Glut3 promotes cellular O-GlcNAcylation as a distinctive tumor-supportive feature in Treg cells

Glut3 as a potential candidate with tumor-Treg-specific functional relevance

We investigated the relevance of two T lymphocyte-expressed facilitative glucose transporters, GLUT1 and GLUT3 (encoded by SLC2A1 and SLC2A3, respectively), via in silico methods and analyzed their correlation with Treg and non-Treg infiltration levels across different tumor types. Analysis of the Cancer Genome Atlas (TCGA) human cancer data via TIMER 2.0 (http://timer.cistrome.org) [14] revealed a strong correlation between SLC2A3 expression and Treg infiltration in several queried cancer types. These included skin cutaneous melanoma (SKCM), stomach adenocarcinoma (STAD), and colon adenocarcinoma (COAD) (Fig. S1A, left panel). However, this correlation was lost when we queried for its correlation with CD4⁺FOXP3^- non-Treg infiltration (Fig. S1A, middle panel). No such correlation was observed between SLC2A1 and Treg infiltration in any of these cancer types (Fig. S1A, right panel). Extending this analysis to all 40 different tumor types in the TCGA database via two different algorithms, we established that, across most tumor types, there was a positive correlation specifically between Treg infiltration and SLC2A3 expression (Fig. S1B, C). In concert with these observations from human cancer databases, upon analyzing previously published gene expression data in mouse tumor models [15], we observed a clear trend toward increased Slc2a3 expression in tumor-infiltrated Tregs compared with conventional effector T cells (Fig. S1D). These data are the first indications of a hitherto unexplored role of Glut3 in Tregs, specifically in the context of antitumor immunity.

Specific requirement of Glut3 for Treg-mediated tumor immune tolerance

To elucidate the physiological importance of Glut3 in Tregs, we generated mice harboring the loxP flanked Slc2a3 allele (Slc2a3^f/f) and crossed them with Foxp3^YFP-Cre mice to delete Slc2a3 specifically in Tregs (Fig. 1A, B). The Slc2a3^f/fFoxp3^YFP-Cre (hereafter called ‘KO’ and the littermate control Slc2a3^+/+Foxp3^YFP-Cre mice called ‘WT’) mice did not display any alterations in the total cell frequency of peripheral lymph nodes, mesenteric lymph nodes or the spleen. Non-Treg CD4⁺Foxp3⁻ T-cell activation and proliferation at the steady-state level, both at young (~8 weeks) and older (~36 weeks) ages, did not change (Fig. 1C, D; and Fig. S2A–C). Additionally, there was no significant difference in inflammatory cytokine production between WT and KO CD4⁺Foxp3⁻ and CD8⁺ T cells (Fig. 1E and Fig. S2D, E). Furthermore, we did not observe any differences in Treg frequency, activation, or proliferation or in the frequency of Tregs expressing functional markers such as CD25, CTLA4, GITR, or CD103 (Fig. 1F–H and Fig. S2F, G). In an in vitro suppression assay, KO Tregs were equally as competent as WT Tregs in suppressing the proliferation of CD4⁺Foxp3⁻ responder T cells (Fig. 1I, J). Therefore, the KO mice did not display any signs of compromised Treg function at steady state.

To further determine whether Treg-specific deletion of Glut3 has any phenotypic consequences under conditions of inflammation, we employed two different models of autoimmunity. Experimental autoimmune encephalomyelitis (EAE) (Fig. 1K and Fig. S3A, B), an animal model of multiple sclerosis, and a T-cell transfer model of colitis (Fig. 1L and Fig. S3C–E) were used. In both cases, the KO Tregs remained equally competent in controlling inflammation as WT Tregs.

However, upon challenge with either B16F10 melanoma or MC38 colon carcinoma, the KO mice displayed significant suppression of tumor growth compared with the WT mice (Fig. 2A and Fig. S3F). To this end, we focused on the B16F10 model to gain insights into the cellular and molecular mechanisms of the phenotype. The compromised tumor growth in the KO mice was accompanied by a tendency toward an increase in the tumor-infiltrating CD8⁺/CD4⁺ T-cell ratio (Fig. 2B), as well as a significant increase in the proliferation of CD4⁺Foxp3⁻ and CD8⁺ cytotoxic T cells, specifically in the TME and not in the spleen, as evidenced by Ki67 expression (Fig. 2C, D). Furthermore, there was a significant increase in the frequency of IFNγ-producing CD8⁺ T cells, specifically within tumor-infiltrated lymphocytes (Fig. 2E).

To comprehensively determine whether Treg-specific deletion of Glut3 results in enhanced T-cell activation within tumor sites, we performed RNA-seq analysis of FACS-sorted tumor-infiltrating CD4⁺Foxp3⁻ effector T cells and CD8⁺ cytotoxic T cells from WT and KO mice (Fig. S4A). Since differences in tumor size can influence the intrinsic properties of tumor-infiltrating lymphocytes due to an altered TME, for all these and subsequent TIL-Treg-derived RNA-seq experiments described later, we adjusted the tumor inoculation point so that at the time of TIL isolation and analyses, both the WT and KO mice had a tumor volume of ~500 mm³. As expected, according to gene ontology analysis (using the Database for Annotation, Visualization and Integrated Discovery (DAVID 6.8) software package https://david.ncifcrf.gov/home.jsp) of the “biological process”, both CD8⁺ T and CD4⁺Foxp3⁻ T cells presented increased differential gene expression (DGE) of T-cell activation-related genes and enrichment of pathways such as ‘adaptive immune response’, ‘inflammatory response’, ‘TCR signaling pathway’, and ‘positive regulation of IFN-γ production’ (Fig. 2F–H). Notably, there was a significant reduction in the expression of Ifngr2 in both CD4⁺Foxp3⁻ and CD8⁺ T cells and Tnf in CD8⁺ T cells. Ifngr2 signaling is largely dispensable for the expansion and differentiation of CD8⁺ T cells [16]. Furthermore, TNF can lead to activation-induced death of CD8⁺ T cells [17]. These findings confirmed that the reduction in tumor growth upon Treg-specific depletion of Glut3 was a consequence of enhanced T-cell-mediated antitumor immunity. Consistently, the frequency of intratumoral Tregs but not of splenic Tregs from tumor-bearing mice (henceforth referred to as ‘TBM’) was dramatically lower in the KO animals than in the WT animals (Fig. 2I). This tumor-specific property of Glut3-deficient Treg cells is cell intrinsic, since Glut3-deficient Slc2a3^f/fFoxp3^YFP-Cre/WT heterozygous female mice, in which, owing to X chromosome inactivation, Glut3-sufficient and deficient Tregs coexist, displayed similar reductions in YFP-Cre⁺ TIL-Treg frequency compared with their counterparts in Slc2a3^+/+Foxp3^YFP-Cre/WT littermate controls (Fig. S4B).

Changes in tumor-specific gene expression in Glut3-deficient TIL-Tregs

To determine Glut3-dependent gene expression changes in Tregs, we next analyzed the transcriptomic profiles of WT and KO CD4⁺YFP⁺ Tregs isolated from the spleens of unchallenged mice at steady state, TIL-Tregs sorted from tumor sites, and splenic Tregs sorted from TBMs (Fig. S4A, C). We observed a progressive increase in DGE in Tregs ranging from untreated mouse spleens to TBM splenocytes and TIL-Tregs, where TIL-Tregs were the most affected (Fig. 2J). Intriguingly, 696 genes were downregulated (DOWN) more than 1.5-fold, specifically in TIL-Tregs, whereas 218 genes were downregulated in TBM splenic Tregs. Gene ontology analysis revealed that this group included genes related to the ‘immune system process’, ‘chemotaxis’, ‘adaptive immune response’, and ‘innate immune response’, among others. The majority of the upregulated (UP) genes in the TIL-Tregs, on the other hand, included genes involved in ‘translation’ and ribosome assembly, presumably indicating increased stress-related protein synthesis (Fig. S4D, E). Interestingly, although the number of ‘UP’ genes in TIL-Tregs was close to the number of ‘UP’ genes in TBM splenic Tregs, qualitatively, these genes displayed differences in gene ontology, where the ‘UP’ genes in the TBM spleen were related primarily to the ‘cell cycle’, ‘mitotic nuclear division’, ‘cell division’, ‘DNA replication’, etc. (Fig. S4F, G). Notably, we did not observe an increase in the expression of Slc2a1 or any other Glut-encoding genes in Glut3-deficient Tregs compared with WT Tregs upon manual inspection of the RNA-seq data, indicating that the effect of Glut3 deficiency is unlikely to be compensated for by their increased expression in any of the above groups (Fig. S5).

Since the TIL-Treg ‘DOWN’ gene set appeared to be the most relevant group in which immune function-related genes were primarily affected, we next focused on these genes to obtain more mechanistic insights. For this purpose, we utilized data from a recent study in which tumor-infiltrating Treg gene expression profiles from both mouse and human tumors were compared with those from normal lymphoid organs or tissues. In this study, upon comprehensive bioinformatics analyses, a set of 108 tumor-infiltrating Treg ‘signature’ genes (TITR gene set) whose transcriptional profiles are upregulated in TIL-Tregs within all analyzed tumor types in both species was reported [15]. Initial manual investigation revealed that many TIL-Treg ‘DOWN’ genes are among the tumor Treg signatures identified in this study (Fig. 2K). To achieve a more comprehensive understanding, we performed a cumulative distribution function (CDF) analysis of the difference in expression for the entire tumor Treg signature gene set on our RNA-seq data. A left or right shift in the CDF curve compared with the cumulative distribution for all expressed genes indicates that most genes within the gene set analyzed are downregulated or upregulated, respectively. By quantitatively extending the findings from the manual examination, we detected significant downregulation of most of the tumor Treg signature genes in Glut3-deficient TIL-Tregs. On the other hand, no significant change in the expression of this gene set was observed in the splenic Tregs of KO TBM (Fig. 2L).

In addition to the above observations, upon a closer candidate-based investigation of the TIL-Treg ‘DOWN’ group, we noticed several genes that are known transcriptional targets of the NF-κB pathway, which was scarcely observed in other differentially expressed gene set categories (Fig. 2M and numbers in brackets in Fig. S4D upper panel compared with lower panel, and Fig. S4F). Among these genes, several have been reported to be involved in Treg homeostasis and function. Costimulation via the adhesion molecule CD44 has been reported to be important for Treg cell persistence and the production of IL-10 [18]. Indeed, a deficiency in CD44 leads to reduced Treg cell numbers and an exacerbation of EAE in mice [19]. Similarly, the chemokine receptor CCR5 has been implicated in the intratumoral migration and suppressive function of Tregs [20]. Both Btk [21] and Cd86 [22] are associated with aTregs, and agonistic stimulation of cell surface TLR2 leads to the proliferation of Tregs [23]. Subsequent KEGG pathway analysis revealed that NF-κB-mediated signaling is indeed a major downregulated pathway in Glut3-deficient TIL-Tregs (Fig. 2N). To strengthen these observations, CDF analyses of the gene set comprising all known NF-κB target genes (https://www.bu.edu/nf-kb/gene-resources/target-genes/) revealed significant downregulation of NF-κB targets, specifically in Glut3-deficient TIL-Tregs (Fig. 2O). Interestingly, even though unlike TIL-Tregs, TBM splenic Tregs did not show substantial changes in the expression of the majority of NF-κB target genes, some of the candidates of major importance were found to be significantly downregulated in this group. These include Rel itself, which codes for the NF-κB canonical subunit c-Rel; the regulator Nfkbiz; and the well-known NF-κB target JunB (Fig. 2P). Notably, c-Rel and JunB have been previously implicated in Treg homeostasis and effector functions in the context of tumor immunity [24, 25]. Taken together, these findings led to two important conclusions. First, within the tumor microenvironment, molecular events downstream of Glut3 are required to shape the tumor-specific transcriptional signature specific for TIL-Tregs. Second, Glut3-dependent alterations in NF-κB transcriptional activity appear to be important determinants of this process. Notably, we observed that 16 out of the 108 tumor Treg signature genes are also known transcriptional targets of NF-κB, suggesting a probable mechanistic link between them.

Glut3 promotes O-GlcNAcylation in TIL-Tregs

Next, we sought to determine the Glut3-dependent molecular mechanisms by which the gene expression profile of Tregs is shaped in the tumor environment. While highly suppressive and phenotypically activated, TIL-Tregs are technically challenging to isolate in large numbers with intact physiologic identity, a requirement that is necessary to address such mechanistic questions. Therefore, to mimic the TIL-Treg phenotype in vivo, we utilized a previously characterized model of Treg activation [26]. Briefly, it is a system in which two consecutive days of diphtheria toxin (DT) administration in Foxp3^IRES-DTR mice results in transient Treg ablation followed by overt immune activation. When DT administration stops, there is a rebound in the proliferation of Tregs in an inflamed environment, making them highly activated with superior suppression capacity [26] (Fig. S6A). Notably, the gene expression profile of these in vivo-activated ‘DT-aTreg’ cells has been previously shown [26] and was also evident from our independently generated RNA-seq data (Fig. S6B) to correlate significantly with the gene expression signature of TIL-Tregs infiltrating established tumors in mice [15]. By bulk RNA-seq experiments, Slc2a3 was found to be highly expressed in DT-aTregs (Fig. S6C). Furthermore, in accordance with the increased expression of Slc2a3, in an adoptive cell transfer-based in vivo 2-NBDG (a surrogate fluorescent glucose analog) uptake assay, we found that the relative glucose uptake capacity of DT-aTregs was greater than that of resting Tregs (Fig. S6D, E). Therefore, throughout this study, we utilized DT-aTregs generated by the Foxp3^DTR system to establish hypotheses that were tested in subsequent experiments on TIL-Tregs isolated from tumors.

We performed a metabolomics experiment in which resting Tregs were compared with activated DT-aTregs generated via the Foxp3^DTR system. We identified uridine diphosphate N-acetylglucosamine (UDP-GlcNAc) as a metabolite specifically enriched in DT-aTregs (Fig. S7A). UDP-GlcNAc is the end product of the so-called hexosamine biosynthetic pathway (HBP), an alternative glucose utilization pathway that consumes approximately 5% of glucose uptake in normal cells. UDP-GlcNAc serves as the raw material for the biosynthesis of various glycoconjugated cellular factors. For example, O-GlcNAcylation of downstream targets has been shown to be dynamically modulated in cancer, which is a direct consequence of altered functionality of such posttranslationally modified cellular components [27]. Two key enzymes responsible for adding and removing the O-GlcNAc moiety from target proteins are O-GlcNAc transferase (OGT) and O-GlcNAcase (OGA), respectively (Fig. S7B). Interestingly, a GLUT3-dependent increase in O-GlcNAcylation has been previously reported in colon and lung cancer cells, cultured HeLa cell lines, and the human brain [28, 29]. Furthermore, posttranslational O-GlcNAcylation of the NF-κB subunit c-Rel has been implicated in enhanced transcriptional activation of its target genes [30]. Therefore, we considered that Glut3-dependent enhancement of the HBP and subsequent posttranslational O-GlcNAcylation-mediated alteration of the functionality of cellular factors, such as c-Rel, may be an important molecular event that distinctively shapes the TIL–Treg transcriptional landscape.

We first compared the total O-GlcNAcylation of cellular proteins in resting and activated T cells generated via the Foxp3^DTR system. We found that global O-GlcNAcylation is indeed significantly greater in DT-aTregs than in internal control CD4⁺Foxp3⁻ cells (Fig. 3A). Furthermore, Treg O-GlcNAcylation was reduced in Glut3-deficient Tregs subjected to an identical inflammatory environment upon adoptive transfer in Treg-depleted Foxp3^DTR mice, indicating that, to a great extent, enhanced O-GlcNAcylation is a Glut3-dependent process (Fig. 3B). Next, we wished to determine whether TIL-Tregs display preferentially enhanced O-GlcNAcylation. Lymphocyte enrichment and isolation from tumor sites require a time-consuming delicate protocol, and O-GlcNAcylation is a dynamic and reversible enzymatic process (Fig. S7B). Therefore, to preserve the pan O-GlcNAcylation of cellular proteins, we took advantage of a widely used specific inhibitor of OGA, Thiamet-G35 (TMG) (Fig. S7B and Fig. 3C), which, in the Foxp3^DTR system, was found to further enhance the O-GlcNAcylation of DT-aTregs (Fig. S7C). Indeed, O-GlcNAcylation of Tregs was found to be most prominent at tumor sites, particularly in TMG-treated animals (Fig. 3D).

To ascertain whether Glut3-dependent glucose uptake directly contributes to the HBP via UDP-GlcNAc, we next performed a metabolic-tracing experiment using uniformly labeled ¹³C₆ glucose-containing media. Owing to the requirement of a large number of cells for such stable isotope tracing experiments, we wished to devise an experimental setup using in vitro-generated iTreg cells. For this purpose, utilizing an adoptive transfer cancer model in Rag1^-/- mice, we first asked whether ex vivo-generated iTreg cells lacking Glut3 are also phenotypically compromised in controlling the antitumor immune response. We cotransferred FACS-sorted CD45.2⁺ iTregs (WT or KO) along with congenically marked effector T cells (CD4⁺CD45.1⁺Foxp3⁻ Te and CD8⁺ Tc). The animals were subsequently challenged with B16F10 melanoma cells the following day (Fig. S8A). Consistent with observations in KO Tregs, Glut3-deficient KO-iTregs failed to promote tumor growth (Fig. S8B). This effect was associated with increased proliferation of tumor-infiltrating effector T cells, as evidenced by increased Ki67 expression (Fig. S8C, D). Upon in vitro restimulation, both tumor-infiltrating CD4⁺Foxp3⁻ Te and CD8⁺ Tc cells exhibited significantly increased IFNγ production (Fig. S8E, F). Furthermore, there was a marked reduction in the frequency of the CD4⁺CD45.2⁺ transferred iTreg population at tumor sites (Fig. S8G). Notably, owing to increased CpG methylation, primarily at the conserved noncoding region 2 (CNS2) of the Foxp3 locus, in vitro-generated iTregs are inherently unstable in terms of maintaining long-term expression of Foxp3 [31, 32]. In concert, we observed that, out of the CD4⁺CD45.2⁺ transferred iTreg population, only 30% of the cells retained Foxp3 at the time of analysis. Nevertheless, among the Foxp3⁺ population, a significant decrease in global O-GlcNAcylation was observed in Foxp3⁺ KO-iTregs, similar to that observed in KO-Tregs (Fig. S8H). Taken together, these results validated the suitability of utilizing iTreg cells for in vitro metabolic tracing experiments.

To model the tumor microenvironment in vitro, we performed experiments in TCM generated from overnight cultures of single-cell suspensions of B16F10 tumors resected from tumor-bearing mice. In brief, YFP⁺ iTreg cells generated from magnetically purified CD4⁺CD25^- littermate WT or KO mice were sorted after four days of differentiation, at which point Slc2a3 expression was already more than ~20-fold lower in the KO iTregs than in the WT iTregs (Fig. S8I, J). Sorted iTregs were cultured in 80% TCM for two additional days followed by 24 h of ¹³C₆-glucose labeling. Under this experimental setting, only a small amount of the ¹³C₆ label was found to be incorporated into glycolytic and TCA cycle intermediates such as pyruvate, lactate, malate or citrate (Fig. 3E–H). However, we observed reduced (M+0) lactate in KO iTregs compared with WT iTregs, indicating reduced overall aerobic glycolysis (Fig. 3F). On the other hand, most of the ¹³C label was found to be incorporated into the pentose phosphate pathway intermediate sedoheptulose 7 phosphate, a phenomenon that was significantly greater in KO iTregs (Fig. 3I). Importantly, there was an ~30% reduction in ¹³C-label incorporation of UDP-GlcNAc (M+6) in KO iTregs, indicating that Glut3-mediated glucose uptake is indeed required for optimum metabolic flux toward the HBP.

These results indicate that Glut3-dependent glucose sensing in Tregs promotes the O-GlcNAcylation of cellular proteins in the TME. Given that NF-κB is a known O-GlcNAcylated transcription factor whose targets are preferentially downregulated in Glut3-deficient TIL-Tregs, an intriguing possibility is that Glut3-dependent c-Rel O-GlcNAcylation is a specific determinant of TIL-Treg functionality. To test this, we utilized a technical variant of the so-called proximity ligation assay (PLA), in which the physical proximity of two different antibodies detecting O-GlcNAcylated residue(s) and epitopes of a protein of interest can be detected by confocal microscopy or flow cytometry-based detection of an incorporated fluorescent dye (Fig. S9A). Initial standardization of the protocol using the Foxp3^DTR system indeed demonstrated that c-Rel is preferentially O-GlcNAcylated in DT-aTregs compared with their resting counterparts (Fig. S9B, S9C left and middle panels). TMG treatment resulted in more widespread detection of the PLA signals in Foxp3⁻ non-Treg cells as well, underscoring the specificity of the approach (Fig. S9C, right panel). Next, we used the PLA to determine whether c-Rel is preferentially O-GlcNAcylated in TIL-Tregs. To obtain statistically interpretable data and circumvent other technical difficulties during the tumor lymphocyte isolation process, we employed PLA followed by flow cytometry for these experiments. Indeed, TIL-Tregs from mouse melanoma patients presented significantly greater O-GlcNAcylation of c-Rel than CD4⁺Foxp3⁻ Teff cells did, which was further augmented by TMG treatment (Fig. S9D).

If Glut3-dependent O-GlcNAcylation of downstream factors, an important candidate being c-Rel, is indeed a determining molecular event for TIL-Treg-mediated tumor immune suppression, then experimentally enhancing global O-GlcNAcylation in Glut3-deficient Tregs is expected to substantially rescue TIL-Treg function in KO mice. We reasoned that treating the KO mice with TMG may be instrumental in this goal by restoring the otherwise diminished rate of the forward enzymatic reaction (Fig. S7B). We realize that a potential caveat of this strategy is the possibility that systemic administration of TMG may affect tumor growth by extrinsically affecting other non-Treg populations in general. However, in preliminary experiments in which a TMG treatment regimen was established in WT tumor-bearing mice, we did not observe any difference in overall tumor growth. Therefore, TMG treatment per se did not directly influence the growth rate of B16F10 melanoma cells in wild-type mice within this experimental time frame (Fig. 3C, lower panel).

TMG treatment restored melanoma growth in the KO animals (Fig. 4A). Effector cytokine production by Teff and CD8⁺ T cells mirrored the “rescue” effect of TMG on tumor progression in KO mice. IFNγ, TNF, Granzyme B, and perforin production, which was increased in KO TILs, was reduced to WT levels upon TMG treatment (Fig. 4B–D and Fig. S10A–C). By Ki67 expression analysis, we observed an effect of TMG on the baseline proliferation of all the lymphocyte compartments in both WT and KO mice, and the differences in proliferation between WT and KO Teff and CD8⁺ TILs were alleviated upon TMG treatment (Fig. 4E). In addition, the intratumoral Treg frequency (Fig. 4F), pancellular O-GlcNAcylation and c-Rel O-GlcNAcylation, which were otherwise diminished in TIL-Tregs from KO mice, were restored (Fig. 4G, H).

Finally, we sought to determine whether and to what extent glucose uptake and glycolysis were affected in Glut3-deficient Treg cells compared with WT Tregs, with or without TMG treatment. To assess glucose uptake by TIL-Tregs, we incubated total TILs isolated from different groups with the cell permeable β-glucose fluorescent tracker glucose-Cy5 followed by flow cytometry. Compared with WT Tregs, Glut3-deficient Tregs presented severe defects in glucose uptake. A similar trend was observed when mitochondrial metabolism was blocked with rotenone and antimycin A. Interestingly, while blockade of oxidative phosphorylation with oligomycin strongly promoted glucose uptake in WT-TIL Tregs in the TMG-treated group, glucose uptake was only marginally increased in the KO group (Fig. 4I, J). To determine glycolytic rates, we next performed Seahorse metabolic flux analysis utilizing a similar iTreg-based in vitro experimental system previously introduced in Fig. S8I. Interestingly, both the oxygen consumption rate (OCR) and glycolytic rates were found to be severely diminished in KO iTregs in TCM, with no compensation upon TMG treatment (Fig. 4K, L). Therefore, TMG-mediated rescue of the TIL-Treg phenotype is unlikely to be attributed to compensatory glycolysis, which is mediated by increased glucose uptake.

To determine the extent to which TMG treatment restores gene expression in Glut3-deficient Tregs, we performed RNA-seq analysis on sorted B16F10 melanoma-infiltrating Tregs from control and TMG-treated KO mice. Indeed, by candidate-based (Fig. 5A, B) and CDF analysis (Fig. 5C), we found significant restoration of the expression of NF-κB target genes. Furthermore, several TIL-Treg signature genes, several of which are known targets of NF-κB, were found to be upregulated in TMG-treated KO-Tregs (Fig. 5A). Extending this analysis to global differentially expressed TIL-Treg gene sets (as shown in Fig. 2J), we found that, in addition to known NF-κB targets, the expression of nearly 45–50% of all Glut3-dependent UP or DOWN genes was restored, as is evident by the opposite DGE pattern, in KO-Tregs upon TMG treatment (Fig. 5D‒G). More precisely, 314 out of the 696 Glut3-dependent DOWN genes displayed the opposite trend and were UP in KO TIL-Tregs upon TMG treatment (Fig. 5D). On the other hand, only 8 of these genes were further downregulated in this context (Fig. 5E). Similarly, whereas only 6 Glut3-dependent UP genes were further upregulated (Fig. 5F), 418 of the 802 genes in this category were downregulated after TMG treatment (Fig. 5G). Therefore, Glut3-dependent downstream posttranslational modification-based O-GlcNAcylation of cellular factors appears to be a general mechanism for distinguishing the Treg gene expression network at tumor sites.

Relevance of GLUT3 in human cancers

Next, we explored the prognostic significance of high GLUT3 expression in human TIL-Tregs and analyzed disease-free survival and hazard ratios (HRs) in survival data from human patients with tumors. For this purpose, we used the web-based analysis platform https://kmplot.com [33]. Increased intratumoral SLC2A3 expression in patients with STAD was correlated with a reduced survival probability and increased HR, indicating an unfavorable prognosis. Interestingly, the prognostic outcomes worsened even more when only ‘Treg-enriched’ tumors were included in the analyses. Patients with ‘Treg-decreased’ tumors did not show such a correlation (Fig. 6A, top panel). SLC2A1 expression, on the other hand, was not significantly correlated (Fig. 6A, lower panel). A similar unfavorable prognosis was recorded for other human tumors, such as head and neck cancer (HNSC) and kidney renal clear cell carcinoma (KIRC) (Fig. S11A, B). Moreover, for cancer types such as breast invasive carcinoma (BRCA) and liver hepatocellular carcinoma (LIHC), higher SLC2A3 expression in patients with Treg-deficient tumors was correlated with a favorable prognosis, which was lost in patients with Treg-enriched tumors (Fig. S11C, D). Interestingly, this in silico approach of gathering correlative evidence of SLC2A3 functionality in human TIL-Tregs upon a comparison between the prognostic outcome in Treg-enriched versus Treg-decreased tumors could be extended to a majority of cancer types present in the TCGA database (Fig. 6B). As a proof of concept of this analytical approach, we confirmed that higher expression of a recently identified functionally relevant TIL-Treg-expressed gene, CD36 [34], correlates with unfavorable prognosis (i.e., increased HR), specifically in Treg-enriched human tumors (Fig. S11E, top panel). In contrast, as a negative control, a similar analysis with CLEC9A, a gene not expressed in Tregs whose high expression was previously shown to correlate with favorable prognosis in several human cancers [35], revealed a lower HR across different tumor types. As expected, the prognostic outcome for CLEC9A was not affected by whether Treg-enriched or Treg-depleted tumors were queried (Fig. S11E, lower panel).

We next explored the possibility that increased intratumoral SLC2A3 expression in Tregs is correlated with increased O-GlcNAcylation of cellular proteins in human cancers. For this purpose, we first inspected the expression patterns of OGA and OGT in mice and asked whether they are correlated with a reduced O-GlcNAcylation status in Glut3-deficient TIL-Tregs. Our RNA-seq data revealed a clear trend toward reduced expression of these genes in KO TIL-Tregs compared with WT TIL-Tregs (Fig. S11F). Interestingly, somewhat counterintuitively, despite promoting opposing enzymatic events, the expression of OGA and OGT is subjected to tightly coordinated regulation, as they are sensitive to cellular O-GlcNAc levels. Pharmacological inhibition of OGA has been demonstrated to downregulate OGT expression [36]. Conversely, OGT inhibition [37] or OGT knockdown [38] results in OGA downregulation. Therefore, OGA and OGT levels are dynamically adjusted in tandem to cope with the internal O-GlcNAcylation status of cellular proteins. The mechanism of such coordinated regulation is not well understood. Nevertheless, this observation suggested that the coordinated expression patterns of OGA and OGT may be utilized to predict O-GlcNAcylation status and its plausible correlation with SLC2A3 expression in human cancers. Indeed, TIMER2.0 analysis revealed that in most tumors, SLC2A3 expression is strongly correlated with the expression of OGA and OGT (Fig. S11G). More interestingly, as was observed for SLC2A3 (Fig. S1B, C), a remarkable correlation in the expression of both genes was observed across the majority of cancer types, specifically with the level of Treg infiltration. No such correlation was observed for CD4⁺ non-Treg infiltrates (Fig. 6C). On the basis of these data and our earlier observation of TIL-Treg-specific O-GlcNAcylation of c-Rel, we finally asked whether O-GlcNAcylation of c-REL is a distinctive feature of Tregs at human tumor sites. For this purpose, we analyzed c-REL O-GlcNAcylation in sections of human colon adenocarcinoma (COAD) (Table S1), a relevant cancer type in which high SLC2A3 expression normalized to that of FOXP3 is correlated with a reduced survival probability (Fig. 6D). Indeed, we found that TIL-Tregs in human patient colon adenocarcinoma sections had highly significant c-REL O-GlcNAcylation, as determined by PLA. On the other hand, negligible PLA signals were observed in paired healthy control sections prepared from the same patient (Fig. 6E, F). Therefore, enhanced O-GlcNAcylation of cellular proteins, exemplified by c-REL, appears to be a distinctive feature of tumor-infiltrating Tregs in human cancers. Furthermore, increased SLC2A3 expression in Treg-enriched tumors and its prognostic significance provide strong correlative evidence that altered glucose utilization downstream of GLUT3 critically contributes to this process.

Mice

All genetic model mice were on a C57BL/6 background. Foxp3^IRES-GFP-DTR [56], Foxp3^IRES-YFP-Cre [57], and Foxp3^IRES-Thy1.1 [58] mice are described elsewhere. B6.129S7-Rag1^tm1Mom/J (Rag1^−/−) mice were purchased from Jackson Laboratories. Slc2a3^f/f mice were generated at Biocytogen and crossed with Foxp3^IRES-YFP-Cre mice to generate Treg-specific Slc2a3-deficient mice. Both male and female mice were used for analysis and quantification. All the mice used were 8–15-week-old, except for aged Slc2a3^f/fFoxp3^IRES-YFP-Cre mice, which were 36 weeks of age. All animals were maintained in specific pathogen–free barrier facilities and were used in accordance with protocols approved by the POSTECH Institutional Animal Care and Use Committee.

Mouse tumor models

Both the B16F10 and MC38 cell lines were purchased from ATCC and maintained in DMEM supplemented with 10% FBS. To establish a mouse tumor model, 1 × 10⁵ B16F10 cells or 2 × 10⁵ MC38 colon adenocarcinoma cells were subcutaneously injected into the right flank region on day 0. From day 7 onward, tumor growth was monitored on alternate days. Vernier calipers were used to measure the length and width of the tumors, and tumor volumes were calculated via the following formula: volume (mm³) = length × width × [(length × width)^0.5] × π/6. Euthanasia was performed at the end of the experiment or if the tumor volume exceeded 2000 mm³ or 15 mm in length on any side.

EAE

Sex- and age-matched (6–13-week-old) WT or KO mice were injected with 0.2 ml of emulsion subcutaneously (s.c.) in the shaved backs of the mice (distributed on either side of the midline on the lower back), with 200 μg of MOG_35–55 peptide (MEVGWYRSPFSRVVHLYRNGK, GL Biochem (Shanghai) Ltd.) emulsified in complete Freund’s adjuvant (CFA) (Sigma) containing 400 μg of heat-killed Mycobacterium tuberculosis (Difco). In addition, the mice were administered 600 ng of pertussis toxin (List Biological Laboratories) intraperitoneally (i.p.) on days 0 and 300 ng on day 2. The severity of EAE was monitored every day from day 7 on the basis of a clinical score ranging from 0 to 5: 0 = no clinical signs; 1 = limp tail; 2 = paraparesis (incomplete paralysis of one or two hind limbs, limited mobility); 3 =  paraplegia (complete paralysis of two hind limbs); 4 = paraplegia with forelimb weakness or paralysis; and 5 = death. Spinal cords were harvested from the mice for cellular analysis on days 28–30.

CD4⁺CD45RB^hi T-cell transfer-induced colitis model

Splenic CD4⁺ Treg-depleted cells from Foxp3 reporter mice were enriched with columns (Miltenyi Biotec), and CD4⁺Foxp3⁻CD45RB^hi T cells were FACS-sorted with a BD FACSAria Fusion Flow Cytometer. Treg cells (CD4⁺CD25⁺YFP⁺) from the LN/mLN/spleen of WT or KO mice were directly sorted by FACS after CD4 enrichment. Approximately 7.5 × 10⁵ CD45RB^hi splenocytes alone or with 2.5 × 10⁵ WT or KO Treg cells were transferred into Rag1^−/− recipient mice by i.v. injection. The recipient mice were weighed twice every week to measure the percentage change in body weight. The mice were harvested at the end of the experiment for analysis.

Adoptive T-cell transfer melanoma model

Naïve CD4⁺ T cells were isolated from the lymph nodes and spleens of knockout (KO) and wild-type (WT) mice via the EasySep™ Mouse CD4⁺ T-Cell Isolation Kit (catalog number 19,852; Stemcell Technologies), followed by FACS sorting. FACS-sorted naïve CD4⁺ T cells were then differentiated into inducible T regulatory cells (iTregs) via activation in the presence of 4 ng/mL TGF-β and 50 IU/mL IL-2 for 48 h. The cells were subsequently cultured for an additional 4 days in RPMI supplemented with 50 IU/mL IL-2. Postculture, WT and KO CD4+Foxp3+ iTregs were subjected to FACS. CD4⁺Foxp3⁻ and CD8⁺ T cells from CD45.1Foxp3^GFP mice were used as effector cells. A total of 1 million GFP⁺ iTregs (either KO or WT), 1 million CD8⁺ T cells, and 2 million Foxp3⁻CD4⁺ T cells were mixed and transferred into Rag1^–/– recipient mice [59]. The following day, the mice were subcutaneously implanted with 0.1 million B16F10 melanoma cells in the right flank. Tumor sizes were measured as described earlier. The mice were sacrificed at the end of the experiment for subsequent analysis.

Cellular analysis

To isolate tumor-infiltrated lymphocytes, approximately 500 mg of tumor tissue was chopped into small pieces and incubated in prewarmed 10 ml of RPMI 1640 containing 20 mM HEPES, 40 μg/ml DNAse I (10104159001, Roche), and 500 μg/ml Liberase TL (05401020001, Roche) for 15 min at 37 °C in a shaking incubator (200 rpm). Digestion was stopped by adding 10 ml of ice-cold RPMI 1640 supplemented with 5% FBS. The suspension was passed through a 100 μm strainer, and the remaining tissue chunks were crushed gently over the strainer to release the cells into the media. The cells were pelleted and subjected to gradient separation with 47% Percoll (Cytiva 17-0891-01, Sigma‒Aldrich). The cell pellet was collected after centrifugation and washed with RPMI 1640 containing 10% FBS. If needed, the RBCs were lysed with ACK RBC lysis buffer for two minutes. The cells from spinal cords were isolated in a similar manner except for the Liberase digestion duration was 10 min.

To isolate intestinal lamina propria lymphocytes, the colon and small intestines were collected and rinsed in cold PBS. Peyer’s patches were removed, and the intestines were cut lengthwise and chopped into small pieces (approximately 0.5 inches). The tissue pieces were incubated in a solution of PBS containing 10% FBS, 25 mM HEPES, 1 mM EDTA, 1 mM DTT, penicillin, and streptomycin at 37 °C in an orbital shaker for 10 min. The EDTA solution was then discarded, and the tissues were washed with RPMI containing 5% FBS. The intestinal tissues were minced further and digested in a solution of 1 mg/mL collagenase A and 40 μg/ml DNAse I in RPMI containing 10% FBS and 25 mM HEPES for 10 min at 37 °C in an orbital shaker. The digested tissue was filtered through a 100 µm cell strainer, and lymphocytes were enriched via Percoll gradient separation. The cells were washed and prepared for downstream analysis.

To isolate cells from secondary lymphoid organs (spleen and lymph nodes), the tissues were gently crushed over a 100 µM strainer and washed with complete RPMI supplemented with 10% FBS. The cells were pelleted and resuspended in complete RPMI. For cytokine production, the cells were stimulated for 5 h with PMA and ionomycin in the presence of Golgi Plug (555029, BD).

For flow cytometric analysis, the cells were stained with viability dye (Ghost Dye Violet 510, 13–0870–T500, Tonbo or Fixable Viability Dye eFluor 780, 65–0865–14, eBioscience) in PBS. The cells were subsequently stained for surface markers in 0.01% BSA in PBS. The cells were fixed and permeabilized with Foxp3 staining buffer (00–5523–00, eBioscience) and stained for intracellular proteins. For O-linked N-acetylglucosamine staining, the cells were blocked with a 2.4G2 Fc block for 10 min at room temperature, followed by staining with a primary anti-O-linked N-acetylglucosamine antibody (Clone RL2, dilution 1:400; ab2739, Abcam) for 30 min at 4 °C. The cells were washed and blocked with 2.4G2 Fc block for 10 min, followed by staining with secondary goat anti-mouse IgG1-PE (Polyclonal, dilution 1:500, sc-3738, SCBT) for 30 min at 4 °C. In some cases, an Alexa Fluor 647-conjugated O-GlcNAc monoclonal antibody (clone RL2, dilution 1:400; ab2739, eBioscience™) was used. Finally, the cells were stained for other markers.

The following FACS antibodies were used: Biolegend - Anti-mouse CD45 (Clone 30-F11), CD3 (Clone 17A2), CD4 (Clone RM4-5, Clone GK1.5), CD8α (Clone 53-6.7), CD44 (Clone IM7), CD45.1 (Clone A20), CD45.2 (Clone 104), CD25 (PC61), GITR (Clone DT1), CTLA4 (Clone UC10-4B9), CD103 (Clone 2E7), IFN-γ (Clone XMG1.2), IL-4 (Clone 11B11), IL-17 (Clone TC11-18H10.1), IL-2 (Clone JES6-5H4), Ki-67 (Clone 16A8); anti-human FOXP3–AF488 antibody (Clone 259D); eBioScience – anti-mouse Foxp3 (Clone FJK-16 s); BD Bioscience – anti-mouse CD62L (Clone MEL-14), CD44 (Clone IM7).

In vivo generation of activated Tregs

Six- to eight-week-old Foxp3^IRES-GFP-DTR mice of either sex were injected with 1 µg/animal diphtheria toxin (DT) from Corynebacterium diphtheriae (D-0564, Sigma) intraperitoneally on days 0 and 1 to deplete the Tregs. The animals were sacrificed on day 11 to collect the spleen, and peripheral lymph nodes and lymphocytes were isolated.

In experiments where activation of adoptively transferred Tregs was sought, Foxp3^IRES-GFP-DTR mice were administered DT on Day 0 and Day 1. Tregs were adoptively transferred on Day 1, and DT was subsequently administered every other day until Day 11, when the animals were euthanized, and the tissues were collected.

In vivo 2-NBDG uptake assay

CD45.1-Foxp3^IRES-GFP-DTR mice were injected with 1 µg/animal DT intraperitoneally on day 0 and day 1 to deplete endogenous Tregs. One million FACS-sorted Thy1.1⁺ Tregs from CD45.2- Foxp3^IRES-Thy1.1 mice were adoptively transferred to DT-treated CD45.1-Foxp3^IRES-GFP-DTR host mice on day 1. The animals subsequently received 1 µg/animal DT intraperitoneally on alternate days. On day 11, the animals received a retro-orbital intravenous injection of 100 µL of 1 mg/mL 2-NBDG (11046, Cayman Chemical) or PBS. After 30 min, the animals were sacrificed, the lymphoid organs were collected, and single-cell suspensions were generated. The cells were stained for surface molecules and immediately analyzed by flow cytometry for 2-NBDG uptake.

RNA sequencing

CD4⁺GFP⁺ or CD4⁺YFP⁺ Tregs from Foxp3^IRES-GFP-DTR or Slc2a3^f/fFoxp3^IRES-YFP-Cre mice or Slc2a3^+/+Foxp3^{IRES-YFP-Cre mice}, respectively, were FACS-sorted at >95% purity via a BD Moflo-XDP cell sorter and stored at −80 °C in TRIzol reagent (15596026, Invitrogen). Total RNA was extracted via an RNeasy Micro Kit (74004; Qiagen). The library was prepared with a TruSeq RNA Library Preparation v2 Kit, and sequencing was performed with a SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing according to the manufacturer’s protocols. The quality of the sequenced reads was measured via FastQC (www.bioinformatics.babraham.ac.uk/projects/fastqc/), and low-quality portions of the sequenced reads were trimmed via TrimGalore (version 0.4.2; www.bioinformatics.babraham.ac.uk/projects/trim:galore/). The quality of the RNA-Seq data was evaluated via RSeQC [60] (http://code.google.com/p/rseqc/). The trimmed reads were mapped to the mouse genome (reference assembly mm10) via STAR (version 2.5.1) [61]. The abundance of reference genes was estimated via StringTie, and differential gene expression between conditions was estimated via the RUVSeq and DESeq2 programs, with an FDR-adjusted cutoff P value of 0.05. All genes with an average Log2 TPM value of ≥2 were considered expressed genes.

Metabolomics

aTregs and rTregs were subjected to FACS, and the resulting pellets were snap frozen in Liq. N2 and stored at −80 °C until extraction. Frozen cell pellets (≥ 2 × 106 cells) were thawed on ice and extracted with 1.2 ml of extraction solution (chloroform:methanol: water = 3:4:1 (v/v), 5 nmol/L warfarin spiked in) and vortexed, followed by the addition of 0.15 ml of ice-cold water and vortexing. The cells were homogenized via a BeadRuptor (OMNI International, Kennesaw GA) for 3 cycles, namely, 10 s at 8 m/s each and 3 s in between cycles, and then centrifuged at 13700 rpm at 4 °C for 10 min. A volume of approximately 800 µL aqueous phase was transferred to new tubes for lyophilization, and the organic phase was transferred to new tubes for evaporation under a stream of nitrogen at room temperature. Tubes containing no cells were processed in the same way as blanks. Untargeted lipids and metabolites were analyzed by LC‒MS. The organic phases were reconstituted in 30 µL of dichloromethane/methanol (2:1, v/v). A 6 μl sample was injected into a Vanquish Horizon UPLC system equipped with a Thermo Scientific Accucore Vanquish C18⁺ column (2.1 × 100 mm, 1.5 μm) and eluted at 0.2 ml/min 0–1 min 40% B, 1–15 min 40–100% B, 15–18 min 100% B, 0.3 ml/min 18–22 min 100–40% B (A, acetonitrile: water = 4:6 (v/v), B, acetonitrile:isopropanol = 1:9 (v/v)). A Thermo Scientific Q Exactive hybrid quadrupole-Orbitrap mass spectrometer (QE-MS) equipped with a heated electrospray ionization (HESI) probe was operated via a combined full mass scan and data-dependent MS/MS (full MS/dd-MS2) approach. The HESI probe was run in both positive and negative modes: the heated capillary temperature was held at 320 °C; the sheath gas and aux gas flow rates were set at 45 and 8 units, respectively; and the spray voltage was 3.6 kV for positive mode and 2.5 kV for negative mode. The full MS data were acquired at a resolution of 140,000, the m/z ranged from 100–1500, and the automatic gain control (AGC) target was 1 × 10⁶. The dd-MS2 data were acquired at a resolution of 17,500, an AGC target of 1 × 10⁵, a loop count of 15 and an isolation window of 1.0 m/z.

The aqueous phases were reconstituted in 25 μl of acetonitrile/water (5:95, v/v) containing 10 mM ammonium acetate. A 4 μl sample was injected into a UPLC instrument equipped with a Waters Acquity UPLC BEH amide column (1.0 × 150 mm, 1.7 μm) and eluted at 0.05 ml/min for 0–16.5 min at 99–50% B, 16.5–18 min at 50–1% B, 18–36 min at 1% B, and 36–39 min at 1–99% B (A, water containing 10 mM ammonium acetate; B, acetonitrile). The MS conditions were the same as those used for lipid profiling. The HESI probe was run in both positive and negative modes: the heated capillary temperature was held at 320 °C; the sheath gas and aux gas flow rates were set at 20 and 5 units, respectively; and the spray voltage was 3.6 kV for positive mode and 2.5 kV for negative mode. The full MS data were acquired at a resolution of 140,000, the m/z ranged from 60–900, and the AGC target was 3 × 10⁶. The dd-MS2 data were acquired at a resolution of 35000, an AGC target of 1 × 10⁵, a loop count of 10 and an isolation window of 1.0 m/z. The data were processed via Thermo Fisher Scientific LipidSearch 4.1 and Compound Discover 3.1 as previously described [62].

Analysis of differentially abundant metabolite production

Statistical analysis was carried out via R on the generated table (with the features in the row and samples in the column). The imported table was first transposed, and data normalization was carried out (log10 transformation and autoscaling by using the scale() function of R). Log2FoldChange was calculated between rTreg vs aTreg, and P values were calculated via t-tests. Further features were filtered on the basis of a Log2FoldChange > 1 or <-1, a P value < 0.05, and well-defined feature names (metabolite names), which led to 41 metabolites of interest (26 upregulated and 15 downregulated in aTregs). The upregulated metabolites were subjected to integrated pathway analysis via the IMPaLA web server (http://impala.molgen.mpg.de/) [63].

In vitro suppression assay

CD4⁺YFP⁺ Tregs from Slc2a3^f/fFoxp3^IRES-YFP-Cre and Slc2a3^+/+Foxp3^IRES-YFP-Cre animals and congenically marked naïve responder cells (CD45.1⁺ CD4⁺ Foxp3⁻ CD62L^hiCD44^lo) were FACS sorted. Responder cells were pulsed with 5 μM CTV (C34557, Invitrogen) for 10 min at 37 °C. A total of 1 × 10⁵ T-cell-depleted APCs and 5 × 10⁴ CTV-pulsed responder cells were mixed with the indicated numbers of Tregs and incubated for 96 h in the presence of 0.3 µg/mL anti-mouse CD3e antibody (145-2C11, BioXCell). After incubation, responder cell proliferation was measured as CTV dilution via flow cytometry.

Proximity ligation assay

c-Rel/c-REL and O-GlcNAc proximity ligation assays [64] were performed according to the manufacturer’s protocol via a Duolink® Flow Cytometry Kit or an in situ fluorescence detection kit (Sigma‒Aldrich) with certain modifications.

For the mouse TIL-Treg FACS PLA, CD45⁺ lymphoid cells were enriched with mouse CD45 (TIL) microbeads (130–110–618, Miltenyi Biotec) according to the manufacturer’s protocol. The cells were counted and fixed overnight at 4 °C with Foxp3 staining buffer (00-5523-00, eBioscience). Before the PLA was started, the cells were permeabilized for 5 min with Foxp3 staining buffer (00–5523–00, eBioscience). A total of 100,000 cells were transferred to a V bottom 96-well plate and washed twice with 200 µL of Duolink® Wash Buffer (DUO82047, Sigma‒Aldrich). The cells were blocked with 30–60 µL of Duolink® Blocking Solution (DUO82007, Sigma‒Aldrich) at 37 °C for 60 min. After incubation, the cells were centrifuged at 500 × g, and the blocking solution was removed. The cells were incubated with primary anti-mouse/human c-Rel antibodies (goat polyclonal, dilution 1:40; AF2699, R&D Systems) and anti-O-linked N-acetylglucosamine antibodies (clone RL2, dilution 1:200; ab2739, Abcam) in Duolink® Antibody Diluent (DUO82008, Sigma‒Aldrich) at 4 °C for 60 min. The cells were washed twice with Duolink® Wash Buffer and incubated with 50 µL of 1:5 Duolink® PLA® probe anti-Goat PLUS (DUO82003) and Duolink® PLA® probe anti-mouse MINUS (DUO82004) at 37 °C for 60 min. The oligonucleotides were ligated, amplified, and detected via the Duolink® flowPLA detection kit Red (DUO94001) according to the manufacturer’s instructions. After PLA, the cells were washed with PBS and incubated with anti-mouse CD4–APC/Cy7 (Clone GK1.5, dilution 1:400; 100413, BioLegend), anti-mouse CD8α–BV421 (Clone 53–6.7, dilution 1:400; 100738, BioLegend) and anti-mouse Foxp3–FITC (Clone FJK-16 s, dilution 1:200; 11–5773–82, eBioscience) in PBS at 4 °C for 30 min. The cells were washed with PBS and resuspended in 200 µL of PBS. The samples were analyzed with a BD LSR Fortessa cell analyzer.

For in situ PLA of mouse T cells, CD4⁺ T cells were enriched via the EasySep™ Mouse CD4⁺ T-Cell Isolation Kit (19852, Stemcell Technologies). Approximately 100,000 cells were allowed to adhere to poly-L-lysine-coated glass slides at 37 °C for 30 min in a cell culture incubator. Subsequently, the cells were washed with PBS and fixed at 4 °C with Foxp3 staining buffer for 30 min. Fixed cells were permeabilized for 5 min with Foxp3 staining buffer. The cells were washed twice with Duolink® in situ wash buffer A (DUO82049, Sigma‒Aldrich) and blocked with 50 µL of blocking solution at 37 °C for 60 min in a humidified chamber. After blocking, the cells were incubated with primary antibodies, namely, anti-mouse/human c-Rel antibody (goat polyclonal, dilution 1:40; AF2699, R&D Systems) and anti-O-linked N-acetylglucosamine antibody (clone RL2, dilution 1:200; ab2739, Abcam), in 100 µL of Duolink® Antibody Diluent (DUO82008, Sigma‒Aldrich) at 4 °C for 60 min. The cells were washed twice with wash buffer A and incubated with 50 µL of 1:5 Duolink® PLA® probe anti-Goat PLUS (DUO82003) and Duolink® PLA® probe anti-mouse MINUS (DUO82004) at 37 °C for 60 min in a humidified chamber. The bound probe oligonucleotides were ligated, amplified, and detected via Duolink® In Situ Detection Reagents Red (DUO92008) according to the manufacturer’s instructions. The cells were subsequently washed twice with Duolink® in situ wash buffer B (DUO82049, Sigma‒Aldrich) and once with 0.01X wash buffer B. Subsequently, the cells were incubated with anti-mouse Foxp3–FITC (Clone FJK-16 s, dilution 1:100; 11–5773–82, eBioscience) in PBS at 4 °C for 60 min. The cells were washed twice with PBS and wet mounted with Duolink® In Situ Mounting Medium with DAPI (DUO82040, Sigma‒Aldrich). The fluorescence images were captured via a TCS SP5 II MP with an SMD confocal microscope (Leica Microsystems Ltd.).

To perform in situ PLA on paired human cancer and healthy tissue sections, tissue cryosections were fixed with Foxp3 staining buffer for 60 min at 4 °C. The sections were permeabilized for 10 min with Foxp3 permeabilization buffer for 10 min at room temperature. After being washed twice with wash buffer A, the sections were blocked with 50 µL of Duolink blocking buffer at 37 °C for 60 min in a humidified chamber. Afterwards, the sections were incubated with primary antibodies, namely, anti-mouse/human c-Rel antibody (goat polyclonal, dilution 1:40; AF2699, R&D Systems) and anti-O-linked N-acetylglucosamine antibody (clone RL2, dilution 1:200; ab2739, Abcam), in 100 µL of Duolink® Antibody Diluent (DUO82008, Sigma‒Aldrich) overnight at 4 °C in a humidified chamber. The remaining PLA process was similar to that of in situ PLA of mouse T cells except that the sections were incubated with an anti-human FOXP3–AF488 antibody (Clone 259D, BioLegend). Confocal images were captured by a TCS SP5 II MP with an SMD confocal microscope. Images were analyzed via Leica Application Suite X. Briefly, regions of interest (ROIs) were drawn around each FOXP3 fluorescence overlapping nuclear stain per image. The average MFI of PLA fluorescence per view was calculated from these ROIs.

Visualization of RNA-seq data

Integrated Genomics Viewer (v2.13, https://software.broadinstitute.org/software/igv/) was used to generate browser snapshots via BigWig files. The loss of reads pertaining to Slc2a3 exons 2 and 3 was confirmed via the Treg RNA-Seq data from Slc2a3^f/fFoxp3^IRES-YFP-Cre.

Analysis of the associations of gene expression with Treg infiltration in human tumors

Survival analysis of human patients with high SLC2A3 or SLC2A1 expression was performed via Kaplan‒Meier plotter (https://kmplot.com/) [33] with the Pancancer RNA-seq module. The analyses were classified into Treg-enriched or Treg-decreased tumors for each cancer type by ‘restrict analysis on the basis of cellular content’ options.

The gene module of the TIMER2.0 (http://timer.cistrome.org/) [14] web server was used to assess the associations of glucose transporters (SLC2A3 and SLC2A1) and OGT/OGA pairs, which encode key enzymes for the addition and removal of O-GlcNAc modifications, with Treg and non-Treg CD4 T cells in multiple human solid tumors. TIMER 2.0 can be used to statistically deconvolute expression data for the approximate calculation of immune cell infiltration. Correlation analysis between survival in human colon adenocarcinoma patients and the expression of SLC2A3 and SLC2A1 with and without normalization to the FOXP3 gene was performed via GEPIA2 (http://gepia2.cancer-pku.cn/#index) [65].

Preparation of tumor-conditioned media

B16F10 (0.5 × 10⁶) tumor cells were injected subcutaneously into the flanks of wild-type mice, and tumors were harvested when they reached 1000–1500 mm3 in size. A single-cell suspension was prepared, and 1 × 10⁷ cells/mL were plated in complete RPMI media and incubated overnight in a CO₂ incubator. The culture media were collected after 24 h, centrifuged, and filtered before being stored in aliquots at −80 °C. TCMs thus prepared were assayed for glucose and lactate using a Glucose-Glo Assay Kit (Promega, J6021) and a Lactate-Glo Assay Kit (Promega, J5021), respectively, to confirm an increase in lactate and a decrease in glucose concentration compared with those in RPMI, as a readout of a typical tumor microenvironment, as reported previously [66]. Prior to use, aliquots were thawed and mixed with complete 10% FBS containing RPMI to prepare 80% TCM for immediate use.

In vivo glucose-Cy5 uptake assay

Tumor-infiltrating lymphocytes from B16F10 melanomas in mice were isolated via Liberase as described earlier. Approximately 1 M cells were pretreated with complete RPMI with or without 2 µM oligomycin or 0.5 µM rotenone/antimycin A (both from Agilent, Seahorse, mito stress test kit, 103015-100) for 30 min at 37 °C in a cell culture incubator. The cells were subsequently washed and incubated with 0.4 µM glucose-Cy5 (Sigma, SML3233) in serum-free RPMI with or without 2 µM oligomycin or 0.5 µM rotenone/antimycin A at 37 °C in a cell culture incubator for 25 min. The cells were washed and stained for cell surface markers for 30 min at 4 °C. The cells were analyzed by flow cytometry for glucose‒Cy5 uptake.

Metabolic flux assay

Magnetically purified CD4⁺CD25^- cells were activated in vitro in the presence of 4 ng/ml TGFβ (Novoprotein CA59). After 4 days of induction, iTregs (CD4⁺YFP⁺) were sorted via a FACS cell sorter and then incubated in 80% TCM with or without 10 µM TMG for 48 h. After incubation, the cells were harvested in XF assay medium supplemented with 1 mM pyruvate, 2 mM glutamine and 10 mM glucose for a glycolytic rate assay to determine the glycolytic proton efflux rate (glycoPER), as described previously [13]. Briefly, iTregs were seeded into a Cell-Tak (Corning, 354240)-coated 96-well Seahorse plate at a density of 250 K/well. The cells were treated with 0.5 µM rotenone plus 0.5 µM antimycin A (Rot/AA, rotenone, Selleck, S2348; antimycin A, ENZO, ALX-380-075-M010) and 50 mM 2-deoxyglucose (2-DG, MCE, HY-13966-10 g) following the manufacturer’s instructions.

Stable isotope tracing analysis

Sorted iTregs (CD4⁺YFP⁺) were cultured in 80% TCM for 48 h and then transferred to glucose-free RPMI (Invitrogen) medium supplemented with 10% dialyzed FBS plus tracer U-¹³C glucose (Sigma) for isotope carbon labeling. After 24 h, the cells (3 × 10⁶ cells/sample) were washed with ice-cold PBS. One milliliter of extraction solvent (methanol:water, 4:1, v/v, prefrozen at −80 °C for 1 h) was added to each sample, and the mixture was quenched in liquid nitrogen for 10 mins. The samples were stored at −80 °C for subsequent metabolite detection and analysis.

Untargeted metabolic flux analysis was conducted at LipidALL Technologies as described previously [67]. Polar metabolites were extracted using ice-cold methanol containing phenylhydrazine. The samples were vortexed and kept at −20 °C for 30 min for derivatization of alpha-keto acids [68]. Following derivatization, the samples were incubated at 1500 rpm for 30 min at 4 °C. At the end of the incubation, the samples were centrifuged for 10 min at 12,000 rpm at 4 °C. The clean supernatant was transferred to a new tube and dried in a SpeedVac in H₂O mode. The dried extract was reconstituted in 5% acetonitrile in water prior to LC‒MS analysis on an Agilent 1290 II UPLC coupled with a Sciex 5600+ quadrupole-TOF MS. For reversed-phase chromatography, polar metabolites were separated on a Waters ACQUITY HSS-T3 column (3.0 × 100 mm, 1.8 µm), whereas a Waters ACQUITY BEH amide column (2.1 × 100 mm, 1.7 µm) was utilized for hydrophilic interaction liquid chromatographic separation. The MS parameters for detection were as follows: ESI source voltage negative ion mode −4.5 kV; vaporizer temperature, 500 °C; drying gas (N2) pressure, 50 psi; nebulizer gas (N2) pressure, 50 psi; curtain gas (N2) pressure, 35 psi; and scan range, m/z 60–850 [69]. The information-dependent acquisition mode was used for MS/MS analyses of the metabolites. The collision energy was set at (-) 35 ± 15 eV. Data acquisition and processing were performed via Analyst® TF 1.7.1 software (AB Sciex, Concord, ON, Canada). All the detected ions were extracted via MarkerView 1.3 (AB Sciex, Concord, ON, Canada) in Excel in the form of a two-dimensional matrix, including the mass-to-charge ratio (m/z), retention time and peak areas. PeakView 2.2 (AB Sciex, Concord, ON, Canada) was applied to extract MS/MS data and perform comparisons with the Metabolites database (AB Sciex, Concord, ON, Canada), HMDB and standard references to annotate ion identities [69]. L-Isoleucine-d10 was used as an internal standard for normalization of peak areas across samples.

Real-time PCR

Total RNA from FACS-sorted Treg cells was extracted via the ReliaPrep RNA cell miniprep system (Z6010; Promega) according to the manufacturer’s instructions and quantified via a Nanodrop (Thermo Fisher). cDNA was synthesized from 500 ng of total RNA, oligo(dT) primers (Promega) and Improm-II Reverse Transcriptase (Promega). cDNA was amplified with SYBR Premix Ex Taq (Takara, Japan) via DNA Engine with Rotor-gene Q (Qiagen, Valencia, USA). The mouse Slc2a3 forward primer ATGGGGACAACGAAGGTGAC and the reverse primer CAGGTGCATTGATGACTCCAG were used for Slc2a3 amplification.

Experimental design

No statistical methods were used to predetermine the sample size. Sample sizes were determined on the basis of previous experience with the chosen animal models. For in vivo experiments, age- and sex-matched mice were randomized into treatment groups. The investigator/s were not blinded to group allocation. No data were excluded, and details on experimental replication are given in the respective figure legends.

Statistical analysis

The data are presented as the means ± SDs or the means ±SEMs as mentioned in the figure legends. The data were analyzed via GraphPad Prism v9. Multiple group comparisons in in vivo tumor models were performed via two-way ANOVA. Single comparisons were performed via unpaired two-tailed t-tests or paired two-tailed t tests. Nonparametric analyses for CDF were performed via the Kolmogorov‒Smirnov test. For metabolic labeling experiments, two-sided Dunn’s test was used for post hoc pairwise comparisons. The human tumor patient survival data are presented as Kaplan‒Meier survival curves, and statistical analysis was performed via the Mantel‒Cox log rank test. P values < 0.05 were considered statistically significant.