Histone H4K5 lactylation-driven feedback loop modulates glycerol-linked immunosuppressive activity of tumor-associated T regulatory cells

Abstract

Tumor metastasis is the leading cause of death for more than 90% of cancer patients, with colorectal cancer (CRC) mainly spreading to the liver. The colon and liver possess distinct nutritional microenvironments, yet how these differences shape the tumor immune landscape remains unclear. Using single-cell analysis of autologous samples from CRC patients with liver metastases, we identified a distinct CD177⁺GK⁺ regulatory T (Treg) cell subset in metastatic lesions. Spatial transcriptomics revealed that these GK⁺ Tregs localize at the tumor-adjacent interface and interact with immunosuppressive macrophages. Moreover, a selective clonal selection was observed in the metastatic niche, leading to the super-amplification of metastasis-associated clones of GK⁺ Tregs, suggesting a dominant role of these clones in metastatic immune evasion. Mechanistically, glycerol-derived lactate promotes H4K5 histone lactylation, which upregulates FOXP3 and subsequently GK expression, creating a positive feedback loop that amplifies Treg immunosuppressive activity. Blocking the glycerol metabolic pathway in Tregs combined with anti-PD-1 therapy synergistically curbs tumor progression. Pan-cancer analyses further confirm the potent suppressive activity of CD177⁺GK⁺ Tregs across multiple cancer types. Collectively, our findings highlight glycerol metabolism and histone lactylation as critical drivers of Treg-mediated immune evasion, offering new strategies to reprogram the immunosuppressive tumor microenvironment for enhanced cancer therapy.

Introduction

Colorectal cancer (CRC) is a leading cause of cancer-related mortality worldwide, with a particularly high incidence in developed countries, and frequently metastasizes to the liver, which is the most common site of distant spread. Upon colonization, the tumor microenvironment (TME) within the liver evolves into a highly immunosuppressive niche that not only supports tumor growth but also contributes to therapeutic resistance, including resistance to immune checkpoint blockade.^1,2 Despite surgical resection of primary tumors, distant recurrences—particularly in the liver—remain a major clinical challenge, as they account for the majority of CRC-related deaths.³ The immunosuppressive nature of the metastatic TME is shaped by dynamic interactions among tumor cells, stromal components, and various immune populations, creating a complex network of pro-tumor and anti-tumor signals.⁴ Of these, regulatory T cells (Tregs) play a dual role: they are essential for maintaining immune homeostasis and preventing autoimmunity, but in the context of cancer, they serve as central mediators of immunosuppression, dampening antitumor immune responses.^5–9

Tregs exhibit a remarkable ability to thrive in metabolically hostile TMEs, which are characterized by hypoxia, acidosis, and intense nutrient competition—conditions that typically impair effector T cell function and induce apoptosis.^10–12 Paradoxically, these harsh conditions appear to enhance Treg survival and functional activity, allowing them to outcompete other immune cells. This metabolic flexibility enables Tregs to utilize alternative energy sources such as lactate, a byproduct of aerobic glycolysis, to reinforce their immunosuppressive capacity.^13–15 Although lactate is known to inhibit CD8⁺ T cell cytotoxicity and promote pro-tumor macrophage polarization,^16,17 emerging evidence suggests it may also act as a signaling molecule that modulates epigenetic programs through lysine lactylation, a recently identified form of post-translational modification that is distinct from acetylation and methylation.^18–20 In particular, histone lactylation has been implicated in metabolic adaptation and immune cell plasticity under various pathological conditions, including cancer and inflammation.^19,21–24 Despite these advances, it remains unclear how tissue-specific nutrient availability—such as the high glycerol and lactate levels characteristic of the liver—modulates Treg function through metabolic and epigenetic reprogramming in the metastatic niche. Moreover, the potential role of lactylation in fine-tuning Treg stability and immunosuppressive activity within metastatic sites is poorly understood, leaving a critical gap in our knowledge of immune evasion. Therefore, elucidating the mechanistic basis of lactylation-driven epigenetic regulation in Tregs may help identify novel therapeutic targets. Such insights could facilitate the development of molecularly targeted strategies aimed at disrupting immunosuppressive niches and improving the efficacy of existing immunotherapies, particularly for patients with liver metastases.

In this study, we integrate single-cell transcriptomics (scRNA-seq) and functional assays to identify a distinct CD177⁺ Treg subset that is markedly enriched in metastatic lesions of colorectal liver metastasis (CRLM) patients. This subpopulation demonstrates heightened expression of glycerol kinase (GK), a key enzyme in glycerol metabolism, along with classical immunosuppressive markers such as PDCD1 and CTLA4, indicating a potent regulatory phenotype. Using spatial transcriptomics, we found that these Treg cells are predominantly localized at the interface between the tumor and adjacent non-cancerous tissues, a region that may be associated with tumor invasion and metastatic spreading. Moreover, these Treg cells interact with multiple immune cell types, including CXCL12⁺ macrophages, which may serve as one of the contributing factors to the complex and dynamic tumor immunosuppressive microenvironment, potentially through chemokine-mediated crosstalk. Furthermore, in GK⁺ Treg cells with robust glycerol metabolism, certain T-cell clones—most likely targeting metastasis-associated antigens—undergo extreme clonal expansion, gaining a marked growth advantage in the metastatic microenvironment. We show that under low-glucose conditions, these Tregs enhance glycerol metabolism, leading to increased lactate production and subsequent histone H4K5 lactylation, which in turn reinforces FOXP3 binding to the GK promoter. This establishes a positive feedback loop that amplifies and sustains the Treg suppressive phenotype, thereby perpetuating local immunosuppression. Importantly, combination therapy targeting aquaporin-3 (AQP3), a glycerol channel, and anti-PD-1 synergistically reduces tumor growth in MC38 tumor-bearing mice, underscoring the therapeutic potential of disrupting this metabolic-epigenetic pathway. Furthermore, CD177⁺GK⁺ Tregs display a robust immunosuppressive molecular profile across multiple cancer types, indicating a pattern that may be relevant for future translational studies and perhaps serving as a biomarker for patient stratification.

Overall, despite advances in immune checkpoint blockade (ICB), most “immune-cold” tumors—including CRLM—remain refractory to treatment.²⁵ The metabolic adaptability of Tregs, together with tissue-specific nutrient availability, represents a major obstacle to effective immunotherapy, as it enables Tregs to persist and function even under nutrient-deprived conditions. Our work delineates a glycerol-driven metabolic–epigenetic circuit centered on H4K5 lactylation that reinforces Treg-mediated immunosuppression in CRLM, providing a mechanistic framework for understanding how local metabolism shapes immune evasion. These findings not only uncover a context-specific mechanism of immune evasion but also propose innovative combinatory strategies to reprogram the TME and enhance ICB efficacy. This study highlights the critical roles of metabolic flexibility and epigenetic plasticity in cancer progression and provides a strategic roadmap for targeting immune dysfunction in metastatic disease.

Results

Enhanced Suppressive Function of Tregs in Metastatic Niches Is Strongly Associated with Glycerol Kinase Expression

To investigate the immune composition differences between primary colorectal tumors (Pri.CT) and liver metastases (Liv.Mets) in colorectal liver metastasis (CRLM) patients, we analyzed the single-cell data²⁶ from 7 paired Pri.CT and adjacent non-tumor colorectal tissues (Adj.Col) from patients with liver metastasis (E-MTAB-8107), and 10 paired Liv.Mets and adjacent non-tumor liver tissues (Adj.Liv) from 10 CRLM patients (GSE164522) (Fig. 1a), adjacent non-tumor tissues were defined as adjacent normal tissue samples at least 2 cm from their matched tumor tissues. We observed that CD4⁺ cells and Tregs were significantly more abundant in both primary tumors and liver metastases compared to their adjacent tissues, with particularly high infiltration in liver metastatic lesions (Fig. 1b and Supplementary Fig. 1a-c). These Tregs exhibited elevated expression of immunosuppressive markers such as PDCD1 and CTLA4 in liver metastases (Fig. 1c), suggesting that the metastatic microenvironment may enhance Treg suppressive activity. This also implies that the subpopulation of tumor-infiltrating Treg (TI-Treg) cells in liver metastases is likely enriched by the malignancy status. Given that Tregs are known to adapt to nutrient-deprived tumor microenvironments by utilizing alternative energy sources,^13,27,28 we hypothesized that Tregs in liver metastases might exploit unique metabolic pathways available in the liver.

Fig. 1.

Enhanced suppressive function of Tregs in metastatic niches is strongly associated with glycerol kinase expression. a Study design. scRNA-seq datasets were collected from matched samples of CRLM patients including 7 pairs of primary colorectal tumors (Pri.CT) and adjacent non-tumor colorectal tissues (Adj.Col) from patients with liver metastasis (E-MTAB-8107), and 10 pairs of liver metastases (Liv.Mets) and adjacent non-tumor liver tissues (Adj.Liv) from 10 CRLM patients (GSE164522). Figure1a was created using BioRender (https://BioRender.com/b7u86pj). b Abundance of Tregs in Pri.CT and Liv.Mets compared to their respective adjacent tissues (Adj.Col, Adj.Liv). Significantly higher infiltration was observed in Liv.Mets. c Elevated expression of immunosuppressive markers (PDCD1, CTLA4) in Tregs from Liv.Mets compared to Pri.CT. d GSVA analysis revealed significant upregulation of glycerol metabolism pathways (glycerophospholipid and glycerolipid metabolism) in Liv.Mets versus Pri.CT. e Volcano plot showing differentially expressed genes between Tregs from Pri.CT and Liv.Mets. Genes with adjusted p-value < 0.01 (Wilcoxon rank-sum test) and log2(fold change) > 0.3 were color-coded. f Enhanced activity in glycerophospholipid and glycerolipid metabolism pathways in GK⁺ Tregs. g Increased expression of canonical immunosuppressive markers in GK⁺ Tregs. h The spatial expression of EPCAM, GK, Treg, and GK⁺ Treg in Pri.CT(up) and Liv.Mets(down). i Comparison of the abundance of GK⁺ Treg between Pri.CT(CRC) and Liv.Mets(CRLM) patients using a spatial transcriptomics public dataset. See also Supplementary Fig.1

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses revealed that, compared to Treg cells in Pri.CT, two glycerol metabolism pathways, glycerophospholipid and glycerolipid metabolism, were the two most significantly upregulated metabolic pathways in Treg cells from liver metastatic sites. However, in colorectal tissues, no upregulation of these two metabolic pathways was observed in Treg cells from tumors compared to adjacent non-cancerous tissues. In contrast, in liver tissues, these pathways were upregulated in Treg cells from tumors compared to adjacent non-cancerous tissues. This suggests that the unique metabolic environment of the liver becomes more pronounced after tumor development, enabling Treg cells in this site to more efficiently utilize glycerol metabolic pathways (Fig. 1d and Supplementary Fig. 1d, e). Glycerol, a key intermediate in carbohydrate and lipid metabolism, is abundant in the liver, where it serves as a gluconeogenic precursor and is used for triglyceride synthesis. Indeed, we noted that glycerol concentrations were significantly higher in the liver than in the colon in mice (Supplementary Fig. 1f). To identify the key regulators of glycerol metabolism in Tregs, we compared the gene expression data using Tregs from Pri.CT and Liv.Mets in CRLM patients, which highlighted glycerol kinase (GK) as a significantly upregulated gene in liver-metastatic Tregs (Fig. 1e). GK catalyzes the phosphorylation of glycerol to glycerol-3-phosphate (Gro3P), a critical step in glycerol metabolism (Supplementary Fig. 1g). Notably, GK is highly expressed in Treg cells within liver metastases (Supplementary Fig. 1h). Further analysis revealed that GK⁺ Tregs exhibited enhanced activity in glycerol metabolism pathways and stronger immunosuppressive capabilities (Fig. 1f,g and Supplementary Fig. 1i). Colorectal cancer liver metastases display a specific landscape of T cell infiltration patterns in and around the tumor.^29–32 Treg cells also exhibit a high degree of spatial consistency with these patterns. Van den Eynde et al. demonstrated marked heterogeneity in FOXP3⁺ Treg density not only between different metastases from the same patient but also across regions within a single metastasis^.29 Using spatial transcriptomics, we examined the distribution characteristics of GK⁺ Treg cells between the two tissue compartments. Interestingly, we found that these cells were abundant not only inside the tumor but also at the tumor-adjacent interface (Fig. 1h). Moreover, the abundance of these cells was higher in metastatic lesions than in matched primary tumors (Fig. 1i). Published studies have shown that Tregs at the invasive margin are highly activated and proliferative.^33,34 This raises the intriguing possibility that in the nutrient-poor, metabolically hostile tumor microenvironment, GK⁺ Tregs may self-supply energy by exploiting locally available alternative fuels, thereby reinforcing an immune-cold environment at the invasive front and potentially facilitating tumor invasion.

Elevated Glycerol Metabolism and Potent Suppressive Activity of CD177⁺ TI-Tregs in Metastatic Lesions

To further characterize Tregs within metastatic tumors, we analyzed scRNA-seq on Tregs isolated from primary and metastatic tumor sites. Unsupervised clustering identified four distinct Treg subsets (Fig. 2a, b), with Cluster 0 representing the predominant Treg subset in liver metastases and exhibited the highest expression of molecules associated with immunosuppression among Treg populations (Fig. 2c, d). Comparative analysis revealed that Cluster 0 had a pronounced metabolic phenotype (Supplementary Fig. 2a), and Cluster 0 isolated from metastatic tumors exhibited elevated activity in glycerophospholipid and glycerolipid metabolism pathways (Supplementary Fig. 2b). Moreover, genes associated with glycerol metabolism, particularly GK, were markedly upregulated in metastatic cluster 0 Treg subset (Fig. 2e, f). Furthermore, this subset exhibited enhancement in both glycerophospholipid and glycerolipid metabolism pathways (Fig. 2g). A strong positive correlation was noted between GK expression and signature functional molecules of tumor-infiltrating (TI) Treg cells such as CCR8 and CTLA-4 (Fig. 2h), suggesting that metabolic reprogramming reinforces Treg suppressive phenotype and activity. Interestingly, we observed that CCR8⁺GK⁺ and CTLA-4⁺GK⁺ dual-positive cells spatially co-localized with CD177 expression, predominantly within hepatic metastatic lesions (Fig. 2i, j), whereas CD177 is indeed a marker associated with TI-Tregs with potent immunosuppressive functions.³⁵ Recent studies have revealed that glycolysis in tumor cells is related to the suppressive function of intratumoral Tregs.¹³ We further analyzed the expression of glycolysis-related genes in tumor cells from four different sites: primary tumor-adjacent tissues, primary tumor tissues, metastatic tumor-adjacent tissues, and metastatic tumor tissues. We observed that these genes were significantly greater in metastatic tumor sites than in primary tumor and adjacent sites (Supplementary Fig. 2c), implying that the metastatic microenvironment could impose stronger metabolic stress on tumor cells. We asked whether HCC patients, who also possess a glycerol-rich microenvironment, have Tregs capable of utilizing glycerol as a nutrient source, similar to Tregs in the liver metastatic sites of CRLM patients. In this study, we sought to compare the expression of GK and CD177 in Tregs across distinct patient cohorts, including 10 cases of colorectal cancer (CRC), 5 cases of hepatocellular carcinoma (HCC), and a cohort of colorectal liver metastasis (CRLM) patients comprising primary tumor samples from 7 individuals and metastatic liver lesions from 10 patients. Notably, GK was predominantly expressed in Tregs from metastatic tumors of CRLM patients, whereas its expression remained low in primary hepatocellular carcinoma (HCC) lesions. Conversely, CD177 expression was largely restricted to a minor Treg subset within CRLM metastases, the vast majority of GK⁺ Treg cells are CD177 negative (Fig. 2k–m). This spatial heterogeneity may reflect heightened glycolytic activity in metastatic microenvironments-characterized by more severe metabolic stress and elevated lactate accumulation (Supplementary Fig. 2c). Based on the expression of CD177, we divided the Tregs in the metastatic tumors of CRLM patients into two subpopulations: CD177^-GK⁺ and CD177⁺GK⁺ (Supplementary Fig. 2d). The CD177⁺GK⁺ population exhibited enhanced metabolic capacity with concomitant enrichment in glycerophospholipid and glycerolipid metabolism pathways, along with elevated expression of immunosuppressive molecules, potentiating superior immunosuppressive capabilities (Supplementary Fig. 2e, f and Fig. 2n). We observed that in metastases, within GK⁺ Treg cells, the CD177⁺GK⁺ subset displays the strongest glycolytic capacity (Supplementary Fig. 2g). Given that glycolysis-derived lactate acts as a signaling molecule in the tumor microenvironment, it is possible that GK⁺ Treg cells can utilize the glycerol metabolism pathway to produce lactate. This lactate could potentially be transferred to neighboring cells via the lactate transporters MCT1/MCT4, allowing these Tregs to act as “niche” regulators. Through interaction analysis based on single-cell databases (GSE164522), we revealed that GK⁺ Treg subset had a strong interaction with CXCL12⁺ macrophages (Supplementary Fig. 2h). In the vast majority of solid tumors, CXCL12⁺ macrophage has been considered as a core pro-tumor and immunosuppressive cell subset.³⁶ Using spatial transcriptomics, we also showed that GK⁺ Tregs strongly interact with both CXCL12⁺ and SPP1⁺ macrophages (Supplementary Fig. 2i, j), resulting in aggravating immunosuppression. These results demonstrate that CD177⁺GK⁺ Tregs represent a metabolically active and immunosuppressive population critical for maintaining the suppressive microenvironment of metastatic tumors and potential therapeutic targets.

Fig. 2.

Elevated glycerol metabolism and potent suppressive activity of CD177⁺ TI-Tregs in metastatic lesions. a Unsupervised clustering of Tregs from primary and metastatic tumor sites identified four distinct Treg subsets. b Dot plot showing expression levels of specific genes for Treg subsets. Dot size indicates the fraction of expressing cells in each cluster, colored by normalized expression levels. c Cluster 0 was predominant in liver metastases. d Cluster 0 Tregs exhibited the highest expression of immunosuppression-associated molecules among Treg populations. e Marked upregulation of glycerol metabolism genes, particularly GK, in metastatic Cluster 0 Tregs. f Violin plot showing higher expression of GK in liver metastasis tumor. g Enhanced glycerophospholipid and glycerolipid metabolism pathways in Cluster 0 Tregs. h Strong positive correlation between GK expression and TI-Treg functional markers (CCR8, CTLA-4). i Co-localization of CCR8⁺GK⁺ and CTLA-4⁺GK⁺ dual-positive Tregs with CD177 expression within hepatic metastatic lesions. j Feature plot showing the expression of CD177 in GK⁺ Tregs in liver metastases. k UMAP visualization of Tregs from colorectal cancer (CRC, n = 10), hepatocellular carcinoma (HCC, n = 5), and colorectal liver metastasis (CRLM; primary tumors n = 7, metastatic lesions n = 10). l GK and CD177 were predominantly expressed in Tregs from CRLM metastatic tumors. m The proportion of Treg subpopulations including CD177^-GK^- Treg, CD177⁺GK^- Treg, CD177^-GK⁺ Treg, CD177 + GK+ Treg. n CD177⁺GK⁺Tregs exhibited higher expression of immunosuppressive genes. Data are derived from scRNA-seq and experimental validation. See also Supplementary Fig.2

Glucose Deprivation Promotes Glycerol Metabolism and Enhances the Suppressive Function of Tregs

To explore the roles and mechanisms of glycerol metabolism in Tregs, we first isolated natural Tregs (nTregs), naïve CD4⁺ T cells from FOXP3^GFP reporter mice, and induced Tregs (iTregs) in vitro as per the protocol we previously reported.³⁷ Similarly, human nTregs were sorted, and iTregs were induced from naïve CD4⁺ T cells isolated and prepared from peripheral blood as described in the protocol previously^38,39 (Supplementary Fig. 3a). Stimulation with glycerol significantly increased the expression and activity of GK, as well as the levels of glycerol-3-phosphate (Gro3P), in both mouse and human nTregs (Fig. 3a–c and Supplementary Fig. 3b, d, f). We also observed that in vitro culture of murine iTregs exhibited a progressive increase in glycerol utilization capacity over time, with human iTregs similarly demonstrating enhanced glycerol metabolism (Fig. 3d–f and Supplementary Fig. 3c, e, g). To further validate these findings, we treated mouse and human iTregs with a glycerol analog, which also resulted in decreased GK expression and activity, along with Gro3P levels (Fig. 3g–i and Supplementary Fig. 3h–j). To explore the mechanisms underlying glycerol uptake in Tregs, we analyzed RNA-Seq data from naïve CD4⁺ T cells and iTregs, and measured the expression of aquaporin (AQP) channel family members in iTregs. We found that Aqp3 was highly expressed in iTregs (Supplementary Fig. 3k). RT-PCR analysis confirmed that Aqp3 was indeed the highest-expressed aquaporin isoform across evaluated immune cell types (Supplementary Fig. 3l). Consistent with this, interrogation of the Protein Atlas database (www.proteinatlas.org) revealed broad AQP3 expression in diverse immune cells, with notably enriched abundance in Tregs cells (Supplementary Fig. 3m). We also compared the expression of Aqp3 among various CD4⁺ T cells and recovered that both iTregs and nTregs exhibited the highest expression, while naïve CD4⁺ T cells and effector T cells showed weaker expression (Supplementary Fig. 3n). Inhibition of aquaporin-3 (AQP3), a key glycerol uptake channel, significantly reduced glycerol uptake and downstream GK activity, further supporting the importance of glycerol metabolism in both mouse and human iTreg function (Fig. 3j–l and Supplementary Fig. 3o–q). We also revealed that the uptake and utilization of glycerol increased over time in cultured mouse iTregs (Fig. 3d–f), suggesting that nutrient depletion in the culture medium might enhance glycerol uptake. Given that glucose is the primary energy source for living cells, we hypothesized that Tregs might increase glycerol utilization under glucose-deprived conditions, similar to the nutrient-deprived tumor microenvironment. To test this, we established a low-glucose environment in vitro and noted that glucose deprivation significantly enhanced Gk expression and activity, as well as Gro3P levels in mouse iTregs (Fig. 3m–o). We investigated whether glycerol uptake enhances the suppressive function of Tregs. Naïve CD4⁺ T cells were isolated from FOXP3^GFP reporter mice and iTregs were induced in vitro under low-glucose conditions, with or without glycerol stimulation. Compared to the control group, glycerol stimulation markedly increased the expression of Tim3 and Lag3 in Tregs. Moreover, in vitro functional suppression assays demonstrated that glycerol significantly enhanced the suppressive capacity of Tregs toward responder T cells (Fig. 3p–s). However, glycerol alone did not affect the process of induction of iTregs from naïve CD4⁺ T cells (Supplementary Fig. 3r). To better understand the response of Tregs to glycerol in the tumor microenvironment, we established tumors in FOXP3^GFP reporter mice (Fig. 3t). Following tumor establishment, lymphocytes isolated from tumors were cultured in low-glucose medium with or without glycerol stimulation. Flow cytometry analysis demonstrated that glycerol stimulation significantly upregulated immunosuppressive molecules on Tregs and furthermore increased the proportion of exhausted CD8⁺ T cells, whereas conventional CD4⁺ T cells remained unaffected (Fig. 3u, v and Supplementary Fig. 3s). In summary, our findings document that glycerol metabolism plays a critical role in enhancing the suppressive function of Tregs, particularly under glucose-deprived conditions. Our data imply that glycerol metabolic reprogramming allows Tregs to thrive and augment functionality under tumor nutrient stress.

Fig. 3.

Glucose Deprivation Promotes Glycerol Metabolism and Enhances the Suppressive Function of Treg Cells. Glycerol stimulation increases Gk gene expression (a; n = 5), protein expression (b), Gk activity and Gro3P level (c; n = 6) in mouse nTregs. Gk gene expression (d; n = 3), protein expression (e), Gk activity, and Gro3P level (f; n = 3) rise as glycerol treatment time increases in mouse nTregs. g-i Glycerol analog reduces Gk gene expression (g; n = 3), protein expression (h), Gk activity and Gro3P level (i; n = 6) in mouse iTregs. AQP3 inhibition reduces Gk gene expression (j; n = 3), protein expression (k), Gk activity, and Gro3P level (l; n = 6) in mouse iTregs. Glucose deprivation enhances Gk gene expression (m; n = 3), protein expression (n), Gk activity, and Gro3P level (o; n = 6) in iTregs under low-glucose conditions. p and q Glycerol stimulation upregulates Tim3/Lag3 in Tregs induced from naïve CD4⁺ T cells (low glucose), n = 3. r and s CFSE-based functional assay validated that glycerol enhances the suppressive ability of mouse iTregs under low glucose, n = 3. t Tumors were harvested from FOXP3^GFP transgenic mice and digested; TILs were then isolated via Ficoll density gradient centrifugation. Isolated TILs were cultured in vitro with or without glycerol stimulation. Finally, surface molecules of Tregs or CD8⁺ T cells were analyzed by FACS. Figure3t was created using BioRender (https://BioRender.com/4eqscy2). u and v Glycerol stimulation in TILs increases immunosuppressive molecules on Tregs and elevates the exhausted CD8⁺ T cell proportion in low-glucose medium, n = 3. Bar graphs represent the mean ± SEM of technical duplicates per group in one representative experiment (a, c, d, f, g, i, j, l, m, o, q, s–v), with consistent results observed across independent experiments. Statistical analyses were performed as follows: Unpaired t-test (a, c, q, u, v); One-way ANOVA followed by Tukey’s post hoc test (g, i, j, l); Two-way ANOVA followed by Tukey’s post hoc test (d, f, m, o, s). For statistical significance: p ≤ 0.05 was considered significant, and p > 0.05 was considered not significant. See also Supplementary Fig.3

Histone Lactylation by Glycerol-Derived Lactate Promotes GK Expression and Upregulates Glycerol Metabolism

Following glycerol uptake and metabolism, Tregs generate intracellular lactate (Supplementary Fig. 4a). In our previous experiments involving the stimulation of induced Treg (iTreg) cells with glycerol, we measured the levels of pyruvate and lactate in both cell supernatants and intracellular compartments. We observed that not only did the expression of Gk increase over the course of the culture days, but there was also a concomitant accumulation of pyruvate and lactate (Fig. 3d–f and Supplementary Fig. 4b, c). To further investigate the role of lactate in this process, we treated both mouse and human iTregs with NaOxa, an inhibitor of LDHA (lactate dehydrogenase A), which reduces supernatant and intracellular pyruvate and lactate production (Supplementary Fig. 4d). It was noted that the inhibition of lactate production by NaOxa also suppressed the glycerol-induced upregulation of Gk expression and activity, as well as the Gro3P levels (Fig. 4a–c). In an in vitro suppression assay, the enhanced inhibitory function of Tregs induced by glycerol was attenuated when glycerol-stimulated intracellular lactate production was blocked, suggesting that lactate promotes the glycerol-mediated enhancement of Treg function (Fig. 4d, e). Recent studies have identified lysine lactylation (Kla) as a novel post-translational modification that regulates gene expression.¹⁹ We hypothesized that lactate-induced histone lactylation might mediate GK transcriptional regulation. A-485 is a potent and selective catalytic inhibitor of p300/CBP, identified as an inhibitor of lactylation.⁴⁰ To test this, we observed that A-485 indeed had a similar effect to NaOxa, suppressing both glycerol-induced expression of Gk and its enzymatic activity, as well as the downstream metabolic product Gro3P (Fig. 4f–h). A significant decrease was observed in lactate and pyruvate concentrations (intracellular and supernatant) following A-485 administration (Supplementary Fig. 4e). Moreover, we measured the overall lactylation levels in iTregs after glycerol stimulation using a pan-lactylation antibody. We found that lactylation was significantly increased following glycerol stimulation (Fig. 4i). Further screening of specific lactylation sites revealed that the primary lactylation sites upregulated in Tregs after glycerol absorption were H3K9 and H4K5 (Fig. 4j, k). Similar results were observed in mouse nTregs, human nTreg and iTregs, where H3K9 and H4K5 lactylation was also significantly induced by glycerol (Supplementary Fig. 4f and Supplementary Fig. 5a, b). In contrast, glycerol did not affect acetylation or methylation levels (Supplementary Fig. 4g, h). Furthermore, we demonstrated that Gk expression and enzymatic activity, as well as lactylation of H3K9, H4K5 were glycerol-dependent, with their levels increasing in response to higher glycerol concentrations (Fig. 4l, m). When mouse iTregs were treated with NaOxa and A-485, we saw a significant reduction in H3K9 and H4K5 lactylation (Fig. 4n, o). Consistent with these findings, AQP3i and a glycerol analog also suppressed glycerol-induced H3K9, H4K5 lactylation in mouse and human iTregs (Supplementary Fig. 4i, j and Supplementary Fig. 5c, d). NaOxa and A-485 also reduced the glycerol-induced lactylation of H3K9 and H4K5, GK expression/activity, Gro3P levels, and intra-/extracellular pyruvate/lactate in human iTregs (Supplementary Fig. 5e–j). GK seems to use the mechanistic axis through which glycerol metabolism in Treg cells promotes histone lactylation via lactate production, thereby enhancing their suppressive function. Using siRNA-mediated knockdown of Gk in Treg cells, we confirmed that Gk is indeed a critical factor in glycerol metabolism-driven lactylation-mediated enhancement of Treg suppressive function (Fig. 4p–s and Supplementary Fig. 4k, l). Furthermore, overexpression of Gk in Treg cells demonstrated that increased Gk expression promotes lactylation, glycerol metabolism pathway, and the suppressive function of Treg cells (Supplementary Fig. 4m–q). In summary, our findings document that glycerol metabolism, through the generation of lactate and subsequent induction of histone H3K9 and H4K5 lactylation, plays a critical role in upregulating GK expression and activity, allowing Tregs to thrive under nutrient stress conditions where glycerol may be more abundant, such as in specific tumor microenvironments or the liver. In microenvironments rich in glycerol, this lactate-lactylation-GK axis potently augments Treg suppressive activity by sustaining their metabolic fitness and function.

Fig. 4.

Lactate-Mediated Lactylation Regulates Glycerol-Induced Metabolic Reprogramming in Treg Cells. LDHA inhibitor (NaOxa) suppresses glycerol-induced GK gene expression (a; n = 3), protein expression (b), GK activity, and Gro3P level (c; n = 5) in mouse iTregs. d, e CFSE-based functional assay revealed that the LDHA inhibitor (NaOxa) effectively attenuated the glycerol-potentiated suppressive function of mouse iTregs, n = 3. Lactylation inhibitor (A-485) suppresses glycerol-induced GK gene expression (f; n = 3), protein expression (g), Gk activity, and Gro3P level (h; n = 6) in mouse iTregs. i Global lactylation increases in iTregs post-glycerol stimulation (pan-lactylation antibody). j, k Specific lactylation sites H3K9 and H4K5 are upregulated in Tregs after glycerol treatment. l, m Gk, H3K9la and H4K5la expression, Gk activity and Gro3P level (m; n = 6) rises as glycerol concentration increases in mouse nTregs. n NaOxa reduces glycerol-induced H3K9 and H4K5 lactylation in mouse iTregs. o A-485 reduces glycerol-induced H3K9 and H4K5 lactylation in mouse iTregs. p, q Gk knockdown reduces glycerol-induced H3K9 and H4K5 lactylation, Gk activity, and Gro3P levels in mouse iTregs, n = 6. r and s CFSE-based functional assay confirmed that Gk knockdown attenuated the glycerol-augmented suppressive capacity of mouse iTregs, n = 3. Bar graphs represent the mean ± SEM of technical duplicates per group in one representative experiment (a, c, e, f, h, m, q, s), with consistent results observed across independent experiments. Statistical analyses were performed as follows: One-way ANOVA followed by Tukey’s post hoc test (a, c, e, f, h, m, q, s). For statistical significance: p ≤ 0.05 was considered significant, and p > 0.05 was considered not significant. See also Supplementary Figs. 4 and 5

Histone H4K5 Lactylation Orchestrates FOXP3-Mediated GK Upregulation in a Glycerol-Driven Amplifying Loop

To investigate how GK is regulated by histone lactylation, we first examined whether histone lactylation directly influences GK transcription. Using ChIP-PCR, we found that lactylation of histone H3K9 and H4K5 hardly directly affects the transcriptional expression of Gk (Supplementary Fig. 6a, b). We observed that naïve CD4⁺ T cells failed to upregulate GK expression or elevate histone H3K9/H4K5 lactylation upon glycerol stimulation (Fig. 5a–c). The striking difference between naïve CD4⁺ T cells and Tregs was the robust FOXP3 expression in the latter. Glycerol dose-dependent assays further revealed that FOXP3 was specifically induced in iTregs but not in naïve CD4⁺ T cells (Supplementary Fig. 6c, d). We then observed that Tregs in the metastatic lesions of CRLM patients not only expressed higher levels of inhibitory molecules but also exhibited significantly elevated FOXP3 expression (Supplementary Fig. 6e). Given that FOXP3 is a master transcription factor of Tregs that controls Treg cell function by targeting certain genes for activation or repression,^41,42 we hypothesized that FOXP3 might regulate GK transcription. JASPAR-based bioinformatics analysis⁴³ identified one putative FOXP3 binding motif within the GK promoter (−1 to −2000 bp upstream of the transcription start site plus the exon sequence of the 5’ untranslated region (5’UTR) (Fig. 5d). To validate this, we performed a luciferase reporter assay and confirmed that FOXP3 indeed directly targets the GK promoter to regulate its transcription (Fig. 5e). To further pinpoint the specific regions of the GK promoter where FOXP3 binds, we divided the GK promoter into 13 segments, each approximately 180 bp in length. ChIP-qPCR showed that FOXP3 primarily binds to segments 12 and 13 of the GK promoter (Fig. 5f, g). Based on the sequences of fragments 12 and 13, we predicted FOXP3-binding sites within these regions and identified three high-scoring binding motifs (Supplementary Fig. 6f). We subsequently generated deletion mutants of these three predicted binding sites in luciferase reporters and demonstrated that all three sites are critical for FOXP3 binding to the GK promoter, as FOXP3 failed to bind upon their deletion (Fig. 5h). We then explored whether histone lactylation, specifically H3K9 and H4K5 lactylation, influences GK expression through FOXP3. Using ChIP-PCR, we determined that Foxp3 upregulation is primarily mediated by H4K5 lactylation (Fig. 5i, j). Further analysis revealed that Cd177 expression on Gk⁺ Tregs is also directly regulated by Foxp3, independent of histone H4K5 and H3K9 lactylation (Supplementary Fig. 6g, h). To verify that H4K5 is a critical lactylation site regulating Foxp3 expression, we constructed wild-type H4 overexpression plasmids and mutant H4 plasmids in which the fifth lysine residue was replaced with arginine (K5R) (Supplementary Fig. 6i). We found that mutation of this site led to downregulation of both glycerol-induced Foxp3 and Gk expression (Fig. 5k). To further elucidate the mechanistic axis by which the GK-mediated glycerol metabolism pathway promotes FOXP3 expression through histone lactylation, thereby enhancing FOXP3 binding to the GK promoter and amplifying this regulatory circuit, we performed knockdown of Gk and then used an H4K5 lactylation antibody to pull down and detect the bound Foxp3. We found that Gk knockdown reduced the amount of Foxp3 bound to lactylated H4K5 (Fig. 5l). Moreover, when intracellular glycerol-induced lactylation was inhibited, Foxp3 binding to the Gk promoter was also significantly suppressed (Fig. 5m). Following glycerol stimulation, we performed CUT&Tag sequencing for H4K5 lactylation (H4K5la) enrichment. This revealed a significant increase in H4K5la-bound genes post-stimulation (Fig. 5n, o and Supplementary Fig. 6j, k). Notably, glycerol-induced H4K5la upregulated Foxp3 expression (Fig. 5p), while exerting no significant effect on the direct regulation of Gk or Cd177 (Supplementary Fig. 6l, m). Furthermore, among H4K5la-regulated genes, LAG3-a key molecule mediating Treg suppressive function-showed marked upregulation (Fig. 5q). Our findings demonstrate that FOXP3-a key transcription factor in Tregs-directly regulates GK transcription by binding to specific promoter regions. Additionally, we identify that H4K5 lactylation, but not H3K9 lactylation, modulates FOXP3 expression, thereby indirectly influencing GK expression and co-regulating CD177 expression. These results suggest a complex regulatory network in which glycerol-derived lactate induces H4K5 lactylation, enhancing FOXP3 binding to the GK promoter and promoting glycerol metabolism in Tregs (Fig. 5r).

Fig. 5.

FOXP3 Directly Activates GK Transcription via Promoter Binding and Is Modulated by H4K5 Lactylation. a-c Naive CD4⁺ T cells show no upregulation of Gk protein expression or H3K9/H4K5 lactylation upon glycerol stimulation (vs. Tregs). d Bioinformatics prediction of FOXP3 binding motif in GK promoter (−2000 bp to 5′UTR). e Luciferase assay confirms FOXP3 directly targets GK promoter, n = 2. f, g ChIP-qPCR identifies FOXP3 binding to segments 12–13 of GK promoter. Figure 5f was created using BioRender (https://BioRender.com/r0zy1ql), n = 3. h Deletion mutants of three predicted FOXP3-binding sites abolish FOXP3-promoter binding, n = 2. i, j H4K5 lactylation (not H3k9) regulates Foxp3 expression (ChIP-PCR), n = 3. k Overexpression of the pCMV-EGFP-H4C1-3xFlag plasmid enhances glycerol-induced expression of Foxp3 and Gk in mouse iTregs, whereas overexpression of the mutant H4C1 (K5R) plasmid does not exhibit the same enhancing effect as wild-type H4C1. l Glycerol-enhanced H4K5 lactylation regulates Foxp3 expression in a Gk-dependent manner (ChIP-PCR), n = 3. m Glycerol-promoted Foxp3 binding to the Gk promoter is dependent on lactylation regulation (ChIP-PCR), n = 3. n, o Cut&Tag sequencing: The binding density of H4K5la in iTregs between the control and glycerol stimulation group (n). Figure 5n was created using BioRender (https://BioRender.com/z78kpqf). The color depth of the heatmap indicates the relative number of reads, genes with similar distribution patterns are clustered together through a clustering algorithm to show the binding trends of lactylation modifications on all genes (o). Genome browser tracks of Cut&Tag signal at Foxp3 (p) and Lag3 (q). The highlighted rectangles indicate the peak regions of H4K5la in the promoter region. r Glycerol-derived lactate induces H4K5 lactylation in Tregs. Increased H4K5la upregulates FOXP3 and LAG3 expression; subsequently, FOXP3, as a transcription factor, upregulates GK and CD177 expression. Figure 5r was created using BioRender (https://BioRender.com/y60p18). Bar graphs represent the mean ± SEM of technical duplicates per group in one representative experiment (e, g–j, l, m), with consistent results observed across independent experiments. Statistical analyses were performed as follows: One-way ANOVA followed by Tukey’s post hoc test (l, m); two-way ANOVA followed by Tukey’s post hoc test (e, g, h, i, j). For statistical significance: p ≤ 0.05 was considered significant, and p > 0.05 was considered not significant. See also Supplementary Fig. 6

High lactate boosts GK expression and glycerol metabolism via H4K5 lactylation in Tregs

Lactate is transported across cell membranes by a family of hydrogen ion-dependent monocarboxylate transporters (MCTs).⁴⁴ We then used VB124, a selective MCT4 inhibitor that specifically inhibited lactate efflux (Fig. 6a). Blockade of intracellular lactate export via MCT4 inhibition (VB124) caused lactate accumulation, which paradoxically amplified glycerol-stimulated increases in H4K5 lactylation, GK expression, GK enzymatic activity, and glycerol-3-phosphate (Gro3P) levels in both mouse iTregs (Fig. 6c,d and Supplementary Fig. 7a) and human iTregs (Supplementary Fig. 7c–e). We also used AZD3965, a selective MCT1 inhibitor that primarily inhibits lactate uptake (Fig. 6b and Supplementary Fig. 7b), and found that the expression of H4K5 lactylation and GK, the activity of GK, as well as the levels of the glycerol metabolic intermediate Gro3P, were significantly inhibited by AZD3965 in both mouse and human iTregs (Fig. 6e,f and Supplementary Fig. 7f–h). Lactic acid is highly enriched in the tumor microenvironment (TME) and is known to have immunosuppressive effects.⁴⁵ It has been well established that Tregs maintain high suppressive function within the TME.⁶ We hypothesized that high concentrations of lactate in the TME, similar to intracellular lactate produced by glycerol metabolism, could induce histone lactylation, thereby upregulating the expression of GK and promoting the glycerol metabolic pathway in Tregs, ultimately enhancing their survival and function. To investigate this, with increasing concentrations of exogenous lactate, H4K5 lactylation and FOXP3 were robustly enhanced in a concentration-dependent manner. However, we observed that GK expression did not show a clear concentration-dependent response; its induction was only observed at high lactate concentrations (Fig. 6g). This suggests that although lactate can upregulate H4K5 lactylation, the regulation of Gk by Foxp3 under glycerol stimulation may involve additional glycerol-triggered signaling mechanisms distinct from pure lactate stimulation that uniquely enhance Gk expression. To learn the possibility that lactate serves to amplify the effects of glycerol metabolism, we exogenously supplemented of mouse and human iTregs with lactate plus glycerol that indeed induced a significant synergistic enhancement of H4K5 lactylation and GK expression (Fig. 6h and Supplementary Fig. 7i). Nonetheless, GK induction exhibited no lactate concentration-dependence, requiring only that lactate concentrations reached a threshold level to enable maximal expression (Fig. 6h). Intracellular GK activity and Gro3P content peaked under dual lactate-glycerol stimulation, significantly exceeding levels in glycerol-only treated cells (Fig. 6i and Supplementary Fig. 7j). This finding aligns with our observation that Tregs with high GK expression are more prevalent in metastatic sites, where the environment is more hostile and lactate levels are elevated (Supplementary Fig. 2c). Furthermore, when we inhibited lactate uptake using AZD3965, the synergistic effect of glycerol and lactate on H4K5 lactylation, GK expression/activity and Gro3P content was significantly diminished in both mouse and human iTregs (Fig. 6j, k and Supplementary Fig. 7k, l). When we knocked down Gk, the synergistic effect of lactate and glycerol on H4K5-mediated enhancement of Foxp3 transcription was abolished (Fig. 6l, n). Similarly, when lactylation was inhibited, the synergistic effect of lactate and glycerol on Foxp3-mediated upregulation of Gk expression was also reduced (Fig. 6m, o). Furthermore, we found that, compared with glycerol, lactate alone could indeed promote Foxp3 expression through H4K5 lactylation, but it did not exhibit the same pronounced effect on Foxp3 binding to the Gk promoter. These results support our previous hypothesis: although glycerol metabolism promotes GK expression in a manner dependent on H4K5 lactylation, additional regulatory mechanisms triggered by glycerol signaling may also contribute to FOXP3-mediated regulation of GK (Supplementary Fig. 7m).

Fig. 6.

Tumor Microenvironment Lactate Amplifies Glycerol Metabolism in Tregs via MCT1/MCT4-Dependent Lactylation. a, b Schematic of lactate transport: MCT1 inhibitor (AZD3965) blocks lactate uptake, MCT4 inhibitor (VB124) blocks lactate efflux. VB124 induces glycerol-induced H4K5 lactylation and Gk protein expression (c), Gk activity, and Gro3P level (d) in mouse iTregs, n = 6. AZD3965 inhibits glycerol-induced H4K5 lactylation and Gk protein expression (e), Gk activity, and Gro3P level (f) in mouse iTregs, n = 6. g NaLac enhances H4K5 lactylation and the expression of Foxp3 and Gk, with the most prominent effect on H4K5 lactylation. The expression of Foxp3 and Gk is not lactate gradient-dependent, as a marked increase is seen only at the highest concentration. A bubble chart was used for quantitative display. h and i Lactate + glycerol synergy: Threshold-dependent enhancement of H4K5 lactylation and Gk expression (h), Gk activity and Gro3P level (i; n = 6) in mouse iTregs. A bubble chart was used for quantitative display (h). AZD3965 abolishes lactate-glycerol synergy on H4K5 lactylation and Gk expression (j), Gk activity and Gro3P level (k; n = 6) in mouse iTregs. A bubble chart was used for quantitative display (j). ChIP-PCR validated that Gk knockdown abolished the glycerol- and lactate-induced H4K5 binding of Foxp3 (l; n = 3), A-485-mediated inhibition of H4K5 lactylation weakened the glycerol-enhanced binding of Foxp3 to the Gk locus (m; n = 3). n Gk knockdown abolishes lactate-glycerol synergy on H4K5 lactylation and Gk expression in mouse iTregs. o A-485 abolishes lactate-glycerol synergy on H4K5 lactylation and Gk expression in mouse iTregs. Bar graphs represent the mean ± SEM of technical duplicates per group in one representative experiment (d, f, i, k, l, m), with consistent results observed across independent experiments. Statistical analyses were performed as follows: One-way ANOVA followed by Tukey’s post hoc test (d, f, i, k, l, m). For statistical significance: p ≤ 0.05 was considered significant, and p > 0.05 was considered not significant. See also Supplementary Fig. 7

Targeting Treg-Specific Glycerol Metabolism Enhances Anti-PD-1 Immunotherapy

To assess the therapeutic potential of targeting glycerol metabolism, C57BL/6 mice inoculated with colorectal cancer cell line-MC38 cells were fed a high-fat diet (HFD) for 12 weeks to elevate systemic glycerol levels. Currently, no specific inhibitors targeting GK are commercially available. Analysis of bulk RNA-sequencing datasets from peripheral blood mononuclear cell (PBMC)-derived T cells across 157 patients spanning 12 cancer types revealed a statistically significant positive correlation between GK expression and AQP3 (Supplementary Fig. 8a), the glycerol-transporting channel, in T cells. To evaluate the impact of the GK-dependent glycerol metabolic pathway in Tregs on tumor progression, we used the selective AQP3 inhibitor DFP00173 to block glycerol uptake in vivo mouse models. Tumor-bearing mice received intraperitoneal injections of AQP3 inhibitor DFP00173 alone, PD-1 antibody alone, or a combination of both starting at day 3 post-inoculation (Fig. 7a). In HFD-fed mice, DFP00173 monotherapy achieved comparable antitumor efficacy to anti-PD-1 antibody, whereas it exhibited no significant antitumor effect in normal chow diet (NCD)-fed mice. Strikingly, co-treatment with DFP00173 and PD-1 antibody together markedly reduced tumor volumes in HFD-fed mice (Fig. 7b, c). Flow cytometry analysis revealed that monotherapy with either DFP00173 or anti-PD-1 antibody reduced the abundance of TI-Tregs in HFD-fed mice, and the combination treatment further amplified these effects (Fig. 7d, e). To explore the underlying mechanisms, TI-Tregs were isolated, and lactate levels, histone H4K5 lactylation levels, Gk expression, and Gro3P levels were measured. DFP00173 treatment significantly reduced glycerol uptake, Gro3P levels, and lactate levels in Tregs from both normal diet- and high-fat diet-fed mice, whereas PD-1 antibody treatment alone had no effect on glycerol metabolism (Fig. 7f, g and Supplementary Fig. 8b). To determine whether DFP00173 similarly inhibits glycerol uptake in CD8⁺ T cells in tumor, we measured Aqp3 expression in spleen-derived naïve CD8⁺ T cells and effector CD8⁺ T cells from mice. Notably, Aqp3 levels were significantly lower in both subsets compared to those in CD4⁺ iTregs (Supplementary Fig. 8c). To evaluate the specificity of targeted blockade of Treg cell glycerol metabolism, tumor-infiltrating FOXP3⁺ Tregs and CD8⁺ T cells were isolated from MC38-bearing mice and cultured in low-glucose medium. Tregs exhibited robust glycerol uptake and utilization, as evidenced by higher Gk activities and Gro3P levels, while CD8⁺ T cells showed minimal glycerol dependency (Supplementary Fig. 8d). Similarly, in colorectal cancer patient-derived Tregs and CD8⁺ T cells, Tregs maintained superior glycerol absorption under low-glucose conditions, further supporting the selective targeting of Treg metabolism by DFP00173 (Supplementary Fig. 8e). Moreover, biopsy specimens from colorectal cancer patients were stratified into CD177⁺ and CD177⁻ Treg subsets via flow cytometry. The GK activities and Gro3P levels were significantly higher in CD177⁺ TI-Tregs than in CD177⁻ TI-Tregs (Fig. 7h). Further stratification of patients into high- and low-glycerol groups revealed that high-glycerol tumors harbored more CD177⁺ TI-Tregs with elevated GK activities and Gro3P levels, which positively correlated with intratumoral glycerol contents (Fig. 7i, j and Supplementary Fig. 8f, g). Immunoblotting confirmed increased histone H4K5 lactylation and GK expression in high-glycerol patient tissues (Fig. 7k). These findings highlight the key role of AQP3 inhibition in disrupting Treg-mediated immunosuppression and enhancing anti-PD-1 therapy efficacy. The specificity of DFP00173 for Tregs over CD8⁺ T cells underscores its therapeutic potential in glycerol-rich tumors (Supplementary Fig. 8h).

Fig. 7.

Targeting Treg-Specific Glycerol Metabolism Enhances Anti-PD-1 Immunotherapy. a Treatment schema: MC38 tumor-bearing mice fed high-fat diet (HFD) to elevate systemic glycerol; interventions: AQP3 inhibitor (DFP00173) ± αPD-1 antibody. b DFP00173 monotherapy reduces tumor volumes in HFD mice (vs. normal chow diet). c Synergistic antitumor effect of DFP00173 with αPD-1 in HFD mice, n = 4. d and e DFP00173 and/or αPD-1 decrease tumor-infiltrating Tregs in HFD mice, n = 4. DFP00173 monotherapy reduces glycerol-induced H4K5la and GK protein expression (f), GK activity and Gro3P level (g; n = 3) in HFD mice (vs. normal chow diet). h Higher GK activity and Gro3P level in CD177⁺ vs. CD177^- TI-Tregs (CRC patients), n = 6. i High-glycerol tumors harbor more CD177⁺ TI-Tregs with elevated GK activity and Gro3P levels isolated from CRC patients, n = 9. j Glycerol levels were positively correlated with GK activity and Gro3P content in tumor-infiltrating CD177⁺ Tregs isolated from CRC patients, n = 9. k Increased H4K5 lactylation and GK protein expression in high-glycerol CRC tissues. Bar graphs represent the mean ± SEM of technical duplicates per group in one representative experiment (c, e, g–j), with consistent results observed across independent experiments. Statistical analyses were performed as follows: Two-way ANOVA followed by Tukey’s post hoc test (c, e, g, i); Unpaired t-test (h); Pearson correlation analysis (j). For statistical significance: p ≤ 0.05 was considered significant, and p > 0.05 was considered not significant. See also Supplementary Fig. 8

Pan-Cancer Analysis Reveals Strong Correlation Between GK Expression in Treg and Treg Infiltration/Function in the Tumor Microenvironment

To investigate whether a GK expression/activity-dependent glycerol utilization pathway promotes the accumulation and enhanced function of tumor-infiltrating Tregs and enhanced function across multiple cancer types, we analyzed scRNA-seq transcriptomic datasets of T cells from peripheral blood mononuclear cells (PBMCs) from approximately 300 patients across 20 cancer types obtained from the Gene Expression Omnibus (GEO) database (GSE156728). Initial analysis revealed significantly higher GK expression in CD8⁺ T cells compared to CD4⁺ T cells (Fig. 8a). Further clustering of T cell populations demonstrated that GK expression was predominantly localized to Treg cells (Fig. 8b, c). Notably, GK expression was particularly enriched within the c20_Treg_TNFRSF9 and c21_Treg_OAS1 Treg subset, which also exhibited high CD177 expression (Fig. 8d). Analysis of T cell clustering across diverse malignancies yielded results consistent with our findings in CRLM patient liver metastases (Fig. 2l). CD177 and GK were primarily co-expressed on Treg cells, and regions of high CD177 expression generally corresponded with strong GK expression in most cancer types, with the exception of BCL cancer (Fig. 8e, f). Pan-cancer analysis further revealed a positive correlation between elevated GK expression within Tregs and the proportion of Tregs within the TME. This association was most pronounced in CRC, STAD, and CHOL cancers (Fig. 8g). Subsequently, we examined Treg functional molecules and found that the c20 and c21 Treg subsets, characterized by high GK and CD177 expression, also exhibited the strongest expression of Treg functional markers (Fig. 8h). This observation indicates that tumor-infiltrating CD177⁺GK^high Treg cells possess potent suppressive capabilities.

Fig. 8.

Pan-Cancer Analysis Reveals Strong Correlation Between GK Expression in Treg and Treg Infiltration/Function in the Tumor Microenvironment. a Violin plot showing the significant expression of GK between CD4⁺ and CD8⁺ T subsets. b UMAP plot illustrating the subsets of CD4⁺ T cells from different types of cancer. c Feature plot showing the highly specific expression of GK in Treg subsets. d Violin plot showing the highly specific expression of GK and CD177 in TNFRSF9⁺ Treg (C20) and OAS1⁺ Treg (C21), respectively. e Feature plot showing the differential expression of GK across 20 cancer types. f Feature plot showing the differential expression of CD177 across 20 cancer types. g The correlation of GK expression level and Treg proportion across different cancers. h Heatmap showing the differential expression of inhibitory genes across CD4⁺ T subsets. i The distribution of different-sized TCR clones in Treg cells across PN, PT, MN and MT. j Clonal comparison of GK^- Treg and GK⁺ Treg cells between the PT and the MT. k Alluvial plot of clonotype dynamics

Several studies have previously reported a close relationship between lipid metabolism and TCR diversity.^46,47 Whether GK⁺ Treg cells in the tumor microenvironment also influence their TCR diversity through the glycerol metabolic pathway remains an intriguing question. To address this, we analyzed the TCR sequencing data from colorectal cancer liver metastasis patients (GSE164522). We compared the clonal diversity of GK⁺ Treg cells among primary tumors (PT), peritumor normal tissues (PN), and metastases (MT). Notably, the clones in MT-adjacent normal tissue were also more diverse than those in PN, indicating distal immune remodeling after CRC metastasized to the liver (Fig. 8i). When comparing GK^- Treg and GK⁺ Treg clones between PT and MT, both subsets exhibited higher clone counts in MT than in PT, suggesting greater clonal diversity in metastases. Moreover, very few clones overlapped among the four groups (PN, PT, MT-normal, MT), indicating that specific TCR clones emerge during tumor progression (Fig. 8j). In MT, the dominant clones expanded extremely, forming a vastly broader “river” of clonal dominance far exceeding that in PT (Fig. 8k). This suggests that in GK⁺ Treg cells, certain T-cell clones-most likely targeting metastasis-associated antigens-undergo extreme clonal expansion, gaining a marked growth advantage in the metastatic microenvironment. These findings strongly indicate that the metabolic niche (including glycerol and lipid availability) can shape TCR clonal selection and expansion in Treg cells. Our results also provide valuable clues for understanding how obesity and high-fat diet may influence T-cell function in the tumor microenvironment.

Discussion

In this study, we explored the role of GK⁺ Tregs in the tumor microenvironment (TME) of colorectal liver metastases (CRLM) patients and identified a critical link between glycerol metabolism, histone lactylation, and the immunosuppressive function of Tregs. To our knowledge, this finding is the first to show that Tregs in the local microenvironment of metastatic lesions exhibit enhanced inhibitory capabilities by using glycerol and adapting to nutrient-deficient conditions. These innovative insights not only deepen our understanding of Treg cell biology in the TME but also offer potential therapeutic targets for modulating immune responses in cancer.

The liver serves as the primary site for glycerol processing and conversion, functioning as a key gluconeogenic precursor that can be transformed into glucose during hypoglycemia.^48,49 Research on glycerol metabolism in the field of immunology is currently still in an exploratory stage, but its importance has gained increasing prominence in recent years—particularly regarding its roles in the tumor microenvironment and metabolic stress responses. Glycerol metabolism serves as a critical hub for cellular adaptation to microenvironmental pressure by balancing reductive stress and energy stress. Li’s team previously reported that in the hypoxic tumor microenvironment, AMPK regulates glycerol metabolism in tumor cells through phosphorylation of the ALDOB protein,⁵⁰ its regulatory mechanisms offer novel insights for understanding metabolic reprogramming in immune cells. Their study further revealed significantly elevated serum glycerol levels in patients with cervical, lung, and liver cancers compared to healthy individuals, suggesting a potential association between glycerol secretion and tumor progression. However, whether this phenomenon correlates with immune cell infiltration depth or responsiveness to immunotherapy remains unexplored.

In this study, we identified that the glycerol metabolic pathway was highly active in glycerol kinase (GK)-high-expressing Tregs. Interestingly, while Tregs exhibited robust glycerol utilization, CD8⁺ T cells showed much weaker metabolic activity in this regard. This metabolic disparity between Tregs and CD8⁺ T cells may contribute to the immunosuppressive dominance of Tregs in the TME and provides a rationale for selectively targeting Treg cell metabolism to restore anti-tumor immunity. Building on the Tegnér team’s discovery of regulons governing human Treg differentiation through time-resolved transcriptome and proteome analysis, we further determined glycerol kinase (GK) as a key metabolic regulator within their dataset.⁵¹ This finding also implies that GK is crucial for the induction of Tregs from naïve CD4⁺ T cells and explains the superior reliance of Tregs on glycerol metabolism compared to CD8⁺ T cells.

One of the key discoveries is the high expression of glycerol kinase (GK) in Tregs with strong immunosuppressive functions. In addition, using spatial transcriptomics, we found that GK⁺ Treg cells, which are highly expressed in liver metastases, are predominantly localized at the interface between tumor and adjacent non-cancerous tissue. This spatial distribution pattern offers intriguing insights into tumor invasion and metastasis. Additionally, we authenticated that GK⁺ Tregs interact with CXCL12⁺ and SPP1⁺ macrophages, which may serve as an additional contributing factor to a colder tumor immune microenvironment. These cells also express the surface marker CD177, which serves as a potential biomarker for identifying this subset of Tregs. Notably, CD177⁺GK⁺ Tregs demonstrate a superior capacity to absorb and utilize glycerol, particularly under glucose-deprived conditions, highlighting their metabolic adaptability in the harsh TME. We previously reported that the CD177⁺ Treg subset displays superior immunosuppressive capacity in the TME,³⁵ and the current study further clarifies that metabolic alteration indeed shapes CD177⁺ Treg biological activity, helping explain the underlying mechanism by which this subset of Tregs enhances their function in the TME. This metabolic reprogramming not only enhances their inhibitory function but also underscores the importance of glycerol as an alternative energy source in nutrient-scarce environments.

Previous studies have reported regulatory mechanisms of non-histone lactylation on Treg function,^23,24; our current study has provided a new explanation for histone lactylation on Treg phenotype and function. External lactate can synergistically enhance the effect of glycerol on Treg cells, but whereas lactate can upregulate H4K5 lactylation, the regulation of GK by FOXP3 under glycerol stimulation may involve additional glycerol-triggered signaling mechanisms distinct from pure lactate stimulation that uniquely enhance GK expression. FOXP3, a master transcription factor essential for Treg cell development and function, has recently been shown to drive metabolic reprogramming.⁴¹ CAR-T cells engineered to express FOXP3 (CAR-T-Foxp3) exhibited a distinct metabolic shift, characterized by downregulated aerobic glycolysis and oxidative phosphorylation alongside upregulated lipid metabolism, thereby enhancing their adaptability.⁵² In this study, we demonstrate that FOXP3 is essential for metabolic control in Tregs. Importantly, this process is specific to the H4K5 lactylation site, providing a potential therapeutic target for modulating Treg cell function. Interestingly, although both H4K5 and H3K9 lactylation modifications are elevated in the glycerol metabolic pathway, H4K5 lactylation governs glycerol metabolism in Tregs by upregulating FOXP3 expression to transcriptionally control glycerol kinase (GK). Furthermore, glycerol-dependent H4K5 lactylation enhances the expression of LAG3, an inhibitory surface molecule on Tregs. The functional significance of H3K9 lactylation in Tregs remains unclear and warrants further investigation. Our research on lactylation in Tregs has provided a better understanding of the cross-talk between metabolomics and epigenomics. Several studies have reported a close relationship between lipid metabolism and TCR diversity.^46,47 In line with this, we found that in GK⁺ Treg cells, certain T-cell clones—most likely targeting metastasis-associated antigens—undergo extreme clonal expansion, gaining a marked growth advantage in the metastatic microenvironment. This conclusion also provides valuable clues for understanding how obesity and high-fat diet may influence T-cell function in the tumor microenvironment.

In high-fat diet (HFD)-fed obese mice with elevated systemic glycerol levels, blocking the glycerol utilization pathway reduced intratumoral Treg infiltration and inhibited tumor progression, and combining this with anti-PD-1 therapy showed enhanced efficacy compared to monotherapy. However, this synergistic effect was context-dependent, observed only in HFD-fed mice but not in normal diet-fed mice. Importantly, our findings are subject to notable limitations for clinical translation. HFD-fed mice may inherently possess an immunosuppressive state that does not fully recapitulate human conditions. Furthermore, the size of murine tumors is not comparable to that of clinical liver metastases in patients, making it difficult to extrapolate our results directly to human disease. Despite these caveats, our study provides a proof-of-concept that disrupting Treg glycerol metabolism can sensitize tumors to immune checkpoint blockade. Future studies using Gk^fl/flFoxp3^cre mice would provide more definitive evidence for the importance of glycerol metabolism in the immunosuppressive activity of tumor-associated Tregs. For patients with obesity or fatty liver disease, the glycerol utilization pathway of Tregs may be even more pronounced, potentially leading to stronger immunosuppressive activity and making immunotherapy less effective.⁵³ This raises important clinical questions about how to tailor immunotherapeutic strategies for these patient populations. Developing small-molecule drugs or polypeptides targeting GK to block the glycerol metabolic pathway within Tregs could be a promising approach.

In conclusion, our study elucidates a novel mechanism within the hepatic metastases of CRLM patients, by which a distinct CD177⁺GK⁺ Treg subset uniquely exploits glycerol metabolism and histone lactylation to enhance its immunosuppressive function. Glucose deprivation triggers a loop: Glycerol-Lactate-Histone lactylation-Enhanced FOXP3/GK-Potent immunosuppression (Supplementary Fig. 8h). These findings deepen our understanding of Treg cell biology in the TME. Future studies should explore the broader applicability of these findings in other cancer types and investigate the potential of combining metabolic inhibitors with existing immunotherapies to achieve better clinical outcomes.

Materials And Methods

Animals

This study employed C57BL/6 (B6) mice and B6 Foxp3-GFP knock-in mice. These animals were maintained in the animal facility of Songjiang Hospital affiliated with Shanghai Jiao Tong University. All mice were housed under specific pathogen-free (SPF) conditions with a 12 h light/dark cycle, controlled temperature (22 ± 1°C), and humidity (50 ± 5%). Food and water were provided ad libitum. All study protocols involving mice were approved by the Institutional Animal Care and Use Committee of Songjiang Hospital Affiliated to the Shanghai Jiao Tong University and conducted in accordance with China’s governmental regulations for the care and use of animals.

Patients and specimens

Human colorectal cancer specimens from 9 patients were frozen sectioned and prepared for Oil Red O Staining. The samples were classified into high and low glycerol groups according to their measured glycerol content. And 12 samples were classified into high and low glycerol groups according to their measured glycerol content. Additionally, tumor tissues from 12 patients with colorectal cancer were digested into single-cell suspensions. Through flow cytometry sorting, CD8⁺ T cells and Treg cells were isolated, respectively. Furthermore, Treg cells were subdivided into CD177⁺ and CD177^- populations for subsequent measurement of GK activity and Gro3P content. The use of CRC patient specimens and the relevant database was approved by the institutional review board of Songjiang Hospital Affiliated to the Shanghai Jiao Tong University School of Medicine and complied with all relevant ethical regulations (Ethical Approval No. 2024-58-XZ01). All human-derived samples used in this study were obtained after written informed consent was acquired from the participating patients. The detailed information for all patients can be found in the attached Table 1.

Animal experiment ethical approval

Animal experiments were approved by the Institutional Animal Care and Use Committee of Songjiang Hospital Affiliated to the Shanghai Jiao Tong University School of Medicine and conducted in accordance with China’s governmental regulations for the care and use of animals (Ethical Approval No. IACUC-82371817-01, ACE-003-2025). For mouse tumor experiments, all tumor sizes/burdens were permitted by the Institutional Animal Care and Use Committee of Songjiang Hospital Affiliated to the Shanghai Jiao Tong University School of Medicine, and the maximal tumor burden was not exceeded in this study.

Tumor cell culture

The MC38 mouse colon cancer cell line was obtained from the American Type Culture Collection. This cell line was grown in Dulbecco’s Modified Eagle’s Medium (DMEM, Gibco) supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% penicillin-streptomycin (Gibco) under standard cell culture conditions. The incubation was carried out at 37 °C in an atmosphere containing 5% CO₂. When the cell confluency reached 80–90%, sub-culturing was performed to maintain optimal growth conditions for the MC38 cells.

Cell preparation

Enriched T cells were isolated from spleen cells in C57BL/6 (B6) mice or B6 Foxp3-GFP knock-in mice by collecting nylon–wool column nonadherent cells.⁵⁴ Various T-cell populations were acquired by negative selection via the auto-MACS method. Briefly, enriched T cells were first stained with PE-conjugated anti-CD8a, -CD25, -CD11b, and-B220 mAbs and then washed and combined with anti-PE microbeads (Miltenyi Biotec, Auburn, CA, USA). After they passed through the MACS separation columns, the negative exports were collected as CD4 + CD25- cells. Subsequently, naive CD4 + CD25-CD62L + T cells were positively selected from the enriched CD4 + CD25- T-cell fraction by the anti-CD62L antibody and microbeads.

The generation of mouse CD4⁺ iTregs ex vivo

For the induction of CD4⁺ Tregs, naive CD4⁺CD25^-CD62L⁺ T cells were isolated from the spleens of B6 mice or B6 Foxp3-GFP knock-in mice by using a naive CD4⁺ T cell negative isolation kit (Miltenyi Biotec) as described above. Naive CD4⁺CD25^- cells were cultured in 96-well plates and stimulated with anti CD3/28 microbeads (the ratio of beads to cells was 1:5), rhIL-2 (50 U/ml; R&D Systems, Minneapolis, MN, USA), with (CD4_TGFβ) or without rhTGF-β (2 ng/ml; Humanzyme, Chicago, IL) (CD4_Med) for 3 days. RPMI 1640 medium supplemented with 100 U/ml penicillin, 100 mg/ml streptomycin,10 mM HEPES (Invitrogen-Life Technologies, Gaithersburg, MD, USA), and 10% heat-inactivated FCS (HyClone Laboratories, Logan, UT, USA) was used for all cultures. When the cultures were terminated, the cells were harvested, and the beads were removed for secondary culturing. The expression of Foxp3 was determined by flow cytometry.

In vitro suppressive assay

Freshly isolated T cells (responder cells) labeled with CFSE were stimulated with anti-CD3 mAb (0.025 ug/mL) and mitomycin C-treated APCs (1:1 ratio) for 3 days, with or without iTregs generated as described above. (The ratio of Treg: T cell was 1:2, or in graded ratios.) T-cell proliferation was determined by the CFSE dilution rate after 3 days of culture. The suppression function was calculated as baseline minus experimental group.

Generation of human iTregs ex vivo

PBMCs were prepared from the heparinized venous blood of healthy adult volunteers by Ficoll-Hypaque density gradient centrifugation. All protocols that involved human blood donors were approved by the Institutional Review Board at the University of Southern California. T cells were prepared by negative selection to a purity of 95%.⁵⁵ CD4⁺CD45RA⁺ cells were isolated from the CD4⁺CD25^- cells by negative selection, in an approach similar to that used to isolate the mouse cells, except that the antibodies used (described above) were activated with anti-human CD3/CD28 beads 1:10 (1 bead:10 cells) for 5 days in AIM-V serum-free medium (Life Technologies, Carlsbad, CA, USA) containing HEPES buffer (10 mM), sodium pyruvate (1 mM), glutamine, nonessential amino acids, and penicillin and streptomycin. This complete medium was supplemented with IL-2 (50–100 U/ml), with or without TGF-1 (5 ng/ml). Depending on their density, the cells were split, and the culture medium with the corresponding additives was replaced every 3 days.

RNA extraction and Quantitative Real-time PCR

Total RNA was extracted from human and mouse nTreg or iTreg cells by employing the Super Fastpure Cell RNA Isolation Kit (Vazyme) in accordance with the manufacturer’s instructions. The purity and concentration of the isolated RNA were determined spectrophotometrically using a NanoDrop 2000 (Thermo Fisher Scientific). The quality of the RNA was ensured by verifying that the A260/A280 ratios fell within the range of 1.8-2.0. Subsequently, 1 μg of the total RNA was reverse-transcribed into cDNA using a HiScript III All-in-one RT SuperMix Perfect for qPCR Kit (Vazyme).

Quantitative Real-time PCR (qRT-PCR) was carried out in triplicate. The reaction was performed using SYBR Green Master Mix (Vazyme) on a QuantStudio 5 Real-Time PCR System (Applied Biosystems). The reaction mixtures, with a total volume of 10 μL, consisted of 5 μL of SYBR Green mix, 2 μL of cDNA template, 0.25 μM each of forward and reverse primers, and nuclease-free water. The relative mRNA expression levels were normalized to the endogenous control, 18S rRNA, and calculated using the 2^^(−ΔΔCt) method. The sequences of the primers used in this study are presented in Table 2.

Immunoblotting

Total protein extracts were obtained from human and mouse nTreg or iTreg cells by utilizing RIPA lysis buffer (25 mM Tris-HCl at pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, and 0.1% SDS) supplemented with a protease inhibitor cocktail (Beyotime). Protein concentrations were determined through the BCA assay (Thermo Fisher Scientific).

An equal amount of protein (ranging from 20–50 μg per lane) was separated by SDS-PAGE on 15% gels and then transferred onto polyvinylidene fluoride (PVDF) membranes with a 0.2-μm pore size (Millipore). The membranes were blocked with 1% bovine serum albumin (BSA) in phosphate-buffered saline (PBS) for 1 hour at room temperature. Subsequently, they were incubated overnight at 4 °C with primary antibodies. After being washed three times with PBS containing 0.1% Tween-20 (PBST), the membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 hour at room temperature.

Protein bands were visualized using an enhanced chemiluminescence (ECL) substrate (Beyotime) and captured with a ChemiDoc MP Imaging System (Bio-Rad). All primary and secondary antibodies used in this study are listed in Table 3.

Sample preparation and metabolite content measurement

Following the completion of the experiment, Treg cells and corresponding culture supernatants were separately acquired via centrifugation. The levels of metabolites in the cell supernatants and cell extracts were then measured. Specifically, the concentrations of lactate, pyruvate, and glycerol were determined using a colorimetric assay kit (Nanjing Jiancheng Bioengineering Institute); glycerol-3-phosphate (Gro3P) was determined using a colorimetric assay kit (Grace Biotechnology) following the manufacturer’s protocol. The optical density was precisely recorded using a microplate reader manufactured by Thermo Fisher Scientific.

Determination of GK activity

After the experiment was completed, Treg cells were obtained by centrifugation. To determine the activity of glycerol kinase (GK), a colorimetric assay kit (Grace Biotechnology), with all steps adhering to the manufacturer’s protocol. Subsequently, the optical density was accurately measured and recorded using a microplate reader manufactured (Thermo Fisher Scientific).

Motif enrichment analysis

FOXP3 consensus motif enrichment in the GK promoter was predicted using the JASPAR database (http://jaspar.genereg.net/). GC-bias-corrected deviation scores were computed via the chromVAR “computeDeviations” function. Motif enrichment analysis of FOXP3 binding revealed that fragments 12 and 13 exhibited the highest scores. To verify their role in FOXP3 binding, we constructed deletion mutants of GK promoter fragments 12 and 13, respectively. The binding of these fragments to FOXP3 was confirmed by luciferase reporter assays.

Plasmid constructions and mutagenesis

The DNA sequence encoding human FOXP3 was subcloned into the pCMV3 vector, with a FLAG tag fused to the protein. Full-length and deletion-mutant human GK promoters (based on sequences of fragments 12 and 13) were subcloned into the pGL3 vector. Wild-type GK, wild-type H4C1, and mutant H4C1 (K5R) were subcloned into the pCMV-EGFP-3xFlag vector. All plasmids were constructed using the ClonExpress II One Step Cloning Kit (Vazyme, C112-01).

siRNA transfection

GK siRNA was transfected into Tregs using ProteanFect Max (Nanoportalbiotech, PT03) transfection reagent. Transfection complexes were prepared on ice and incubated at 37 °C for 15–30 min in accordance with the manufacturer’s protocol. Following washing, cells were cultured for 24 hours before subsequent analyses.

Luciferase reporter assay

HeLa cells were seeded in a 24-well plate and allowed to reach 80–90% confluency. The cells were then co-transfected with the promoter reporter (GK wild type or GK mutants) along with a gradient of different concentrations of FOXP3-overexpressing plasmid. Renilla luciferase plasmid was included in each well as an internal control. Transfection was performed using an appropriate transfection reagent according to the manufacturer’s instructions. After 24 hours of transfection, the cell culture medium was removed, and the cells were washed once with PBS buffer. Subsequently, 100 μL of 1×Lysis Buffer was added to each well, and the cells were lysed at room temperature for 15 minutes. The lysates were then transferred to 1.5 mL microcentrifuge tubes and centrifuged at 12,000 rpm for 5 minutes to pellet cell debris. The luciferase activity in the supernatant was measured using a Dual-Luciferase Reporter Assay System (Promega) according to the manufacturer’s instructions. The luminescence signal was quantified using a microplate reader. The luciferase activity of each sample was normalized to the Renilla luciferase activity (internal control) to assess the regulatory effects of FOXP3 on the GK promoter.

ChIP-PCR

Treg cells were stimulated with glycerol for 24 hours, whereas 293 T cells were transfected to overexpress pCMV-FOXP3-Flag. Post-transfected, the cells were harvested by centrifugation. For the H3K9la and H4K5la chromatin immunoprecipitation (ChIP) assays, the collected cells were resuspended in 150 μL of ChIP buffer 1 (15 mM Tris-HCl (pH 7.5), 60 mM KCl, 5 mM MgCl₂, 0.1 mM EGTA, and 0.3 M sucrose). Subsequently, 150 μL of buffer 2 (15 mM Tris-HCl (pH 7.5), 60 mM KCl, 5 mM MgCl₂, 0.1 mM EGTA, 0.3 M sucrose, 0.5% NP-40 (v/v), and 1% NaDOC (w/v)), was added to the cell suspension. The mixture was lysed on ice for 15 minutes.

Next, 300 μL of buffer 3 (85 mM Tris-HCl, pH 7.5, 3 mM MgCl₂, 2 mM CaCl₂, 0.3 M sucrose) was incorporated into the lysate, and the sample was pre-warmed at 37 °C for 5 minutes. The reaction system was then supplemented with 0.5 μL of MNase (NEB, M0247S) and immediately vortex-mixed to ensure homogeneous distribution. The enzymatic digestion was carried out at 37 °C for 15 minutes. To halt the reaction, 6 μL of 0.5 M EDTA (pH 7.5) and 6 μL of 100×Roche Complete protease inhibitor cocktail (Roche) was added. After centrifugation at 12,000 rpm for 15 minutes at 4 °C, the collected supernatant was readied for the subsequent chromatin immunoprecipitation steps.

Following the determination and normalization of the DNA content, 1/100 of the volume of each supernatant was collected and designated as the input sample. The remaining supernatants were subjected to immunoprecipitation with either anti-H3K9la antibody (1:100) or anti-H4K5la antibody (1:100). The samples were then placed on a rotator and incubated at 4 °C overnight. ChIP-Grade Protein G Magnetic Beads (CST, #9006, with 3 μL per μg of antibody) were added and incubated at 4 °C for 2 hours. The magnetic beads were isolated using a magnetic frame and washed sequentially: once with 1 mL of low-salt buffer (20 mM Tris-HCl, pH 8.1, 150 mM NaCl, 2 mM EDTA, 0.1% SDS (w/v), 1% Triton X-100 (v/v)), once with 1 mL of high-salt buffer (20 mM Tris-HCl, pH 8.1, 500 mM NaCl, 2 mM EDTA, 0.1% SDS (w/v), 1% Triton X-100 (v/v)), once with 1 mL of LiCl buffer (50 mM Tris-HCl, pH 8.1, 1 mM EDTA, 0.25 M LiCl, 1% NP40 (v/v), 1% NaDOC (w/v)), and twice with 1 mL of TE buffer (10 mM Tris-HCl, pH 8.1, 1 mM EDTA).

After the magnetic beads were washed, 200 μL of elution buffer 1 (1% SDS in TE buffer) was added to elute the protein. The elution was carried out at room temperature on a rotator for 10 minutes, after which the supernatants were collected using a magnetic frame. For decrosslinking, 8 μL of 5 M NaCl was added to the supernatants, and the samples were incubated at 65 °C overnight. Subsequently, 1 μL of RNase A (Sangon Biotech, A005649) was added, and the samples were incubated at 37 °C for 1 hour. Then, 2 μL of proteinase K (Sangon Biotech, A600451), 8 μL of 1 M Tris-HCl (pH 6.5), and 4 μL of 0.5 M EDTA were added to digest the protein at 65 °C overnight. DNA extraction was performed using a DNA purification Kit (TIANGEN, DP204) in accordance with the manufacturer’s instructions. ChIP-qPCR was performed using the primers of the selected genes as listed in Table 2.

Cut & Tag

Treg cells were stimulated with Glycerol for 24 hours. Post-stimulation, the cells were harvested by centrifugation at 600 × g for 5 minutes at room temperature, and the supernatant was carefully aspirated. The Cut&Tag assay was performed using the NovoNGS® Cut&Tag 4.0 High-Sensitivity Kit (for Illumina®, Cat. no. N259-YH01, Novoprotein Scientific Inc.) following the non-formaldehyde crosslinking protocol. Briefly, ConA magnetic beads (10 μL per sample) were retrieved from 4°C, pipetted to disperse clumps, and transferred to 1.5 mL EP tubes. The beads were washed twice with 10×volume of 1×ConA Binding Buffer (prepared by diluting 10×ConA Binding Buffer with ddH₂O), then resuspended in 10 μL of 1×ConA Binding Buffer and equilibrated at room temperature for 5 minutes. The harvested Treg cells were resuspended in 90 μL of 1×Wash Buffer (prepared from the kit’s 10×Wash Buffer (-) plus 50×protease inhibitor and ddH₂O), and 6.5 μL of 5% Digitonin (from the kit) was added to permeabilize the cell membrane, followed by gentle pipetting to mix.

The permeabilized cell suspension was mixed with the pre-washed ConA magnetic beads, and the mixture was incubated at room temperature for 10 minutes with rotary mixing (100 rpm) to ensure efficient cell-bead conjugation. For antibody incubation, 50 μL of pre-cooled Primary Antibody Buffer (2 μL 25×pAb Mix + 50 μL Antibody Buffer; Antibody Buffer: 6.5 μL 5% Digitonin + 650 μL 1 × Wash Buffer) containing anti-H4K5la antibody (1:100 dilution) was added to the cell-bead complexes, and the samples were incubated at 4 °C overnight on a rotator. The next day, the complexes were placed on a magnetic rack to remove the primary antibody supernatant, washed twice with 200 μL of Antibody Buffer (room temperature, 3 minutes each), then incubated with 90 μL of secondary antibody dilution (Goat Anti-Rabbit IgG H&L, Cat. no. N269, Novoprotein; diluted to 10 μg/mL in Antibody Buffer) at room temperature for 1 hour with rotary mixing. After secondary antibody incubation, the complexes were washed twice with 200 μL of ChiTag Buffer (65 mM Tris-HCl pH 7.5, 130 μM protease inhibitor, 13 μM Digitonin; prepared from the kit’s 10×ChiTag Buffer(-) plus 50×protease inhibitor, 5% Digitonin, and ddH₂O), then resuspended in 72 μL of ChiTag transposome dilution (1 μL ChiTag® pAG-Transposome + 80 μL ChiTag Buffer) and incubated at room temperature for 1 hour. For tagmentation, 65 μL of Tagmentation Buffer (70 μL ChiTag Buffer + 1 μL 0.7 M MgCl₂) was added to the complexes, which were then incubated at 37 °C for 1 hour using a PCR thermocycler.

The tagmentation reaction was terminated by adding 5 μL of Stop Buffer and 1 μL of Proteinase K (both from the kit), followed by incubation at 55 °C for 10 minutes. DNA extraction was performed using 2 × volume of Tagment DNA Extract Beads (pre-equilibrated at room temperature for 15 minutes); the mixture was incubated at room temperature for 5 minutes, then placed on a magnetic rack to separate beads. The beads were washed twice with 1 mL of freshly prepared 80% ethanol, air-dried for 4–6 minutes (avoiding over-drying), and DNA was eluted with 22 μL of Elution Buffer. For library construction, a 50 μL PCR system was prepared: 21 μL eluted DNA + 25 μL AmpliMix (from the kit) + 1.5 μL N5 Index primer + 1.5 μL N7 Index primer. The PCR program was set as follows: 72°C for 3 minutes (strand displacement); 98°C for 30 seconds; (98 °C for 5 seconds + 72 °C for 5 seconds) × 12–15 cycles; 72°C for 2 minutes; hold at 10°C. The PCR products were purified with 1.3×volume of DNA Clean Beads (from the kit): the mixture was incubated at room temperature for 5 minutes, washed twice with 200 μL of 80% ethanol, air-dried, and eluted with 20 μL of Elution Buffer. Library quality control was conducted using a Qubit fluorometer for concentration determination and an Agilent Technologies 2100 Bioanalyzer for fragment size analysis (target range: 200–1000 bp). Cut&tag sequencing was performed by Beijing Novogene Bioinformatics Technology Co., Ltd.

Flow cytometry

For surface staining, cells were incubated with antibodies at 4 °C for 30 minutes in FACS buffer (PBS + 2% FBS + 1 mM EDTA). For intracellular staining, cells were fixed in 4% PFA for 15 minutes at room temperature and then in 0.2% Triton-100 for 5 minutes. For intracellular cytokine detection, cells were fixed and permeabilized using the Foxp3/Transcription Factor Staining Buffer Set (eBioscience). FACS data were acquired through the Cytek Aurora cytometer (BD Bioscience) and analyzed using the FlowJo software (Tree Star). All antibodies used in this study are listed in Table 3.

MC38 xenograft tumor models

Logarithmic-phase murine MC38 colon cancer cells were harvested via trypsinization and washed three times with PBS. They were then resuspended in PBS, and 2 × 10⁶ cells were subcutaneously injected into the flanks of 6–8-week-old female mice.

Mice were randomly assigned to experimental groups and allocated to one of two dietary regimens: a standard chow diet (Laboratory Diet 5L0D; 58% of calories from carbohydrates, 13.5% from fat, 28.5% from protein) or a high-fat (HF) diet (37.4% of calories from dietary fat, TD.97222, custom diet, USA).

Therapeutic anti-PD-1 Antibody (pembrolizumab) (100 μg per mouse; Sanyou Biopharmaceuticals Co., Ltd.) or AQP3 inhibitor DFP00173 (50 μg per mouse) was administered intraperitoneally on days 3, 6, 9, 12, 15, and 18 post-cell injection.

Tumors were collected 21 days after tumor cell injection. Tumor volumes were measured using Vernier calipers (length = a, width = b) and calculated with the formula: V = ab²/2. Collected tumors were subsequently used for FACS analysis or Treg isolation.

Oil Red O staining

Human colorectal cancer patient-derived frozen sections were cut to a thickness of 6–8 μm. Fix in formalin (optional, 10 min) followed by rinsing with water. Immerse sections in 70% ethanol for 5 seconds. Subsequently, stain with Oil Red O working solution for 5–10 min. Rinse with 70% ethanol to remove excess dye, then perform 30 seconds of washing under running water. Counterstain with hematoxylin for 2 min, differentiate in hydrochloric acid-alcohol, rinse with distilled water, and blue with a bluing reagent. Blot sections dry with filter paper and mount using glycerin gelatin.

Processing and analysis of single-cell RNA-seq data

Single-cell transcriptomics datasets were collected from public repositories including tumor and adjacent non-tumor liver samples from 5 patients (GSE140228), liver metastasis and adjacent non-tumor liver tissues from 10 patients (GSE164522), primary colorectal tumors and adjacent non-tumor tissues from 10 patients without distant metastasis (GSE146771), 7 paired primary colorectal tumors and adjacent non-tumor colorectal tissues from patients with liver metastasis (E-MTAB-8107), and T cells of PBMC from about 300 patients across 21 cancer types (GSE156728). The scRNA-seq data were aligned and quantified by Cell Ranger (v.5.0.0) with the GRCh38 human reference genome. Cells surpassing the expression threshold of 4000 genes (potentially indicating doublets), falling below 200 expressed genes (suggesting low-quality libraries), or exhibiting mitochondrial UMI counts exceeding 15% (possibly indicative of cell fragments and debris) were excluded from subsequent analysis. Seurat (v4.3.0) was used to perform scRNA-seq data analysis, including data integration, normalization, dimension reduction, and cell clustering. For the visualization of the scRNA-seq data, we employed the t-distributed stochastic neighbor embedding (tSNE) algorithm. Differentially expressed genes (DEGs) were identified through the FindAllMarkers() function and were used to annotate the final clusters. Additionally, these DEGs were used to perform Gene functional enrichment analysis for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and Gene Ontology (GO) analysis. scMetabolism (v2.1.0) with the VISION algorithm was also used to quantify metabolic pathway scores for each cell based on the CancerSEA database and HALLMARK gene sets. The AddModuleScore function was used to score the gene sets related to glycolysis across four cell types.

Cell interaction analysis of Treg and Macrophage

To investigate the intricate network of cellular communication, we employed the CellChat package (version 1.6.1) for cell interaction analysis. Leveraging the extensive ligand-receptor pair data available in CellChatDB, we evaluated the potential interactions among the different cell populations. Specifically, we focused on the datasets pertaining to “secreted signaling,” “ECM-receptor,” and “cell-cell contact” interactions. These selected datasets provided valuable insights into the intricate cell communication occurring between each cluster.

TCR analysis

The TCR data was download form GSE164522. scRepertoire (version 2.2.1) was used to process TCR sequencing data, assign clonotypes based on CDR3 sequences and VDJ genes, and analyze clonal dynamics. Only in-frame TCR a-b pairs were considered to define the dominant TCR of a single cell. Each unique dominant a-b pair was defined as a clonotype. If one clonotype was present in at least two cells, this clonotype would be considered clonal, and the number of cells with such a dominant a-b pair indicated the degree of clonality of the clonotype.

Spatial transcriptomic analysis

Spatial transcriptomic datasets were downloaded from public studies derived from 10x Genomics VISIUM technology. Six slides were left for downstream analysis. Cell2location (version 0.0.5) was first used to perform deconvolution of the spatial transcriptomic data, and gene expression signatures of cell types from an scRNA-seq reference were created. Scanpy.pl.spatial function was used to visualize the infiltration score of each subset. Pearson correlation analysis was applied to calculate the infiltration relationship of different subsets in each spot. EPCAM and KRT8 were used to distinguish the tumor area from the adjacent non-tumor area.

Cut&Tag sequencing and analysis

Three-day in vitro-induced iTregs from human PBMC naive CD4⁺ T cells. and stimulated with Glycerol for 24 hours. Briefly, the Cut&Tag assay was conducted following the manual of hyperactive pG-Tn5/pA-Tn5 transposase for the Cut&Tag kit (for Illumina®, Cat. No. N259-YH01, Novoprotein Scientific Inc.). Library quality control was conducted using a Qubit fluorometer for concentration determination and an Agilent Technologies 2100 Bioanalyzer for fragment size analysis (target range: 200–1000 bp). Cut&tag sequencing was performed by Beijing Novogene Bioinformatics Technology Co., Ltd. The raw sequencing data were subjected to quality trimming using Fastp (version 1.0.1). Reads were aligned to the hg38 genome using Bowtie2 (version2.5.4) and were normalized by calculating the reads per kilobase of transcript per million mapped reads (RPKM). Peak calling was performed using MACS2 (version 2.2.9.1) with FDR < 0.01. Tracks were visualized using Integrative Genomics Viewer (IGV), and peaks were annotated by ChIPseeker (version 1.30.3). Heatmaps were generated using deepTools (version 2.5.7), and the density of histone modifications from 3000 bp upstream of transcription start sites (TSS) to downstream of the transcription end site (TES) was plotted using the plotHeatmap function.

Quantification and statistical analysis

Statistical analysis and graphs were performed using GraphPad Prism 10 software. One-way or Two-way ANOVA with Tukey’s multiple comparisons test was used to evaluate statistical significance. Correlation analysis of the expression levels of the two genes was performed using Pearson correlation analysis. For statistical significance: p ≤ 0.05 was considered significant, and p > 0.05 was considered not significant.

Materials availability

Detailed information on CRC patients, a complete list of primer sequences, and all primary and secondary antibodies used in this study can be found in Table 1, Table 2, and Table 3, respectively.

Author contributions

W.W. conceived the project and mainly designed and performed the experiments. C.S., Y.L., R.L., H.W., Q.L., H.S., Q.H., C.Z., W.H., Y.Z., C.Z., and Y.X. Y.L.S., and X.C. provided experimental assistance for this project. P.W., L.S. and L.Z. collected human samples. W.W. and B.K. contributed to data analysis. C.H. conducted the Spatial transcriptome analysis, the Cut&Tag data analysis, and pan-cancer analysis. K.X. was responsible for the single-cell data analysis, while B.J., H.L.S., S.S. and S.Z. provided intellectual advice and helped with manuscript editing. W.W., B.K. and J.W. supervised the study. W.W. wrote the manuscript with input from all authors. All authors have read and approved the article.

Data availability

All data generated in this study are included in this published article and its supplementary information. The raw Cut&Tag sequencing data generated for this study can be accessed from the Genome Sequence Archive of the Beijing Institute of Genomics (BIG) Data Center, BIG, Chinese Academy of Sciences, under accession number HRA013277 at http://bigd.big.ac.cn/gsa-human. All other study data are included in the article. Uncropped Western blot images can be found in the Supplementary Material. The public scRNA-seq, TCR, and spatial transcriptomics datasets used in this study can be obtained from E-MTAB-8107, GSE164522, GSE140228, GSE146771, GSE245552, GSE298084, GSE156728, and http://biotrainee.cn:19195/CRLM-ST-share/. This paper does not report original code. Any additional information required to re-analyze the data reported in this paper is available from the lead contact upon request.

Competing interests

S.Z., W.W., J.W., B.K., C.S., L.S., G.F., W.H., R.L., L.S., Y.L., H.S., C.Z. and A.C. have filed a provisional patent (202510349047.7) disclosure based on the results in the paper. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41392-026-03039-y.