4me3 levels (Fig. 6h and Supplementary Fig. 7b), only KDM5B knockdown have enhanced trained immune responses against cancer cells, as evidenced by increased phagocytosis in BMDMs (Supplementary Fig. 7c, d) and elevated IL-1β and IL-6 secretion (Fig. 6i and Supplementary Fig. 7e). Consistently, treatment with CPI-455 or the KDM5B-selective inhibitor GSK467 were both able to restore training responses and H3K4me3 levels in DM-α-KG treated cells, confirming a specific role for KDM5B among KDM5 family in epigenetic regulations of trained immunity (Fig. 6j, k and Supplementary Fig. 7f, g). Notably, similar trends were observed in P3C- and poly(I:C) induced training settings (Supplementary Fig. 7h), suggesting that the α-KG–KDM5B axis may be engaged in multiple stimulus-dependent contexts. As we found that SLC1A5_var regulates α-KG availability and epigenetic remodeling in LPS-trained macrophages in vitro, we next tested whether this mechanism also occurs in vivo to enhance antitumor immunity. To this end, in vitro-trained BMDMs were treated with DM-α-KG and/or GSK467 during the resting phase and then adoptively transferred into the recipient mice (Fig. 6l). The tumor-suppressive effect of trained macrophages was abrogated by DM-α-KG treatment, whereas GSK467 further potentiated the training effect (Fig. 6m, n). Importantly, KDM5B inhibition reversed the suppressive effect of DM-α-KG, underscoring the importance of metabolic–epigenetic interplay in governing trained immunity. Flow cytometric analysis of tumor-infiltrating immune cells revealed that DM-α-KG treatment blunted LPS training, reduced the recruitment of inflammatory monocytes and CCR2+ macrophages to the tumor site, and decreased the CD86+/CD206+ tumor-associated macrophage ratio in the TME (Fig. 6o and Supplementary Fig. 7i–k). These effects were reversed by the GSK467-mediated potentiation of LPS training. Consistent with earlier findings, the adoptive transfer of LPS-trained macrophages also increased T-cell infiltration and IFNγ⁺ CTL levels, and these effects were augmented by GSK467 and diminished by DM-α-KG (Supplementary Fig. 7l–o). These findings demonstrate that targeting the α-KG–KDM5B axis modulates the efficiency of LPS-trained macrophages and their antitumor activity in vivo. Collectively, our results thus far support a model in which the SLC1A5_var–α-KG–KDM5B axis functions as a metabolic–epigenetic checkpoint that regulates H3K4me3 and contributes to the extent of trained immunity. Targeted inhibition of SLC1A5_var promotes macrophage training to combat cancer Given the central role of SLC1A5_var in coordinating glutamine metabolism and epigenetic rewiring, we next tested whether its targeted repression could enhance antitumoral macrophage training both in vitro and in vivo. shRNA-mediated knockdown of SLC1A5_var (Supplementary Fig. 8a,b) significantly reduced mitochondrial 3H-glutamine uptake (Supplementary Fig. 8c), reduced intracellular α-KG levels (Supplementary Fig. 8d) and suppressed mitochondrial respiration (Supplementary Fig. 8e,f). These metabolic reprogramming increased global H3K4me3 (Supplementary Fig. 8g) without altering H3K27 acetylation (Supplementary Fig. 8h). Functionally, knockdown of SLC1A5_var enhanced LPS training, as evidenced by increased phagocytosis (Supplementary Fig. 8i) and elevated IL-1α, IL-1β and IL-6 expression (Supplementary Fig. 8j). DM-α-KG was able to counteracted the enhanced phagocytosis observed in trained SLC1A5_var knockdown BMDMs (Supplementary Fig. 8i). Notably, these changes were also evident in naïve BMDMs, indicating that reducing SLC1A5_var partially restructures the α-KG–KDM5B–H3K4me3 axis, thereby mimicking the features of the trained phenotype. To extend these findings in vivo, we generated myeloid-specific SLC1A5_var knockdown mice (sh1A5_varLyzM−cre) to assess macrophage-dependent outcomes while minimizing confounding effects (Supplementary Fig. 9a). SLC1A5_var knockdown was validated in tissue macrophages isolated from spleen and lymph nodes by magnetic sorting, but not in lymphocytes from the same tissues (Supplementary Fig. 9b,c). This macrophage-specific silencing did not affect myeloid development or macrophage populations in bone marrow, spleen, or lymph nodes (Supplementary Fig. 9d). Similarly, SLC1A5_var knockdown was confirmed in BMDMs without altering their differentiation in vitro (Supplementary Fig. 9e–g). Next, we administered endotoxin-free saline or LPS intraperitoneally to control (CtrlCre) and sh1A5_varLyzM−cre mice, followed one week later by subcutaneous inoculation of MC38 cells (Fig. 7a). Following LPS injection, both genotypes showed no signs of septic shock, and the serum IL-6 levels returned to baseline by day 7, indicating the absence of prolonged systemic inflammation (Supplementary Fig. 9h). Compared with LPS-trained control littermates (CtrlCre), LPS-trained shSLC1A5_varLyzM−cre mice exhibited significantly reduced MC38 tumor growth (Fig. 7b,c). Supporting this observation, BMDMs derived from sh1A5_varCre mice also displayed heightened training ex vivo, characterized by elevated cytokine production and increased cancer cell phagocytosis (Supplementary Fig. 9i–k). Tumor immune profiling revealed that tumors from LPS-trained shSLC1A5_varLyzM−cre mice exhibited increased CD8⁺ T-cell infiltration and elevated frequencies of IFNγ⁺ CTL (Fig. 7d and Supplementary Fig. 9l), accompanied by increased CD86+/CD206+ tumor-associated macrophage ratios (Fig. 7e) and enhanced TNF-α+ and IL-6+ tumor-associated macrophages (Fig. 7f). Importantly, macrophages isolated from the tumors of LPS-trained mice showed elevated H3K4me3 enrichment, which was further amplified by SLC1A5_var knockdown (Fig. 7,g), supporting the notion that the repression of glutaminolysis reinforces H3K4me3-driven epigenetic memory in trained macrophages. Consistently, LPS-trained sh1A5_varCre BMDMs displayed reduced mitochondrial glutamine uptake, decreased intracellular α-KG, and increased H3K4me3 enrichment (Supplementary Fig. 9m–o). SLC1A5_var inhibition potentiates macrophage training and antitumor immunity. a–c 6-week-old CtrlLyzM−Cre and sh1A5_varLyzM−Cre mice received a single i.p. injection of PBS or LPS, and 7 days later, the mice were subcutaneously inoculated with MC38 cells (n = 8 mice per group) (a). Tumor volume was monitored for 21 days (b). Tumor weight at the end of the experiment (c). Scale bar, 1 cm. One-way ANOVA test. Exact p-values (p < 0.0001): (b) 3 × 10−5, 1 × 10−7; c 3 × 10−5. d Percentage of IFNγ+ cytotoxic T lymphocytes (CTLs) in tumors from (a–c) (n = 7). e Ratio of CD86+ and CD206+ tumor-associated macrophages (TAM) among tumor-infiltrated immune cells (n = 7). Exact p-values (p < 0.0001): 2 × 10−5. f Representative flow cytometry plots (left) and quantitative analysis (right) of tumor-infiltrated CD11b+F4/80+IL-6+ and CD11b+F4/80+TNF-α+ macrophages in tumors (n = 7). g H3K4me3 levels in tumor-infiltrated macrophages from the experiment in (a–c) (n = 7). h Pathway enrichment analysis of ChIP-seq differential peaks. FDR, false discovery rate. i ChIP‒qPCR analysis of H3K4me3 enrichment at Il6 promoters in bone marrow-derived macrophages (BMDMs) of indicated groups (n = 7 independent biological replicates). Exact p-values from left to right, 6 × 10−4, 3 × 10−5, 6 × 10−4, 2 × 10−4, 7 × 10−4, 5 × 10−4, 2 × 10−3, 5 × 10−7, 4 × 10−4. j IL-6 secretion in BMDMs challenged with MC38 cells (n = 6 independent biological replicates). Exact p-values from left to right, 8 × 10−3, 4 × 10−9, 1 × 10−6, 3 × 10−8, 1 × 10−5, 2 × 10−7, 3 × 10−7, 3 × 10−9, 4 × 10−8, 2 × 10−7, 7 × 10−9. The data are presented as the mean ± SEM. Significance was assessed using a two-tailed Student’s t-test. *p < 0.05, **p < 0.01 and ***p < 0.001. Source data are provided as a Source Data file. (a) is created with BioRender (Created in BioRender. https://BioRender.com/n400kl0). To further dissect how SLC1A5_var loss in macrophages shapes epigenetic memory, we performed H3K4me3 ChIP-seq. Knockdown induced differential peaks at the promoters of key proinflammatory cytokine loci (Supplementary Fig. 9p), and GO analysis revealed enrichment in pathways associated with cytokine production, inflammatory regulation, T-cell activation, and phagocytosis in sh1A5_Var BMDMs (Fig. 7,h and Supplementary Fig. 9q). Consistently, ChIP-qPCR confirmed promoter-specific H3K4me3 enrichment upon SLC1A5_var knockdown, and importantly, DM-α-KG supplementation reversed these effects, reducing H3K4me3 enrichment and suppressing cytokine expression (Fig. 7i,j and Supplementary Fig. 9r, s). These results confirm our findings that the amplified trained phenotype is driven by restriction of mitochondrial glutaminolysis, which limits α-KG availability and thereby strengthens H3K4me3-dependent epigenetic memory. Taken together, these data provide direct evidence that targeting SLC1A5_var enhances the durability and efficacy of trained macrophage responses in vivo, promoting a proinflammatory TME and augmenting both innate and adaptive antitumor immunity. Discussion In this study, we identify a previously unrecognized mechanism of trained immunity in which selective metabolic restriction, rather than activation, establishes durable epigenetic memory in macrophages and promotes antitumor responses. We demonstrate that LPS induces macrophage training through TLR4-NF-κB-mediated repression of the mitochondrial glutamine transporter SLC1A5_var, thereby limiting glutaminolysis and reducing intracellular α-KG levels. This reduction constrains the activity of the α-KG-dependent histone demethylase KDM5B, thereby contributing to the preservation of H3K4me3 and stabilizing transcriptionally active chromatin at inflammatory loci. As a result, LPS-trained macrophages exhibit enhanced cytokine production and phagocytic activity, contributing to remodeling of the TME and suppression of tumor growth. Classical models of trained immunity, exemplified by β-glucan and BCG vaccination, have largely emphasized glycolytic reprogramming and the accumulation of metabolites such as lactate and acetyl-CoA that promote histone acetylation and lactylation30,32,63. While this glycolysis-centric framework has provided a foundation for understanding metabolic–epigenetic coupling, the diversity of training stimuli suggests the existence of additional, context-specific regulatory circuits. Our findings extend this paradigm by identifying a complementary mechanism in which suppression of mitochondrial glutaminolysis reinforces epigenetic memory through α-KG limitation. In this model, trained immunity can arise through two distinct but potentially cooperative metabolic modes: metabolite accumulation and metabolite restriction, depending on the stimulus and cellular context. Consistent with this context-dependent framework, we observed that additional TLR agonists, including P3C and Poly(I:C), but not R848, exhibited patterns broadly consistent with LPS-induced downregulation of SLC1A5_var and enhanced inflammatory responses. This is further supported by ChIP-qPCR analysis demonstrating that NF-κB recruitment to the Slc1a5_var promoter is stimulus-dependent, despite robust recruitment to canonical targets such as Tnfa. These results suggest that the SLC1A5_var- α-KG axis may be selectively engaged in a subset of TLR-driven training conditions rather than representing a universal mechanism. Future studies will be required to define the upstream signaling determinants that regulate SLC1A5_var expression and to clarify how distinct metabolic programs are integrated across diverse innate immune stimuli. Glutamine metabolism is widely recognized as a central component of immunometabolic regulation in macrophages, particularly in shaping activation and polarization states37,38. Increased glutaminolysis and α-KG production have been linked to anti-inflammatory macrophage polarization64 and immunosuppressive TAM phenotypes within the TME65,66. Conversely, pharmacological inhibition of glutamine metabolism has been shown to enhance myeloid-driven antitumor immunity67,68. Building on this framework, our findings identify glutamine metabolism as a regulator of innate immune memory, extending its role beyond polarization toward dynamic epigenetic reprogramming. Specifically, suppression of mitochondrial glutaminolysis enhances macrophage-mediated inflammatory responses and supports antitumor immunity, establishing a functional link between glutamine utilization and trained immunity. However, given the shared dependence on glutamine metabolism between tumor and immune cells, therapeutic modulation of this pathway will require careful consideration of metabolic competition within the TME. In particular, α-KG availability may represent a critical determinant of how glutamine metabolism is partitioned between tumor and immune compartments69. From a translational perspective, these findings have important implications. Pharmacological inhibition of glutaminolysis, or targeting SLC1A5_var itself, enhances trained immune responses, suggesting that agents originally developed as metabolic inhibitors for cancer therapy may be repurposed to reinforce innate immune memory68,70. In this context, the SLC1A5_var–α-KG–KDM5B axis functions as a tunable metabolic–epigenetic checkpoint, revealing a conceptual inversion of conventional metabolic interventions: rather than promoting nutrient utilization, selective metabolic restriction can stabilize and amplify immune activation. Functionally, this metabolic–epigenetic rewiring enhances macrophage effector capacity, as evidenced by increased inflammatory cytokine production and augmented phagocytic activity. The preservation of tumor control in Rag1−/− mice indicates that LPS-induced antitumor effects are primarily mediated by trained innate immunity. In immunocompetent settings, the increased recruitment of cytotoxic T lymphocytes further indicates that trained macrophages can potentiate antitumor responses through innate-adaptive crosstalk. These findings support a model in which metabolically reprogrammed macrophages contribute both directly and indirectly to tumor control. Importantly, myeloid-specific knockdown or pharmacological inhibition of the SLC1A5_var–α-KG–KDM5B axis recapitulated these antitumor effects, whereas α-KG supplementation attenuated them, supporting this pathway as a functionally relevant and potentially targetable regulatory node. This bidirectional modulation further highlights the role of α-KG availability in controlling trained immunity. Despite these promising findings, several limitations should be acknowledged. Most experiments were performed in murine systems, and thus the relevance of this pathway in human innate immunity remains to be established. In addition, although our data support a role for KDM5B-dependent epigenetic regulation, other histone-modifying enzymes may also contribute to this process and warrant further investigation. Reprogramming macrophages within the TME has emerged as a central objective in cancer immunotherapy. Our findings suggest that restricting mitochondrial glutamine flux promotes macrophage effector function and shifts the TME toward a more immune-active state. Such metabolic reprogramming may complement existing immunotherapeutic strategies, including immune checkpoint blockade71 and chimeric antigen receptor (CAR)-engineered macrophage therapies72. Given the limitations of systemic LPS administration, future work should focus on developing safe and targeted approaches to induce trained immunity in human myeloid cells. In summary, repression of mitochondrial glutaminolysis through SLC1A5_var defines the SLC1A5_var–α-KG–KDM5B axis as a metabolic–epigenetic checkpoint that stabilizes inflammatory epigenetic states. By integrating controlled metabolic restriction into the framework of trained immunity, this study expands current understanding of immunometabolic regulation and provides a conceptual basis for exploiting metabolic limitation as a therapeutic strategy in cancer immunotherapy. Methods Mice All the mice used in this study were housed at 5 per cage at 24 °C on a 12-h light/dark cycle. The mice were fed a rodent chow diet (38057, Purina) and provided water ad libitum in a pathogen-free facility, and their health status was monitored via culture, serum analysis, and microscopic examination. C57BL/6N mice were purchased from Orient Corporation (Seongnam, Republic of Korea). Rag1−/− mice and LyzM-Cre mice on a C57BL/6 background were provided by Dr. Dae-Sik Lim. C3H/HeN (TLR4 wild-type) and C3H/HeJ (TLR4LPS−D) mice were purchased from the Jackson Laboratory (Bar Harbor, ME, USA). SLC1A5_var transgenic mice (SLC1A5_var Tg) and LSL-shSLC1A5_var transgenic mice were generated by the GEM division of Macrogen Inc. (Seoul, Republic of Korea) under pathogen-free conditions. Briefly, female C57BL/6 mice were treated with PMSG and hCG. After 48 h, these females were mated with C57BL/6 stud males. The following day, vaginal plug-positive females were sacrificed, and fertilized embryos were collected. The linearized transgene fragment was microinjected into the pronuclei of fertilized one-cell embryos, allowing untargeted integration of the transgene into the genome. The injected embryos were incubated at 37 °C for 1–2 h. Fourteen to sixteen injected one-cell-stage embryos were then transferred into the oviducts of pseudopregnant ICR recipient mice. shSLC1A5_varLyzM−Cre mice were generated by crossing LyzM-Cre mice with LSL-shSLC1A5_var mice. The mice were at least 6 weeks of age, and all the experiments were conducted in accordance with relevant ethical regulations and guidelines at the Korea National Institute of Health (BEC-CAN-2016-001 and 002). All studies were approved by the Institutional Animal Care and Use Committee of Yonsei University (IACUC-A-202204-1455-02, IACUC-A-202405-1856-01, IACUC-A-202508-2098-01). Cell lines Raw264.7 mouse monocytic/macrophages, L-929 mouse fibroblasts, and THP-1 and U937 human monocytes were obtained from ATCC. MC38 mouse colon adenocarcinoma cell was provided by Dr. Do Sik Min. LLC mouse lung adenocarcinoma cells were obtained from Ubigene. All cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS) and maintained at 37 °C with 5% CO2. METHOD DETAILS In vivo macrophage training via LPS Mice were randomly divided into a naïve control group and an LPS-trained group. LPS-trained group mice were i.p. injected with a single dose of LPS (50 μg/kg) (serotype 055LB5; Sigma‒Aldrich) on day 0, followed by subcutaneous inoculation of tumor cells on day 7, 14, or 21. Mice treated with the same volume of PBS were used as naïve controls. Alternatively, mice were euthanized 7 days after i.p. LPS injection to assess BM, spleen, and lymph node phenotype changes. Cells were cultured on plates for further ex vivo restimulation experiments or were stained for surface markers and intracellular cytokines for flow cytometry analysis. In vivo tumor xenograft experiments For the subcutaneous tumor model, 50 μl of tumor cell suspension containing 1 × 106 MC38 cells was subcutaneously inoculated into the right flank of mice on a C57BL/6 background to establish a syngeneic tumor mouse model. Tumor growth was monitored with a caliper for 21 days after inoculation. The tumor volume (TV) was calculated using the formula: TV = 0.5 × length × width2. The maximal tumor volume permitted by the Institutional Animal Care and Use Committee of Yonsei University was 2000 mm3, which was not exceeded during this study. Isolation and preparation of primary macrophages Spleens and lymph nodes collected from mice were mechanically dissociated, and cell suspensions were filtered through a 40-μM cell strainer (Corning). Cells were then incubated with RBC lysis buffer (Invitrogen) for 5 min at 4 °C to remove red blood cells. Macrophages were sorted with a MACS cell separation system (Miltenyi Biotec) with anti-F4/80 MicroBeads (Miltenyi Biotec) according to the manufacturer’s instructions. The isolated splenic and lymph node macrophages were cultured in complete RPMI-1640 medium. BMDMs were prepared by flushing femurs and tibias obtained from 6- to 8-week-old euthanized mice with cold RPMI-1640. The cell suspensions were filtered through a 40-μM cell strainer (Corning). Cells were then incubated with RBC lysis buffer (Invitrogen) for 5 min at 4 °C to remove red blood cells. BMDM was cultured in fresh complete RPMI-1640 with 30% M-CSF derived from L-929. L-929 cultured media will be referred as M-CSF conditioned throughout the text. After 6–7 days of culture, the adherent cells were harvested for further experiments. Peritoneal cells were collected after i.p. injection of 5 ml of ice-cold DPBS. The cell suspension was then seeded in culture dishes, and 2 h later, floating cells were removed by washing with DPBS. The attached cells were considered peritoneal macrophages. Tumor dissociation and preparation of single-cell suspensions Tumor dissociation and enrichment of tumor immune cells in subcutaneous murine tumors were performed as previously described73,74. In Brief, tumors were minced with razor blades or scalpels and digested in precooled RPMI-1640 medium supplemented with 25 μg/ml DNAse (Roche) and 50 μg/ml Liberase DL (Roche) at 37 °C for 45 min with occasional pipetting. Ice-cold RPMI medium supplemented with 2 mM EDTA was used to neutralize the digestion. The samples were then mechanically disaggregated with a syringe plunger through a 40-μM cell strainer and centrifuged at 300 × g for 5 min at 4 °C. Red blood cells were lysed with RBC lysis buffer (Invitrogen) for 5 min at 4 °C. Density gradient centrifugation was performed with Percoll Plus (Cytiva) to enrich immune cells from digested tumor cell suspensions. In brief, the cells were resuspended in 5 ml of 40% Percoll and slowly layered on top of 5 mL of 80% Percoll in a 15-ml conical tube without mixing the two phases. The layered tubes were centrifuged at 320 × g for 25 min at room temperature with no brakes. Leukocytes were collected from the interphase and resuspended in complete RPMI-1640 or FACS buffer for further experiments. Flow cytometry Cell analysis was performed using FACS Aria III (BD, Heidelberg, Germany). Single-cell suspensions were blocked with an anti-mouse CD16/CD32 monoclonal antibody (Thermo Fisher Scientific) for 20 min at 4 °C and then stained with Fixable Viability Dye eFluor 450 (Thermo Fisher Scientific) for 15 min at 4 °C. For cell surface phenotype analysis, a Pacific Blue anti-mouse lineage cocktail (anti-mCD3, anti-mLy6G(Ly6C), anti-mCD11b, anti-mCD45R (B220), and anti-mTer-119; BioLegend), FITC anti-mouse CD45 (BioLegend), PE-Cy7 anti-mouse CD48 (BioLegend), PerCP-Cy5.5 anti-mouse CD150 (BioLegend), APC anti-mouse Sca-1 (Miltenyi Biotec), FITC anti-mouse CD117 (c-Kit) (BioLegend), PE anti-mouse CD135 (BioLegend), APC-Cy7 anti-mouse CX3CR1 (BioLegend), PE anti-mouse CD192 (CCR2) (BioLegend), PE-Cy7 anti-mouse CD3 (BioLegend), PE-Cy5 anti-mouse CD8a (BioLegend), PE anti-mouse CD8a (Thermo Fisher Scientific), PE anti-mouse CD4 (Thermo Fisher Scientific), PE-Cy7 anti-mouse TCR-β (BioLegend), FITC anti-mouse CD25 (BioLegend), FITC anti-mouse CD11b (BioLegend), PE anti-mouse CD11b (Miltenyl Biotec), FITC anti-mouse Ly6G (BioLegend), APC anti-mouse Ly6C (BioLegend), APC anti-mouse CD86 (BioLegend), PE anti-mouse CD206 (BioLegend), and PE-Cy7 anti-mouse F4/80 (Thermo Fisher Scientific) antibodies were used. The cells were suspended in FACS buffer (PBS containing 1% v/v FBS) and incubated with antibodies for 30 min at 4 °C. The Pacific Blue anti-mouse lineage cocktail was added to cells at a ratio of 1:10. Unless mentioned otherwise, the antibodies for surface marker staining were added to samples in the ratio of 1:100. For intracellular staining, single-cell suspensions from tumors, spleen, lymph nodes or BMDMs were first stained with antibodies against cell surface markers as described above, and then fixed and permeabilized using a BD Cytofix/Cytoperm Fixation/Permeabilization kit (BD) according to the manufacturer’s protocol and then stained with antibodies. APC anti-mouse IL-6 (BioLegend), PE anti-mouse IL-1α (BioLegend), APC anti-mouse IL-1β pro-form (Miltenyi Biotec), and APC anti-mouse TNF-α (BioLegend) antibodies were added at a ratio of 1:100 and incubated for 30 min at 4 °C before analysis. Immunoassays The serum concentrations of mouse IL-6, IL-1β and TNF-α were measured using mouse Duoset ELISA kits (R&D Systems) according to the manufacturer’s instructions. For ex vivo restimulation of splenic macrophages, lymph node-derived macrophages and BMDMs, cells were restimulated with MC38 cells at a ratio of 10:1 or with 250 ng/ml Pam3CSK4 (P3C; InvivoGen); then culture supernatants were collected for quantification of IL-6 and IL-1β concentrations using mouse ELISA kits (R&D Systems). For restimulation of in vitro-trained BMDMs, cells were challenged with MC38 or LLC cells at a ratio of 10:1 or with 250 ng/ml P3C; then, culture supernatants were collected for quantification of IL-6 and IL-1β concentrations using mouse Duoset ELISA kits (R&D Systems). A LEGENDplex Mouse Macrophage/Microglia Panel (BioLegend) was used to measure the levels of CXCL1, TNF-α, Ilβ, IL-6, IL-12p70, IL-12p40, IL-18, IL-23, TGF-β1, G-CSF, CCL17, CCL22, and IL-10 according to the manufacturer’s instructions using culture supernatant of naïve or trained BMDMs stimulated with MC38 for 16 h. Data analysis was performed using LEGENDplex data analysis software. Depletion of macrophages The mice were trained with either PBS or LPS on day 0 before macrophage depletion. Macrophages were depleted by i.p. injection of 200 μl of Clodronate Liposomes (Liposoma) per mouse on day 2 and 5. Validation of macrophage depletion was performed by analyzing splenic and peritoneal macrophage populations on day 7 using flow cytometry. Alternatively, mice were challenged with a subcutaneous injection of 1 × 106 MC38 cells on day 7, and tumor growth was monitored with a caliper for 21 days. Phagocytosis assay A flow cytometry-based phagocytosis assay was performed as previously described75 with some modifications. In brief, MC38 or LLC carcinoma cells were labeled with 1 μM of CFSE (Invitrogen) for 15 min at 37 °C in the dark and then washed twice with complete medium to remove excess dye. CFSE-stained cells were added to naïve or LPS-trained BMDMs. After 3 h of coculture with gentle mixing every hour, the cell mixtures were stained with anti-CD11b-PE (BioLegend) and anti-F4/80-PE-Cy7 antibodies (BioLegend) for the identification of the macrophage population and analyzed by flow cytometry. Phagocytosis efficiency was determined as the percentage of macrophages exhibiting FITC fluorescence. In vitro macrophage training Differentiated BMDMs were treated with LPS (2.5 ng/ml) or PBS overnight at 37 °C. The macrophages were then washed twice with DPBS. RPMI-1640 supplemented with 10% M-CSF-conditioned media was then added, and cells were allowed to rest until day 6. During this period, fresh media with M-CSF was added every 3 days. On day 6, the macrophages were restimulated with different stimuli, including MC38 or LLC cells or Pam3CSK (P3C; 250 ng/ml; InvivoGen; tlrl-pms). In some experiments, P3C, poly(I:C) (InvivoGen; tlrl-picw), and R848 (Resiquimod; MedChemExpress; HY-13740) were used as training stimuli, which were all reconstituted in sterile endotoxin-free physiological water. The culture supernatants were collected for cytokine analysis by mouse DuoSet ELISA kits (R&D Systems), or the cells were lysed for RNA extraction for further experiments. In some experiments, cells were pretreated with the TLR4 inhibitor TAK-242 (Resatorvid, 10 nM; MCE), NF-kB inhibitor JSH-23 (10 μM; MCE), or JAK1/2 inhibitor Ruxolitinib (2 μM; MCE) 1 h before LPS treatment. In some other experiments, during the resting phase, BMDMs were treated with the SLC1A5_var inhibitor iMQT_020 (5 μM), the Gls1 inhibitor CB839 (Telaglenastat, 2 μM; MCE), the Glud1 inhibitor R162 (10 μM; MCE), the pan-KDM5 inhibitor CPI-455 (20 μM; MCE), the KDM5B inhibitor GSK467 (20 μM; MCE) or DM-KG (4 mM; Thermo Fisher). Macrophage adoptive transfer models Macrophage adoptive transfer was performed as previously described20,76. In brief, BM cells obtained from donor mice were differentiated and then trained with PBS or LPS in vitro. On day 6 of training, BMDMs were mixed with MC38 cells at a ratio of 1:3, and 1 × 106 total cells were injected subcutaneously into new recipient mice. In some experiments, to test the consequence of manipulating the metabolic–epigenetic regulatory axis by various reagents and inhibitors, LPS-stimulated BMDMs were supplemented with 4 mM DM-α-KG (Thermo Fisher Scientific) and/or 20 μM GSK467 (MCE) during the resting phase. RNA extraction and real-time PCR Total RNA was extracted using a MiniBEST Universal RNA Extraction Kit (Takara) according to the manufacturer’s instructions. RNA was quantified using a NanoDrop spectrophotometer (Thermo Fisher Scientific). Reverse transcription of the total RNA was carried out using the TransScript First-Strand cDNA Synthesis Kit (TransGen). The complete reaction mix was first incubated at room temperature for 10 min and then incubated at 42 °C for another 15 min. The reverse transcriptase was inactivated at 95 °C for 30 s. Polymerase chain reaction was performed using DiaStar Taq DNA polymerase (Solgent) with EvaGreen (Biotium), and gene expression levels were analyzed with the Step One Plus system (Applied Biosystems). The expression of each gene was defined from the threshold cycle (Ct), and the relative levels were calculated using the 2−ΔΔCt method and normalized to the expression levels of Actb. Sequences for all primers used in this study are listed in Supplementary Table 1. RNA-seq Total RNA was extracted with a Universal RNA extraction kit (Takara) according to the manufacturer’s protocol, and quality was assessed with the Agilent 4200 TapeStation System (Agilent Technologies). RNA was quantified using the Invitrogen Qubit 4 or Qubit Flex system (Thermo Fisher). The Poly(A) RNA Selection Kit (LEXOGEN, Inc., Austria) was used for mRNA isolation, followed by library preparation using the CORALL RNA-Seq V2 Library Prep Kit (LEXOGEN). Indexing was performed using Lexogen UDI 12-nt Unique Dual Indexing V2. PCR was carried out for the enrichment step. Libraries were subsequently checked using the TapeStation4200 system (Agilent Technologies, Amstelveen) or Agilent 2100 bioanalyzer (Agilent Technologies) to evaluate the mean fragment size. Quantification was performed using a library quantification kit and the Invitrogen Qubit Flex system (Thermo Fisher). High-throughput sequencing was performed as paired-end 150 sequencing using NextSeq 2000 (Illumina). Quality control of the raw sequencing data was performed using FastQC. Adapter and low-quality reads were removed using Fastp. Then, the trimmed reads were mapped to the reference genome using STAR.77 Quantification of the reads was performed using Salmon. Data mining and graphic visualization were performed using ExDEGA (Ebiogen). GSEA was used to analyze DEGs in naïve and LPS-stimulated BMDMs. LC‒MS-based metabolomics and quantification of metabolite abundance Whole-cell metabolite profiling was performed with 5 × 106 naïve or LPS-trained BMDMs for nontargeted metabolomic analysis (n = 3 biological replicates). For metabolic flux analysis, cells were treated with uniformly labeled 13C-glutamine or 15N-glutamine in glutamine-free medium or uniformly labeled 13C-glucose in glucose-free medium for 6 h (n = 3 biological replicates). At collection, the cells were placed immediately on ice and washed with ice-cold DPBS two times and once with Optima LC‒MS water to remove residual medium. Then, the cells were scraped into 1 ml of 80% Optima LC‒MS methanol in water and vortexed vigorously for 1 min. The samples were then centrifuged at 17,500 × g for 20 min, and the supernatants were analyzed by LC‒MS. The metabolite extract was subsequently diluted with 20% (v/v) acetonitrile for ultra-performance liquid chromatography‒triple quadrupole mass spectrometry (UPLC‒TQ‒MS). The Ultimate 3000 RSLC instrument equipped with an Acquity UPLC BEH C18 column (1.7 μm, 2.1 × 100 mm) and Q Exactive Orbitrap Plus MS system (Thermo Fisher) was used for analysis. Compound Discoverer (v3.0; Thermo Fisher) software was used for data analysis. MS was performed with the following settings: gas temperature, 37 °C; isolation window, 2.0 m/z; collision energy, 30; dynamic exclusion, 5 s; sheath gas flow rate, 50; aux gas flow rate, 13; and spray voltage, 2.5 kV. Peaks were quantified via the area under the curve. The fold change in expression in LPS-trained versus naïve BMDMs was calculated. Metabolic pathway enrichment analysis was performed using MetaboAnalyst 6.0 software. H2DCFDA and MitoSOX assays Intracellular ROS levels were determined by staining cells with 1 mM H2DCFDA (Invitrogen) according to the manufacturer’s instructions and then analyzing cells via flow cytometry. MitoROS levels were determined by staining cells with MitoSOX Red (Invitrogen) at 5 mM for 10 min followed by flow cytometry. Changes in the mitochondrial membrane potential were analyzed by staining cells with tetramethylrhodamine ethyl ester (TMRE, 20 nM; Invitrogen) followed by flow cytometry. All data were analyzed using FlowJoTM Software v10.10 (BD Life Sciences). Seahorse XF assay The OCRs and ECARs of the macrophages were determined with an XFe24 extracellular flux analyzer (Agilent Technologies) as previously described78. In brief, in vivo-or in vitro-trained BMDMs or naïve BMDMs were seeded at 1.0 × 105 cells per well in 24-well micro cell culture plates for Seahorse analysis (Agilent Technologies); unbuffered RPMI medium supplemented with 2 mM L-glutamine was added 1 h before the assay. The Seahorse XF Cell Mito Stress Test Kit (Agilent Technologies) and Seahorse XF Glycolysis Stress Test Kit (Agilent Technologies) were used according to the manufacturer’s instructions, and the final concentrations of oligomycin, FCCP, and antimycin A/rotenone were 1.5 μM, 3.5 μM and 1.5 μM, respectively. In some experiments, glucose (2.5 mM), DM-KG (2 mM or 4 mM), DE-succinate (3 mM), and DM-fumarate (100 μM) were added 1 h before the assay. Each measurement cycle consisted of 2 min of mixing, 2 min of waiting, and 4 min of measurement. The OCRs and ECARs were normalized to the number of cells, and bioenergetics parameters were then measured according to the manufacturer’s instructions. Western blotting Cells were washed once with DPBS and then lysed using a RIPA lysis kit (ATTO) containing a protease inhibitor cocktail and phosphatase inhibitor, agitated for 20 min at 4 °C, and subsequently sonicated on ice. Unless mentioned otherwise, samples used for immunoblot analysis of SLC1A5_var were not heated to more than 40 °C or frozen. The protein concentration was measured with a BCA assay kit (Intron). A total of 20 μg of protein was subjected to SDS‒PAGE and transferred to a polyvinylidene difluoride membrane (Merck) using the Trans-BlotR Turbo Blotting System (Bio-Rad). The membrane was blocked in 5% (w/v) BSA or 2% (w/v) skim milk in Tris-buffered saline-Tween for 30 min and incubated with the appropriate primary antibodies for 18 h. The membrane was subsequently incubated with the appropriate HRP-conjugated secondary antibodies (anti-mouse or anti-rabbit IgG conjugated with HRP; Invitrogen) for 1 h. The immunoblot signals were detected with a Fusion Solo S instrument (Vilber) with enhanced chemiluminescence reagents (EzWestLumi, ATTO) and quantified via densitometric analysis of the protein bands with ImageJ. The immunoblots are representative of three independent experiments. Unprocessed original blots are provided in Source Data. Purification of mitochondria from cells and mitochondrial amino acid uptake assay Mitochondria were prepared in accordance with methods previously described by Wieckowski et al., with slight modifications79. A total of 5 × 106 naïve or LPS-trained BMDMs were washed twice with DPBS and then gently scraped into 1 mL of KPBS. The cells were subsequently centrifuged at 900 × g for 3 min, after which the pellet was resuspended in 1 mL of KPBS. The cell mixture was carefully homogenized with 50 strokes in a 2-mL Dounce homogenizer to avoid air bubbles. The homogenate was subsequently centrifuged at 1000 × g for 2 min. Mitochondrial amino acid uptake was measured by first diluting the purified mitochondria in 50 μL of KPBS. Radiolabeled glutamine (3H-glutamine) (PerkinElmer) (3 μCi/mL) was added to the samples, which were subsequently incubated at 37 °C for 1 h. Uptake was stopped by incubating the tubes at 4 °C. The samples were then centrifuged at 12,000 × g for 15 min, after which the pellets were washed twice with KPBS. Lysis buffer (0.2% SDS) was used for extraction and mixed with 3 mL of AquaLight+ (HIDEX). All measurements were analyzed by a liquid scintillation counter (HIDEX). The mitochondrial protein concentration of each sample was subsequently measured with a BCA assay kit (Intron). Amino acid uptake was calculated from the counts per minute per sample and normalized to the mitochondrial protein content. Migration assay Transwell migration assays were performed using Transwell inserts with filter membranes with a pore size of 8 μM (SPL). For the T-cell migration assay, CD8+ T cells were sorted from splenocytes isolated from C57BL/6 mice using a MACS cell separation system (Miltenyi Biotec) with anti-CD8 MicroBeads (Miltenyi Biotec). CD8+ T cells were activated with anti-CD3 and anti-CD28 antibodies and recombinant IL-2 and seeded into the upper chamber (RPMI-1640 containing 0.5% FBS), whereas different groups of BMDMs and MC38 cells were seeded in the lower compartment (RPMI-1640 containing 10% FBS). After 24 h, the migrated cells in the lower compartment were analyzed by flow cytometry. For the macrophage migration assay, BMDMs from different groups were seeded into the upper chamber with RPMI-1640 medium supplemented with 0.5% FBS, and MC38 cells were seeded in the lower compartment with RPMI-1640 medium supplemented with 10% FBS. After 24 h, the migrated cells in the lower compartment were analyzed by flow cytometry. Expression constructs and transfection Mouse p65, p50, and KDM5B cDNA were obtained from the Korea Human Gene Bank (Medical Genomics Research Center, KRIBB), and Gls1 cDNA was purchased from Addgene. For expression studies, the coding sequences of mouse SLC1A5_var and Glud1 were cloned and inserted into the pCMV SPORT6 vector. To generate the WT SLC1A5_var promoter construct with NF-κB-RE (pVar) or without NF-κB-RE (pVar_s) for the transcriptional activity assay, the 5’ upstream region (957 bp or 500 bp) of the SLC1A5_var locus from RAW264.7 cells was cloned and inserted into the pGL4.1 vector. The SLC1A5_var mutant NF-κB-RE promoter was generated using a site-directed mutagenesis kit (NEB). Plasmids were transfected using TurboFect (Thermo Fisher Scientific) or subjected to electrophoresis using a Neon NxT Electroporation system (Invitrogen). siRNA-mediated knockdown of KDM5A, KDM5B and KDM5C in trained BMDMs was performed 24 h after the initiation of the resting period. Cells were transfected with siRNAs at a final concentration of 50 nM using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer’s instructions. After 24 h, the medium was replaced with fresh RPMI-1640 supplemented with M-CSF-conditioned media from L-929 culture, and cells were cultured until day 6 for downstream assays. Knockdown efficiency was confirmed by qPCR and immunoblotting. Nucleotide sequences are provided in Supplementary Table 2. Lentiviral production and cell transduction shRNA sequences specifically targeting mouse Slc1a5_var were cloned and inserted into pLKO.1 and pSICO plasmids (Addgene) following the manufacturer’s instructions. HEK293T cells were transfected with pLKO.1-shSLC1A5_var#1 or pLKO.1-shSLC1A5_var#2 with packaging and envelope plasmids (Addgene) using TurboFect reagent. Virus-containing supernatants were harvested after 48 h. For knockdown of Slc1a5_var, lentiviral transduction was performed during BMDM differentiation. Bone marrow cells were transduced on day 4 of differentiation in the presence of 8 μg mL−1 polybrene. After 12 h, the medium was replaced with fresh RPMI-1640 with M-CSF-conditioned media, and cells were further differentiated and subjected to training and downstream analyses. Target sequences for shRNAs are provided in Supplementary Table 2. Dual-luciferase assay Cells plated in 12-well plates were cotransfected with 0.9 μg of the pGL4.10 firefly luciferase construct containing the SLC1A5_var promoter and 0.1 μg of the pGL4.7 control Renilla construct (Promega) using TurboFect (Invitrogen). After transfection, the cells were incubated for the desired time or under the desired condition, and the luciferase activity was measured using the Dual-Luciferase® Reporter Assay System (Promega) and a microplate reader (Infinite 200 PRO, TECAN). Firefly luciferase activity was normalized to Renilla luciferase activity. Analysis of the histone modification markers via flow cytometry Analysis of histone modification was performed as previously described80,81. In brief, cells were first stained with antibodies against surface markers as described earlier, and then fixed and permeabilized using a BD Cytofix/Cytoperm Fixation/Permeabilization kit (BD) according to the manufacturer’s protocol. The cells were then incubated with primary monoclonal antibodies (anti-H3K4me1, anti-H3K4me3, or anti-H3K27Ac) for 1 h at 4 °C, washed with permeabilization buffer, and incubated with Alexa Fluor 488-conjugated goat anti-rabbit IgG (Invitrogen) as a secondary antibody for 45 min at 4 °C. Chromatin immunoprecipitation assay A total of 5 × 106 cells were fixed with 1% formaldehyde for 15 min at room temperature, and the reaction was quenched with glycine (125 mM). Cells were then sonicated to shear the chromatin into fragments of 200~800 bp and subjected to immunoprecipitation using antibodies against control IgG, H3K4me3, and H3K27Ac (Abcam) complexed with protein A/G agarose beads. Decrosslinking of immunoprecipitated DNA was performed with overnight heating at 65 °C followed by incubation with proteinase K at 45 °C for 1 h. Sheared chromatin that was not immunoprecipitated was subjected to the same procedure to obtain a 1% input control. Eluted DNA was processed further for qPCR analysis using primers specific to the promoter regions of genes of interest. ChIP-seq Approximately 5 × 106 BMDMs were fixed, and the nuclei were collected, after which the chromatin was isolated and immunoprecipitated using an anti-histone H3 (trimethylated K4) antibody (Abcam) and a high-sensitivity ChIP kit (Abcam) according to the manufacturer’s instructions. Purified DNA was subjected to library construction using the NEBNext® UltraTM DNA Library Prep Kit for Illumina (New England Biolabs). In brief, the chipped DNA was ligated with adapters for PCR with index primers for multiplexing sequencing. High-throughput sequencing was performed as paired-end 150-bp sequencing using NovaSeq X Plus (Illumina). Raw sequencing reads were filtered using fastp v0.23.1 and then aligned to the reference genome using Bowtie2 (v 2.3.4.3). Peak calling was performed using MACS2 (v2.2.8), and the ChIPseeker R package (v1.36.0) was used for peak annotation with genomic features. Specific peaks were visualized using IGV (v.2.11.4). ATAC-seq Library construction was performed with the ATAC-Seq Library Prep Kit for Illumina (Active Motif). Approximately 1 × 105 fresh cells were washed twice with ice-cold PBS and resuspended in 100 µl of lysis buffer for 10 min to extract the nuclei. The nuclei were then incubated with Tagmentation Master Mix for 30 min at 37 °C. PCR was performed to amplify the library, and then SPRI beads (Beckman Coulter) were used to clear the contaminating primer dimers. A library quantity check was performed with TapeStation HSD5000 Screen Tape (Agilent Technologies). All libraries were sequenced on the NovaSeq XP system with 150-bp paired-end reads (Illumina). Raw sequencing reads were trimmed using fastp (v0.23.1) and then aligned to the reference genome using Bowtie2 (v.2.3.4.3). Picard (v2.27.5) was used to improve data quality by removing duplicate PCR reads and reads aligning to low mapping quality. Read coordinates were adjusted with the alignmentSieve function from deepTools to account for Tn5 transposase insertion bias. Peak calling was performed using MACS2 (v2.2.8), and the ChIPseeker R package (v1.36.0) was used for peak annotation with genomic features. For differential accessibility analysis, the edgeR package (v3.42.4) was used, and TMM-normalized log2(CPM + 1) transformation was applied. Homer (v4.11.1) was used for motif enrichment analysis to identify binding motifs with accessible chromatin regions for transcription factors. Quantification of intracellular metabolite levels The concentrations of intracellular metabolites were determined with an α-KG assay kit (Abcam), a glutamine/glutamate-Glo assay (Promega), a succinate assay colorimetric assay kit (BioVision) and a fumarate assay kit (BioVision) in accordance with the manufacturer’s instructions. A microplate reader (Infinite 200 PRO, TECAN) was used for analysis, and the absorbance or luminescence of each control well was measured for background signal subtraction. Statistical analysis Statistical analyses were performed with GraphPad Prism 8 software. Sample normality was analyzed using unpaired two-tailed Student’s t-test and one-way analysis of variance (ANOVA) with Tukey or Dunnett post hoc test. A P-value < 0.05 was considered to indicate significance. All data in graphs are presented as the mean ± SEM. Biological replicates were included for each experiment. For all animal studies, the experiments were replicated across multiple batches. Age-matched and sex-matched mice were randomly assigned to groups. No statistical method was used to predetermine the sample size. No data were excluded from the analyses. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary information Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgements The authors thank the staff of the Yonsei Center for Research Facilities (YCRF) for technical assistance. The authors also thank the Yonsei Institute of Pharmaceutical Sciences for its support. Author contributions Y.C.Y., conceptualization, experimental design, collection and analysis of the data and manuscript writing; Y.S., experimental design, collection and analysis of the data; S.L., J.A.K., and M.J.L., collection and analysis of the data; H.Y.K., experimental design, collection of the data; S.K., K.K., and D.S.M., supervision; H.C.Y., conceptualization and supervision; J.M.H., supervision and final approval of the manuscript. Peer review Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Funding This work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF), funded by the Korean government (MSIT) (RS-2023-00219297, RS-2025-00573098, RS-2025-18362970, RS-2026-25510253, J.M.H.). This research was also supported by the Korea Drug Development Fund funded by the Ministry of Science and ICT; the Ministry of Trade, Industry, and Energy; and the Ministry of Health and Welfare (RS-2024-00334337, J.M.H.). Data availability Source data are provided with this paper. All sequencing data generated in this study have been deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under BioProject accession numbers PRJEB104000, PRJEB110966, and PRJEB111004. All LC-MS metabolomics data generated in this study are available in the MassIVE database under accession code MSV000100549 (https://doi.org/10.25345/C5KH0FC5P). Source data are provided with this paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76757-0. References Associated Data Supplementary Materials Supplementary information Reporting Summary Transparent Peer Review file Source Data Data Availability Statement Source data are provided with this paper. All sequencing data generated in this study have been deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under BioProject accession numbers PRJEB104000, PRJEB110966, and PRJEB111004. All LC-MS metabolomics data generated in this study are available in the MassIVE database under accession code MSV000100549 (https://doi.org/10.25345/C5KH0FC5P). Source data are provided with this paper.