Stable human regulatory T cells switch to glycolysis following TNF receptor 2 costimulation

Following activation, conventional T (Tconv) cells undergo an mTOR-driven glycolytic switch. Regulatory T (Treg) cells reportedly repress the mTOR pathway and avoid glycolysis. However, here we demonstrate that human thymus-derived Treg (tTreg) cells can become glycolytic in response to tumour necrosis factor receptor 2 (TNFR2) costimulation. This costimulus increases proliferation and induces a glycolytic switch in CD3-activated tTreg cells, but not in Tconv cells. Glycolysis in CD3–TNFR2-activated tTreg cells is driven by PI3-kinase–mTOR signalling and supports tTreg cell identity and suppressive function. In contrast to glycolytic Tconv cells, glycolytic tTreg cells do not show net lactate secretion and shuttle glucose-derived carbon into the tricarboxylic acid cycle. Ex vivo characterization of blood-derived TNFR2hiCD4+CD25hiCD127lo effector T cells, which were FOXP3+IKZF2+, revealed an increase in glucose consumption and intracellular lactate levels, thus identifying them as glycolytic tTreg cells. Our study links TNFR2 costimulation in human tTreg cells to metabolic remodelling, providing an additional avenue for drug targeting. After activation, conventional T cells undergo metabolic reprogramming. de Kivit et al. show that in human thymic regulatory T cells, TNFR2 stimulation promotes a glycolytic switch with a preferential glucose-derived carbon flux into the TCA cycle to support suppressive functions.

T conv cells, CD3-CD28 signalling strongly activates the PI3K-Akt-mTOR pathway, which promotes glycolysis 16,17 . T reg cells, in contrast, are thought to disfavour glycolysis [18][19][20][21][22] . Forced expression of FOXP3 in T conv cells suppresses glycolysis and promotes fatty-acid oxidation (FAO)-fuelled oxidative phosphorylation (OXPHOS) 23,24 . Drivers of glycolysis reportedly antagonize FOXP3 expression and thereby compromise T reg cell stability and function 19,25,26 . However, other studies suggest that T reg cells require mTOR activity for in vivo function 27 and that human T reg cells are highly glycolytic ex vivo 28 and require glycolysis to support FOXP3 expression and function 29 . The complexity of the cell populations studied may explain these discrepancies: tT reg and pT reg cells may have distinct metabolic programmes, and in vitro induced T reg (iT reg ) cells may not faithfully represent pT reg cells. Furthermore, pT reg cells may (partly) convert back to T conv cells in the assays employed.
We here studied the metabolism of pure human tT reg cells and considered that a glycolytic switch may be induced by specific costimulatory receptors. We compared responses to CD28 and tumour necrosis factor receptor 2 (TNFR2, also known as TNFRSF1B or CD120b) costimulation in tT reg and T conv cells. TNFR2 was previously shown to be important for T reg cell responses and protection against autoimmunity in humans and mice 30,31 , and is considered a clinical target for selective T reg expansion or inhibition in transplant rejection, autoimmunity or cancer 9,10 . We here report that CD3-activated tT reg cells selectively respond to TNFR2 costimulation by proliferation and a PI3K-mTOR-driven glycolytic switch that is important for tT reg cell identity and function. We also identify unique elements of the glycolytic programme in tT reg cells and validate our findings in tT reg cells directly isolated from human blood.

Results
A new strategy allows for stable human T reg cell expansion in the absence of rapamycin. Human T reg cells occur in low frequency in the blood and, therefore, expansion protocols are used for clinical application 32 . In such protocols, T reg cells are sorted by flow cytometry and expanded in presence of the mTOR inhibitor rapamycin that selectively inhibits proliferation of contaminating T conv cells 12 . However, since rapamycin affects many aspects of metabolism, these expansion protocols are not suitable to generate T reg cells for metabolic studies. Also, such cultures may still be contaminated with pT reg cells that can convert back to T conv cells and confound data interpretation. We therefore employed a new method to purify stable human tT reg cells, based on the marker glycoprotein A33 (GPA33) 33 .
Among CD4 + T cells, naive T conv cells were purified by flow cytometry on the basis of a CD25 lo CD127 hi CD45RA + GPA33 int , phenotype and naive tT reg cells on the basis of a CD25 hi CD127 lo CD45RA + GPA33 hi phenotype (Extended Data Fig. 1a). Phenotypic analysis of these populations indicated that the naive tT reg cells could be discriminated from T conv cells as previously defined 34 by expression of FOXP3, IKZF2 (HELIOS) and CTLA4 (Extended Data Fig. 1b). Sorted T cells were activated with agonistic monoclonal antibodies against CD3 and CD28 and were expanded in the presence of interleukin-2 (IL-2) for 14 d. Prior to analysis or restimulation, T cells were cultured from day 14 to day 18 in fresh medium with only IL-2 (Extended Data Fig. 2a). Data from multiple donors indicated that T reg cells reactivated with monoclonal antibodies against CD3 and CD28 uniquely expressed FOXP3, IKZF2 and high levels of CTLA4 (Extended Data Fig. 2b,c). Expanded T reg cells suppressed both CD4 + and CD8 + T conv cell proliferation in a conventional suppression assay, wherein the T cells were activated to proliferate with anti-CD3 monoclonal antibody (Extended Data Fig. 2d). These data indicate that this GPA33-based, rapamycin-free expansion protocol yields a functional and stable human T reg cell population of high purity.

Global changes in T conv and tT reg cells on CD3-CD28-mediated activation.
We examined the response of T conv and tT reg cells to activation via CD3 and CD28, as this efficiently brings about the glycolytic switch in T conv cells 15,16 . For this purpose, the two different cell types were generated according to the described expansion protocol. On day 18, the cells were restimulated with anti-CD3-CD28 monoclonal antibodies, in the presence of IL-2. CD28 expression levels were similar on both cell types (Fig. 1a). The expanded cells did not divide, unless they were stimulated with agonist monoclonal antibodies (Fig. 1b), indicating that they had become quiescent after withdrawal of anti-CD3-CD28 monoclonal antibodies on day 14 of the expansion protocol. Although T conv cells already proliferated in response to CD3 triggering only, tT reg cells were largely reliant on CD28 costimulation to do so (Fig. 1b). These data show that, under these in vitro circumstances, tT reg cells depend more on CD28 costimulation than T conv cells do to proliferate, which is in line with earlier findings 35 .
To gain insight into the metabolic programme engaged by T conv and tT reg cells on CD3-CD28-mediated activation, we performed unbiased transcriptomic and metabolomic analyses. Expanded T conv and tT reg cells were rested and restimulated for 24 h with anti-CD3-CD28 monoclonal antibodies or not. Gene-expression profiling revealed that both T conv and tT reg cells readily responded to CD3-CD28-mediated reactivation, as indicated globally by principal component analysis (PCA). For both T conv and tT reg cells, the messenger-RNA profiles of CD3-CD28-stimulated populations were distinct from those of the unstimulated populations (Fig. 1c). Indicative of the functional distinction between T conv and tT reg cells, the genes responding to CD3-CD28-stimulation in both cell populations showed only a partial overlap (Extended Data Fig. 3a). Metabolomic analysis identified 109 water-soluble metabolites, including metabolites in central-carbon, nucleotide, amino-acid, carnitine and redox metabolism. The metabolite profile in CD3-CD28-stimulated T conv cells was distinct from that in the unstimulated populations (Fig. 1c). In tT reg cells, in contrast, CD3-CD28-mediated reactivation induced negligible differences in metabolite levels, as indicated by PCA (Fig. 1c) and the representation of altered metabolites in a Venn diagram (Extended Data Fig. 3a). These data indicate that expanded T conv and tT reg cells respond to CD3-CD28-mediated reactivation by proliferation and changes in gene expression, but tT reg cells do not change their metabolic programme like T conv cells.
T conv but not tT reg cells become overtly glycolytic upon CD3-CD28-mediated activation. We next zoomed in on the glycolytic pathway. In T conv cells, several glycolytic intermediates increased significantly in abundance after CD3-CD28-mediated activation, including hexose phosphate (HexP) (that is, glucose-6P and fructose-6P), glyceraldehyde-3-phosphate, dihydroxyacetone phosphate (DHAP) and lactate. In tT reg cells, however, among glycolytic intermediates only DHAP levels were significantly increased ( Fig. 1d and Extended Data Fig. 4). Assays with the fluorescent glucose analogue 6-NBDG revealed that glucose uptake was increased in both T conv and tT reg cells upon CD3-CD28-mediated activation. However, glucose uptake was significantly higher in T conv cells than in tT reg cells (Fig. 1e).
Glycolytic flux is predominantly controlled by the expression of two glycolytic enzymes, hexokinase (HK) and phosphofructokinase (PFK) 36 . HK2 mRNA expression was significantly increased in T conv cells, but not in tT reg cells on CD3-CD28-mediated activation (Fig. 1f). A proteomics data set of expanded human T conv and T reg cells that we had generated earlier 14 confirmed selective upregulation of HK2 in T conv cells at the protein level after CD3-CD28 stimulation (Fig. 1g). Isoforms of PFK were upregulated at the mRNA level in both activated T conv and tT reg cells, but not at the protein level (Extended Data Fig. 3b,c). CD3-CD28-mediated activation led to immediate acidification of the growth medium and increased lactate secretion by T conv cells, whereas no such changes were detected for tT reg cells (Fig. 3h, i). The combined data indicate that, upon CD3-CD28-mediated activation, T conv cells overtly increase their glycolytic activity, but tT reg cells do not, in agreement with earlier reports 18,37 .
TNFR2 costimulation promotes proliferation and maintains identity of tT reg cells. Many data suggest that TNFR2 costimulation is important for T reg cells and for maintenance of self-tolerance in humans 9,10,38 . A protocol for therapeutic T reg expansion has been suggested, making use of TNFR2-agonist monoclnal antibody 39 . In our hands, in vitro-expanded tT reg cells, but not T conv cells, strongly upregulated cell-surface expression of TNFR2 upon reactivation via CD3 (Fig. 2a). Upon CD3-mediated activation, tT reg cells clearly responded to TNFR2 costimulation by increased proliferation, whereas T conv cells did not (Fig. 2b). Thus, costimulation via CD28 and TNFR2 induced cell proliferation in tT reg cells that were activated via CD3 (Figs. 1b and 2b). TNFR2 activation alone, in the absence of CD3 stimulation, did not induce cell proliferation (Extended Data Fig. 5a).
We next examined the effects of CD3-mediated activation and CD28 or TNFR2 costimulation on tT reg cell identity. As compared with activation via CD3 alone, costimulation via TNFR2 significantly increased FOXP3, IKZF2 and cell-surface CD25 protein expression, while total CTLA4 protein expression was not affected (Fig. 2c,d). TNFR2 activation alone did not affect expression levels of FOXP3, IKZF2 or CTLA4 (Extended Data Fig. 5a). In T conv cells, TNFR2 and CD28 costimulation did not alter FOXP3, IKZF2, CD25 or CTLA4 protein levels as compared with levels   . c, PCA plots of transcriptomic (left, n = 4) and liquid chromatography-mass spectrometry (LC-MS)-based metabolomic (right, n = 4) analyses of expanded T conv and tT reg cells at 24 h after CD3-CD28-mediated restimulation or control (-). d, Volcano plots showing changes in the levels of 109 water-soluble metabolites in T conv (top) and tT reg (bottom) cells upon CD3-CD28-mediated restimulation (n = 4). each circle represents one metabolite, with red colour denoting responding metabolites involved in glycolysis or the TCA cycle. Statistical analysis was done using an unpaired two-sided Student's t-test, and P = 0.05 is indicated by the dotted line. α-KG, α-ketoglutarate; G3P, glyceraldehyde-3-phosphate. e, Left, flow-cytometric analysis of glucose (6-NBDG) uptake at 24 h after restimulation as indicated. The dotted line indicates signal in absence of 6-NBDG. Right, quantification of 6-NBDG-uptake data based on geometric mean fluorescence intensity (geoMFI) normalized to unstimulated T conv cells (n = 5), **P = 0.0067. f, HK2 mRNA levels derived from data set described in c, expressed in transcripts per million (TPM) (n = 4), *P = 0.026. g, HK2 protein levels, expressed in label-free quantification (LFQ) intensity, as detected in an earlier proteome data set 14 (n = 3), *P = 0.0189. h, Left, representative real-time measurement of extracellular acidification rate (eCAR) in T conv and tT reg cell cultures after CD3-CD28-mediated restimulation. Right, quantification of area under the curve (AUC) normalized to the baseline eCAR (n = 3, paired two-sided Student's t-test, **P = 0.007). i, Lactate secretion as measured by LC-MS in the culture medium, expressed as fold increase relative to levels in culture medium without cells (n = 5), *P = 0.0274. e-i, Two-way analysis of variance (ANOVA) with Tukey's post hoc test was used for statistical analysis unless stated otherwise. Data are presented as mean ± s.e.m. n represents cells from individual donors, analysed in independent experiments (a-i). t T re g a n t i-C D 3 t T re g a n t i-C   Normalized read counts were analysed for log 2 (read counts per million (CPM)) and differential expression by log 2 (fold change) (false-discovery rate (FDR) < 0.01). Red and blue dots indicate transcripts with significant differential expression. b, Analysis of genes that were differentially expressed between CD3-and CD3-TNFR2-activated tT reg cells, with enrichment among high-or low-ranked genes. Ranking on the x axis was based on the log 2 (fold change) in gene expression between CD3-CD28-activated and unstimulated T conv cells (grey bar), from the transcriptome data set described in Fig. 1. Genes that were differentially expressed between CD3-and CD3-TNFR2-activated tT reg cells were tested for enrichment (y axis) in the ranked list of genes expressed by T conv cells. Vertical bars correspond to genes that had higher expression levels in either CD3-(blue) or CD3-TNFR2-activated (red) tT reg cells, and are positioned according to the expression of those genes in T conv cells. Differential expression between CD3-and CD3-TNFR2-activated tT reg cells was determined by edgeR normalization and exact test (FDR < 0.01). enrichment is indicated by graphs with scores higher than the dashed horizontal lines (FRY test). c, evaluation of the 1,294 differentially expressed genes in tT reg cells upon CD3-TNFR2-mediated activation versus CD3-mediated activation alone (unpaired two-sided Student's t-test with Benjamini-Hochberg method; FDR < 0.05), showing the top ten significant pathways predicted to be up-or downregulated by IPA (right-tailed Fisher's exact test with Benjamini-Hochberg method; FDR < 0.05). Positive and negative z scores indicate up-and downregulated pathways, respectively. Numbers indicate quantity of differentially expressed genes involved in each pathway. d, Differential expression of indicated genes involved in glycolysis between T conv and tT reg cells that were stimulated for 24 h with agonistic mAbs against CD3, CD3-CD28 or CD3-TNFR2. z scores are colour-coded. Statistical evaluation was performed with an ANOVA with the Benjamini-Hochberg method; FDR < 0.05.
in CD3-stimulated cells (Fig. 2c,d). As FOXP3 mRNA expression is epigenetically maintained by demethylation of the T reg -specific demethylated region (TSDR) in the FOXP3 gene 5 , we analysed the methylation status of the TSDR in tT reg cells that received CD28 or TNFR2 costimulation. As expected 5 , the TSDR was highly demethylated in unstimulated tT reg cells and methylated in T conv cells (Fig. 2e). This epigenetic status of the TSDR in tT reg cells was preserved when the cells were activated via CD3 alone or in conjunction with CD28 and TNFR2 costimulation (Fig. 2e).
These data indicate that costimulation via CD28 or TNFR2 does not destabilize tT reg cells in terms of expression of their master regulator FOXP3. Accordingly, the suppressive capacity of tT reg cells costimulated either via CD28 or TNFR2 was similar to the suppressive capacity of those stimulated via only CD3 (Fig. 2f). From these data, we conclude that TNFR2 delivers costimulatory signals for tT reg cells, while preserving their identity and suppressive function.
TNFR2 costimulation induces expression of glycolysis-driving enzymes in tT reg cells. We next performed transcriptomics to gain insights into the effects of TNFR2 costimulation on tT reg and T conv cells. This analysis showed in an unbiased manner that TNFR2 costimulation strongly affected gene expression in tT reg cells, but not in T conv cells ( Fig. 3a and Extended Data Fig. 6a,b). Strikingly, gene-set enrichment analysis (GSEA) showed that the gene-expression profile of tT reg cells that had been activated via CD3 and TNFR2 became more similar to the gene-expression profile of T conv cells that had been activated via CD3 and CD28 (Fig. 3b). In contrast, the gene-expression profile of tT reg cells that were stimulated via CD3 alone was enriched in the profile of unstimulated T conv cells (Fig. 3b). These data suggest that tT reg cells require TNFR2 costimulation to undergo the same type of transcriptional changes that T conv cells undergo after activation via CD3 and CD28.
Ingenuity pathway analysis (IPA) was performed on the transcriptome data to examine which biological processes were affected in tT reg cells by TNFR2 costimulation. IPA revealed that glycolysis is a significantly upregulated process in tT reg cells upon TNFR2 costimulation (Fig. 3c). We zoomed into these data with the question of how CD3-activated tT reg and T conv cells alter the mRNA expression of glycolysis-pathway components after CD28 versus TNFR2 costimulation. Unsupervised hierarchical clustering ( Fig. 3d) revealed that, after TNFR2 costimulation, tT reg cells strongly upregulated specific molecules involved in glycolysis (cluster 1 and 2), of which some were already highly expressed in CD3-CD28-activated T conv cells (cluster 2). Other molecules involved in glycolysis were differentially expressed in tT reg and T conv cells at the mRNA level, regardless of the stimulus (cluster 3 and 4). These data support the idea that TNFR2 costimulation induces a glycolytic switch in tT reg cells.
tT reg cells become glycolytic on TNFR2 costimulation, but do not show net lactate secretion. We next performed untargeted metabolomics to examine the metabolic changes in T conv and tT reg cells that occurred as a result of TNFR2 costimulation. In tT reg cells, TNFR2 costimulation significantly altered the levels of 26 metabolites, as compared with CD3-mediated activation alone. These metabolites included intermediates of glycolysis, as well as the pentose-phosphate-and nucleotide-synthesis pathways. The main indicators of glycolytic flux fructose-1,6-bisphosphate (F-1,6-BP) and DHAP 36 were among the significantly changed metabolites in TNFR2-costimulated tT reg , but not T conv , cells, suggesting that glycolysis was upregulated by TNFR2 costimulation in tT reg cells specifically. In T conv cells, in contrast, TNFR2 costimulation significantly altered the levels of only seven metabolites ( Fig. 4a and Extended Data Fig. 7). Interestingly, TNFR2 costimulation upregulated the same metabolites in tT reg cells as did CD3 and CD28 stimulation in T conv cells (Fig. 4b).
To confirm that tT reg cells became glycolytic on TNFR2 costimulation, assays using the fluorescent glucose analogue 6-NBDG were performed. Indeed, TNFR2 costimulation significantly increased glucose-uptake activity in CD3-activated tT reg cells (Fig. 4c). Glucose uptake in tT reg cells after CD3-TNFR2-mediated activation reached levels similar to those in T conv cells after CD3-or CD3-and CD28-mediated activation (Fig. 4c). TNFR2 stimulation alone, in the absence of CD3 stimulation, did not increase glucose-uptake activity by tT reg cells (Extended Data Fig. 5b). In T conv cells, increased glucose consumption occurs following translocation of the glucose transporter GLUT1 to the plasma membrane 16,40 . Activation of T conv cells via CD3 resulted in GLUT1 translocation to the plasma membrane, whereas CD28 or TNFR2 costimulation had no additional effect (Fig. 4d). Interestingly, CD3-activated tT reg cells showed only strong GLUT1 translocation to the plasma membrane upon TNFR2 costimulation, whereas CD28 costimulation had a more modest effect.
Strikingly, CD3-TNFR2-mediated activation did not result in immediate acidification of the growth medium ( Fig. 4e) or net lactate release by tT reg cells (Fig. 4f), whereas these events did occur in cultures of CD3-activated T conv cells. Altogether, these results indicate that TNFR2 costimulation induces a glycolytic switch in CD3-activated tT reg cells, as does CD3-mediated activation with or without CD28 costimulation in T conv cells. Yet, tT reg cells appear to employ a different metabolic programme downstream of glycolysis from that used by T conv cells.

TNFR2-costimulated tT reg cells complete the glycolytic pathway.
To determine how tT reg cells metabolize glucose as compared with how T conv cells do this, we performed tracing experiments using [ 13 C 6 ]glucose (Fig. 5a). The data obtained clearly showed that TNFR2 costimulation activates the glycolytic pathway in CD3-activated tT reg cells, since the levels of 13 C-labelled HexP, F-1,6-BP, DHAP, phosphoenolpyruvate (PEP), pyruvate and lactate were all significantly increased after TNFR2 costimulation (Fig. 5b). T conv cells were already glycolytic after CD3 stimulation alone, and TNFR2 costimulation had no significant additional effects (Fig. 5b). Notably, upon TNFR2 costimulation, tT reg cells built up higher intracellular levels of 13 C-labelled pyruvate (P < 0.01) and similar levels of 13 C-labelled lactate as compared with those in CD3-activated T conv cells (Fig. 5b). These results indicate that tT reg cells can produce lactate from glucose and engage the complete glycolytic pathway upon TNFR2 costimulation, but that this does not result in net lactate secretion (Fig. 4e,f).
Both T conv and tT reg cells likely take up extracellular unlabelled ( 12 C) pyruvate from the cell-culture medium and convert this into lactate, as evidenced by the appearance of unlabelled pyruvate and lactate in these cells (M + 0, Extended Data Fig. 8). However, the levels of unlabelled pyruvate and lactate do not increase on stimulation with CD3 or CD3 and TNFR2 in either cell type. These data suggest that TNFR2 costimulation in tT reg cells promotes glycolytic flux, but not catabolism, of extracellular pyruvate.
TNFR2-costimulated tT reg cells feed glucose-derived carbon into the TCA cycle. Both pyruvate and lactate 41 can be further metabolized in the tricarboxylic (TCA) cycle. In T conv cells, stimulation via CD3 or CD3 and TNFR2 did not significantly increase the levels of labelled TCA-cycle intermediates, even though glycolytic flux was increased under these conditions (Fig. 5b,c). In contrast, the increase in glycolysis in TNFR2-costimulated tT reg cells was accompanied by significantly higher levels of labelled TCA-cycle intermediates, compared with those accompanying no stimulation or CD3 stimulation alone. This suggests that in activated tT reg cells, the increased glycolytic flux helps to sustain an increase in TCA-cycle flux. However, in T conv cells, glycolysis appears to be uncoupled from the TCA cycle, and the increase in glycolytic flux instead results in the secretion of lactate (Fig. 4e,f). Of note, whereas the total levels of citrate and cis-aconitate did not differ significantly between activated T conv and tT reg cells, the levels of downstream α-ketoglutarate and malate were significantly lower in tT reg cells compared with T conv cells in all conditions tested (Extended Data Fig. 8). This result suggests that in tT reg cells, part of the produced citrate leaves the TCA cycle for anabolic processes, instead of being converted to α-ketoglutarate. Altogether, these data indicate that tT reg cells costimulated via TNFR2 increase their glycolytic activity, but use a different metabolic programme downstream of glycolysis than do T conv cells.
Glycolysis induced by TNFR2 costimulation supports tT reg cell function. To address the functional consequence of increased glycolysis in tT reg cells following TNFR2 costimulation, restimulation experiments were performed in the presence of 2-deoxy-d-glucose (2-DG), a competitive inhibitor of hexokinase. TNFR2 costimulation significantly increased FOXP3 and IKZF2 expression by tT reg cells (Fig. 2d). However, in the presence of 2-DG, FOXP3 expression was significantly reduced in tT reg cells stimulated via CD3 and TNFR2, while IKZF2 expression remained unaffected (Fig. 6a, b). The expression of CTLA4 was reduced as well (Fig. 6a, b). The presence of 2-DG did not affect the viability of the cells (data not shown).
In activated tT reg cells costimulated via TNFR2 in the presence of 2-DG, reduced FOXP3 and CTLA4 expression was accompanied by reduced suppressive function (Fig. 6c). We therefore conclude that glycolysis supports the identity and suppressive function of CD3-TNFR2-stimulated tT reg cells.
Costimulation via TNFR2 activates the mTOR pathway in tT reg cells. The PI3K-Akt-mTOR pathway is known to be a key driver of glycolysis in T conv cells following activation via CD3 and CD28 (ref. 15 ). Since TNFR2 costimulation increased gene expression of glycolytic enzymes in tT reg cells, we questioned whether TNFR2 costimulation activated the mTOR pathway. IPA predicted that the mTORC1 complex is an active upstream regulator of genes that were differentially expressed in TNFR2-costimulated tT reg cells (z = 2.613, P = 4.82 × 10 −4 ). GSEA confirmed that genes known to be upregulated via the PI3K-Akt-mTORC1 signalling pathway were enriched in CD3-TNFR2-stimulated tT reg cells as compared with CD3 activation alone (Fig. 7a,b). Furthermore, expression of genes involved in nuclear factor-κB (NF-κB) signalling was enriched in CD3-TNFR2-activated tT reg cells (Fig. 7c). These data suggest that PI3K-Akt-mTOR and NF-κB signalling are candidates for driving glycolysis in tT reg cells upon TNFR2 costimulation. We first addressed whether TNFR2 costimulation activates mTOR signalling. Flow-cytometric analysis of phosphorylation of mTOR (Ser2448) and its downstream target ribosomal protein S6 (Ser235/Ser236) showed that TNFR2 costimulation significantly enhanced mTOR signalling in tT reg cells, as compared with CD3 activation alone (Fig. 7d,e). CD28 costimulation weakly activated mTOR and did not alter S6 phosphorylation in CD3-activated tT reg cells. In T conv cells, both mTOR and S6 were phosphorylated following CD28 or TNFR2 costimulation, although this was not significantly higher than following CD3 activation alone (Extended Data Fig. 9a,b). Similar results were observed when freshly isolated, CD3-activated tT reg cells were costimulated via TNFR2, which was accompanied by increased glycolytic activity (Extended Data Fig. 9c-e). These data show that TNFR2 costimulation specifically activates the mTOR pathway in CD3-activated tT reg cells.
To probe the mechanism by which TNFR2 costimulation drives mTOR signalling in tT reg cells, we tested the impact of small-molecule inhibitors for PI3K (LY294002) and NF-κB-inducing kinase (NIK; NIK-SMI1) on the phosphorylation of mTOR and S6. Inhibition of PI3K, but not NIK, abrogated TNFR2-induced mTOR and S6 phosphorylation in CD3-activated tT reg cells (Fig. 7f,g). As expected, blocking PI3K activity in CD3-activated T conv cells decreased phosphorylation of mTOR and S6 (Extended Data Fig. 9f,g). The inhibitors did not affect cell viability (data not shown). Moreover, TNFR2 costimulation did not increase glucose consumption in tT reg cells when mTOR or PI3K signalling was inhibited by rapamycin or LY294002, respectively, whereas NIK inhibition had no effect (Fig. 7h,i). These data indicate that TNFR2 costimulation induces a glycolytic switch in CD3-activated tT reg cells by activation of mTORC1 signalling via PI3K.

TNFR2 expression level identifies glycolytic tT reg cells in vivo.
Finally, we aimed to address the connection between TNFR2 and glycolysis in tT reg cells in vivo. For this purpose, we first discriminated CD4 + CD25 hi CD127 lo putative T reg cells in human blood on the basis of naive and effector phenotype and GPA33 levels (Fig.  8a, top). CD45RA + naive tT reg cells were identified by high GPA33 expression (population 1), and CD45RAeffector phenotype cells could be subdivided into three populations with high, intermediate or low GPA33 expression. TNFR2 expression inversely correlated with GPA33 expression, with the highest TNFR2 expression on CD45RA -GPA33 lo effector cells (population 4) (Fig. 8a, bottom). Currently, there are no defining cell-surface markers for effector tT reg cells, but we considered that a TNFR2 hi phenotype identifies these cells. Within the CD45RA -GPA33 lo population, we sorted two subsets with either high or low TNFR2 expression (Fig. 8b, top). The TNFR2 hi subset was uniformly FOXP3 + and mainly IKZF2 + (Fig. 8c, top, and 8d) and had uniformly high expression of CTLA4 (Fig. 8c, top, and 8e), indicating that this population was primarily composed of effector tT reg cells. The TNFR2 lo subset included FOXP3 -T conv cells, as well as FOXP3 + IKZF2and FOXP3 + IKZF2 + T reg cells (Fig. 8c, top, and 8d). Within the CD4 + CD25 lo CD127 hi G PA33 lo effector T conv cell population, no TNFR2 expression was detected (Fig. 8b, bottom), and cells were predominantly FOXP3 -, IKZF2and CTLA4 - (Fig. 8c, bottom, and 8d,e).
To address whether TNFR2 hi effector T reg cells, as defined in Fig. 8c were glycolytic, we first assessed the expression levels of GLUT1 directly ex vivo. Effector T conv and TNFR2 hi effector T reg cells had higher GLUT1 expression than the naive T conv and T reg cell populations, respectively (Fig. 8f). This was paralleled by increased glucose consumption, as reflected by increased uptake of 6-NBDG (Fig. 8g). These data suggested that in vivo, TNFR2 hi effector T reg cells exhibit increased glycolytic activity. To support this finding, we performed targeted metabolomics in ex-vivo-isolated TNFR2 hi effector T reg cells. Compared with naive tT reg cells, TNFR2 hi effector T reg cells had increased levels of intracellular F-1,6-BP and lactate (Fig. 8h). We therefore conclude that a TNFR2 hi phenotype identifies effector tT reg cells in human blood with high glycolytic activity.

Discussion
Rapidly dividing cells switch their metabolic reliance from OXPHOS to aerobic glycolysis, which is less efficient in generating ATP, but glucose-derived carbon is used for the generation of nucleotides, amino acids and lipids that rapidly dividing cells require 42 . The glycolytic pathway and the connected TCA cycle also generate metabolites that support anabolic processes, or act as messengers or cofactors to direct cellular differentiation and function 43 . We here investigated whether well-defined, human thymus-derived T reg cells can undergo a glycolytic switch when they are induced to proliferate.
T conv cells switch to glycolysis upon TCR-CD3-mediated activation. CD28 is a costimulatory receptor that enforces the CD3 signal and promotes T-cell cycling, survival and glycolysis 16,44 . We found, in agreement with published data, that in vitro, human tT reg cells do not efficiently proliferate upon CD3-mediated activation, but do so when additionally costimulated via CD28 (ref. 35 ). Our data on glucose uptake, GLUT1 translocation, metabolomics and lactate secretion indicate that tT reg cells do not become overtly glycolytic under these conditions, in contrast to T conv cells. Transcriptome analysis supported this conclusion, since expression of glycolytic enzymes was low in CD3-activated tT reg cells and did not change upon CD28 costimulation. Hence, our analyses suggest that human tT reg and T conv cells use different metabolic programmes following CD3-CD28-mediated activation.
We found that, upon activation with anti-CD3 antibody, tT reg cells responded to TNFR2 costimulation, whereas T conv cells were inert to this costimulus. CD3-activated tT reg cells switched to glycolysis upon TNFR2 costimulation, but not upon CD28 costimulation, while they proliferated following both. It is not known how these distinct costimulatory pathways are used by human tT reg cells in vivo. CTLA4 serves to attenuate CD28 costimulation 45 , suggesting that activated tT reg cells that express high CTLA4 levels may avoid this costimulatory pathway. Our finding that high TNFR2 levels identify highly glycolytic effector phenotype tT reg cells in human blood provided in vivo relevance for our findings and points to an important role of TNFR2 in human tT reg cell physiology. TNFR2 signalling was shown by genetic studies in human and mice to be     = 6). b, Top, sorting strategy for the isolation of TNFR2 hi and TNFR2 lo cells from the CD45RA -GPA33 lo putative effector T reg cell population (population 4). Bottom, TNFR2 and CD25 expression on CD4 + CD25 lo CD127 hi CD45RA -GPA33 lo effector T conv cells (representative of n = 6). c, Flow-cytometric analysis showing FOXP3, IKZF2 and total CTLA4 expression within the TNFR2 hi and TNFR2 lo subpopulations of the putative effector T reg cell population (top) and on effector T conv cells (bottom), as described in b (representative of n = 6). d,e, Quantification of data shown in c (n = 6). f, evaluation of total GLUT1 levels in the indicated T conv and T reg cell populations, based on the MFI as analysed by flow cytometry (n = 3), **P = 0.0021. g, Flow-cytometric assessment of 6-NBDG uptake in the indicated T conv and T reg cell populations, as analysed directly following cell sorting (n = 2 for naive T conv cells, n = 6 for effector T conv , naive tT reg and TNFR2 hi effector T reg cells), ****P = 3.25 × 10 −5 . h, Analysis of the intracellular levels of F-1,6-BP and lactate as determined by targeted metabolomics in freshly isolated naive tTreg cells and TNFR2 hi effector T reg cells (n = 3 independent donors), *P = 0.0283, **P = 0.0075. f-h, An unpaired two-sided Student's t test was used for statistical analysis of naive tT reg cells and TNFR2 hi effector T reg cells. Data are presented as mean ± s.e.m. Sample size (n) represents cells from individual donors, analysed in independent experiments (a-g).
important for maintenance of self-tolerance 30,31 . In addition, TNFR2 was shown to be uniquely important for driving T reg responses and is therefore proposed as a clinical target 9,10 . TNFR2, as opposed to TNFR1, preferentially responds to membrane-bound TNF rather than to soluble TNF 46 . Currently, the tissue context in which T reg cells receive costimulatory input via TNFR2 is not known, except that membrane-bound TNF on T reg cells themselves 47 and on tolerogenic monocyte-derived dendritic cells 48 can promote T reg cell expansion. TNFR2 costimulation also drives T reg expansion in vivo in mice 49 . Agonistic monoclonal antibody against TNFR2 has therefore been used to expand CD3-activated human T reg cells in vitro 39 .
In expansion protocols of human tT reg cells intended for therapeutic purposes, rapamycin is included to avoid outgrowth of contaminating T conv cells 11 . Recently, we identified the type I transmembrane molecule GPA33 as a novel surface marker expressed exclusively by naive tT reg cells 14 that can be used for isolation of these cells from human peripheral blood 33 . GPA33 is an Ig superfamily member originally found predominantly on colon carcinoma cells 50 , but its function in T cells is unknown. In the current study, we have used GPA33 in a new purification protocol for human tT reg cells allowing in-depth metabolic analyses in these cells without confounding effects of rapamycin on their metabolism. The current literature data on human T reg cell metabolism is confusing, because different (mixed) cell populations are studied. For example, freshly isolated human T reg cells stimulated via CD3 and CD28 were found to rely on both glycolysis and fatty-acid oxidation 28 . In that study, T reg cells partially lost their suppressive function and FOXP3 expression upon stimulation, which suggests presence of pT reg cells in the test material. In in vitro-induced T reg cells, in contrast, glycolysis supported FOXP3 expression after CD3-CD28-mediated activation 29 . We demonstrate that TNFR2 costimulation reinforces tT reg cell identity by upregulation of FOXP3 and IKZF2 and find that glycolysis is required for this effect, as well as for the suppressive function of TNFR2-costimulated tT reg cells. Interestingly, TNFR2-costimulated tT reg cells relied on PI3K and mTOR to activate glycolysis, as do T conv cells after CD3 and CD28 stimulation 16,40 . TNFR2 uses a very different signalling mechanism than CD3 and CD28 do, which is based on TRAF and ubiquitin signalling and activity of serine-threonine kinase, rather than tyrosine-kinase activity 38,41 . TNF receptor family members activate canonical and non-canonical NF-κB signalling via NIK, which is linked to cell survival 9,10,38 . We found no impact of NIK inhibition on the TNFR2-induced glycolytic switch in T reg cells. Further study will need to point out how activation of PI3K-mTOR signalling is enabled in tT reg cells by CD3-TNFR2-mediated activation.
We found that in glycolytic tT reg cells, the fate of glucose-derived carbon was different than in glycolytic T conv cells. TNFR2-costimulated tT reg cells produced lactate from glucose, but did not show net lactate secretion. Lactate is actively shuttled in and out of cells by specific membrane transporters and can be used in the TCA cycle or other metabolic pathways. However, it can also act as an intracellular or extracellular signalling molecule and, among other activities, modulate immune-cell function by regulating gene expression 51 . In glycolytic tT reg cells, the increase in glycolysis was coupled to an increased shuttling of glycolysis intermediates into the TCA cycle.
The TCA cycle also generates precursors for a number of biosynthetic pathways. Citrate can exit the mitochondria and can be converted to acetyl-CoA in the cytosol by ATP citrate lyase. Thereby, citrate serves as an important precursor for both fatty-acid synthesis and the mevalonate pathway that produces cholesterol 52,53 . Glycolytic tT reg and T conv cells had similar levels of citrate, but the levels of downstream TCA-cycle intermediates such as α-ketoglutarate and malate were significantly lower in tT reg cells, suggesting that part of the produced citrate is indeed funnelled into anabolic processes. In murine T reg cells, the mevalonate pathway was shown to be important in coordinating T reg proliferation, suppressive capacity and lineage stability 54,55 . Moreover, mTORC1 was recently shown to be important for upregulation of mitochondrial metabolic pathways, including the TCA cycle 27 , and to promote cholesterol synthesis via the mevalonate pathway in T reg cells 55 . We propose that TNFR2 costimulation in tT reg cells may support the mevalonate pathway by upregulating TCA-cycle activity. Our transcriptome analysis indeed suggests that TNFR2 costimulation supports cholesterol biosynthesis and geranylgeranyl diphosphate biosynthesis pathways in proliferating tT reg cells, both of which require the mevalonate pathway.
In conclusion, our study identified a new role for TNFR2 costimulation in regulating glucose metabolism in human tT reg cells. In addition, we provide evidence that tT reg cells have metabolic adaptions downstream of glycolysis that may be related to tT reg -cell functionality. Further understanding of the key signalling events and metabolic adaptions in tT reg cells following TNFR2 activation may reveal unique targets to specifically modulate tT reg cell function in transplant rejection, autoimmunity and cancer.

Methods
Cell isolation and flow cytometric sorting. Human materials were obtained in accordance with the Declaration of Helsinki and the Dutch rules with respect to the use of human materials from volunteer donors. Buffy coats from healthy anonymized male donors were obtained after their written informed consent, as approved by Sanquin's internal ethical board. Human PBMCs were isolated from buffy coats using Ficoll-Paque Plus density gradient centrifugation (GE Healthcare). Subsequently, total CD4 + T cells were isolated by using CD4 magnetic MicroBeads (MACS, Miltenyi Biotec) according to the manufacturer's protocol. Alternatively, CD4 + T cells were isolated directly from the buffy coat using the StraightFrom Buffy Coat CD4 MicroBead kit (Miltenyi Biotec). For sorting, T conv and tT reg cells were stained with combinations of CD4-PE-Cy7/CD4-BB700, CD127-BV421 (BioLegend), CD25-PE (BD Biosciences), CD45RA-FITC/ CD45RA-APC-Cy7/CD45RA-BV650, GPA33-AlexaFluor647 (ref. 50 ) and TNFR2-PE-Cy7 monoclonal antibodies as indicated. Detailed information regarding these antibodies can be found in the Reporting Summary. Cells were sorted on a MoFlo Astrios using Summit software version 6.2 (Beckman Coulter) or BD FACS Aria II using FACSDiva software version 8 (BD Biosciences). Propidium iodide (PI) (Sigma) or the Near-IR Dead Cell Stain Kit (Invitrogen) was used to exclude dead cells.
Suppression assay. Whole PBMCs were labelled using CellTrace-Violet (CTV; Invitrogen). In short, PBMCs were washed and resuspended in PBS and incubated for 8 min using 5 μM CTV. Following labelling, an equivalent volume of FCS was added, and cells were washed twice in IMDM/8% FCS. Labelled PBMCs were cocultured with expanded tT reg cells that were prestimulated for 24 h or were not, as indicated. Cell cultures were stimulated using anti-CD3 monoclonal antibodies (0.05 μg ml -1 ). Proliferation of CD4 + and CD8 + T cells was analysed after 4 d of coculture by flow cytometry on a BD LSR Fortessa or BD LSR II cell analyser. For short-term suppression assays, tT reg cells were prestimulated via CD3 and TNFR2 in the presence or absence of 2-DG (25 mM) for 24 h. Equal numbers of live cells were cocultured overnight with CTV-labelled PBMCs. Cocultures were stained using CD3-FITC, CD69-PerCP-Cy5.5 and CD25-PE monoclonal antibodies. Near-IR Dead Cell Stain Kit (Invitrogen) was used to exclude dead cells. Detailed information regarding these antibodies can be found in the Reporting Summary. The percentage suppression was determined using the formula: 100 -((MFI of CD25 or CD69 in presence of T reg cells)/(MFI of CD25 or CD69 in absence of T reg cells)) × 100. Flow cytometry was performed using a BD LSR Fortessa or BD LSR II cell analyser (BD Biosciences). Data were analysed using FlowJo software version 10.5.3. For gating strategies, see Supplementary Fig. 1.
Transcriptomics. Expanded T cells (1 × 10 5 ) were restimulated for 24 h, washed in ice-cold PBS and resuspended in RLT buffer (Qiagen). Total RNA isolation was performed according to the manufacturer's protocol using the RNeasy MinElute Cleanup Kit (Qiagen), including an on-column DNAse digestion (Qiagen). Quality and quantity of the total RNA were assessed on a 2100 Bioanalyzer using a Nano chip (Agilent). From RNA samples with a measured RNA Integrity Number (RIN) between 8.0 and 10.0, strand-specific libraries were generated using the TruSeq Stranded mRNA sample preparation kit (Illumina), according to manufacturer's instructions (Illumina, part no. 15031047 Rev. E). Polyadenylated RNA from intact total RNA was purified using oligo-dT beads. Following purification, the RNA was fragmented, randomly primed and reverse-transcribed using SuperScript II Reverse Transcriptase (Invitrogen). Second-strand synthesis was performed using polymerase I and RNase H with replacement of dTTP for dUTP. The generated complementary DNA fragments were 3′-end adenylated and ligated to Illumina paired-end sequencing adapters and subsequently amplified by 12 cycles of PCR. The libraries were analysed on a 2100 Bioanalyzer using a 7500 chip (Agilent), diluted and pooled in equimolar ratios into a multiplex sequencing pool. The pooled libraries were eventually sequenced with 65-base single reads on a HiSeq2500 using V4 chemistry (Illumina).
RNA-sequencing analysis. The 65-bp single-end reads were mapped to the human reference genome (hg38) using TopHat (version 2.1.0), which allows exon-exon splice junctions to be spanned. TopHat was supplied with a known set of gene models based on Ensembl gene transfer format (GTF) version 77. The samples were generated using a stranded protocol, which means that TopHat was guided to use the first strand as the library type. Furthermore, TopHat was run with Bowtie 1 and uses the prefilter multihits and no coverage as additional arguments. In order to count the number of reads per gene, a custom script (Itreecount) was used. This script is based on the same ideas as HTSeq-count and has comparable output. Itreecount generates a list of the total number of uniquely mapped sequencing reads for each gene that is present in the GTF file.
Differential-expression analysis and hierarchical clustering were performed in Qlucore Omics Explorer (version 3.4), using the trimmed mean of log expression ratios method (TMM). GRCh38.77.gtf was used as reference genome for alignment. Genes were excluded from downstream expression analysis if they failed to have ten reads in at least six samples. Differential expression with a two-group comparison (Student's t test with the Benjamini-Hochberg method for multiple testing correction) was considered significant at FDR < 0.05. For a multigroup comparison (ANOVA with the Benjamini-Hochberg method), the threshold was set at FDR < 0.005. PCA was used to visualize the data set in three-dimensional space, after filtering out variables with low overall variance due to the impact of noise and centering and scaling the remaining variables to zero mean and unit variance.
TPM were calculated following read count and length normalization for individual genes for each RNA-sequencing sample using R (version 3.5.1). Only the genes with a CPM value >2 in all samples were subjected to differential-expression analysis. MA plots were generated following edgeR (version 3.8.6) normalization and exact test 57,58 , in which genes with FDR < 0.01 were considered differentially expressed. The FRY test 59 (limma 60 , version 3.22.7), which is a fast approximation to ROAST, was used to test for enrichment of a-priori-defined gene sets among genes that are high-or low-ranked on the basis of differential expression between two biological states. This test was performed on edgeR-normalized data to compare the gene-expression profiles of CD3-or CD3-TNFR2-activated tT reg cells with those of unstimulated or CD3-CD28-activated T conv cells. Data were depicted as a barcode plot including an enrichment score. In addition, GSEA software (version 4.0.0, http://broadinstitute.org/gsea) was employed to test the hallmark gene-set collection (Molecular Signatures Database; MSigDB 61 ) for enrichment in the gene-expression profiles of CD3-or CD3-TNFR2-activated tT reg cells, using CPM, to identify involved biological processes.
Along with GSEA, IPA (version 52912811, Qiagen) was used to identify biological processes that were affected by differentially expressed genes between tT reg cells that were activated via CD3 or CD3 and TNFR2 for 24 h as determined by Qlucore Omics Explorer (FDR < 0.05). Pathways were tested for enrichment using the right-tailed Fisher's exact test with the Benjamini-Hochberg method (FDR < 0.05). Detailed information regarding software can be found in the Reporting Summary.
LC-MS and metabolomics. Expanded T cells (5 × 10 5 ) were restimulated, collected after 24 h and centrifuged for 5 min at 1,000g. Medium samples were collected and cell pellets were washed with ice-cold PBS. For metabolic-flux analysis, medium was formulated to match the composition of IMDM. Medium consisted of DMEM (lacking glucose and pyruvate), supplemented with additional non-essential amino acids, 1 mM pyruvate, 25 mM U-13 C 6 -labelled d-glucose (Cambridge Isotopes), 8% FCS (Sigma) and penicillin/streptomycin. T cells (5 × 10 5 ) were cultured at a density of 1 × 10 6 cells ml -1 for 24 h prior to collection. Metabolites were extracted by adding 50 μl ice-cold MS lysis buffer (2:2:1 methanol/acetonitrile/ ultrapure LC-MS-grade water) to the cell pellet. Extracellular metabolites were extracted by adding 25 μl medium to 100 μl methanol/acetonitrile (1:1). Samples were shaken for 10 min at 4 °C and centrifuged at 14,000g for 15 min at 4 °C, and supernatants were collected for LC-MS analysis. LC-MS analysis was performed on an Exactive mass spectrometer (Thermo Scientific) coupled to a Dionex Ultimate 3000 autosampler and pump (Thermo Scientific). The MS operated in polarity-switching mode with spray voltages of 4.5 kV and −3.5 kV. Metabolites were separated using a Sequant ZIC-pHILIC column (2.1 × 150 mm, 5 μm, guard column 2.1 × 20 mm, 5 μm; Merck) using a linear gradient of acetonitrile and eluent A (20 mM (NH 4 ) 2 CO 3 , 0.1% NH 4 OH in ultrapure LC-MS-grade water; Biosolve). Flow rate was set at 150 μl min -1 . Metabolites were identified on the basis of exact mass within 5 ppm and were further validated by concordance with retention times of standards. Metabolites were quantified using LCquan software (Thermo Scientific, version 2.9). Cells were resuspended at equal cell densities, and no cell growth occurred in T cells up to 24 h after restimulation. Therefore, samples were assumed to contain equal cell numbers. To correct for technical variations during mass-spectrometry analysis, peak areas of intracellular metabolites were additionally normalized on the basais of total peak intensity of all identified metabolites (untargeted metabolomics) or essential amino acids (metabolic-flux analysis). Isotopomer distributions were corrected for natural abundance.
For targeted metabolomics of freshly isolated T cells (5 × 10 5 ) (Fig. 8h), metabolites were extracted from cells as indicated above, with 2:2:1 methanol/ acetonitrile/water. The supernatant was dried under nitrogen, reconstituted in 60 μl water and ultrasonicated for 5 min. Per sample, 10 μl was used for analysis. According to Ross et. al. 62 , ultra-high-performance liquid chromatography was done using a Phenomenex Aqua C18 column (Phenomenex) on a Nexera X2 system (Shimadzu). The column was eluted with a gradient of water with 0.1% formic acid (LC-MS 98%, Honeywell, Fluka; eluent A) and acetonitrile with 0.1% formic acid (eluent B) with a column flow of 400 µl min -1 as follows: 0 min 0% B, 3.5 min 0% B, 4 min 90% B, 5 min 90% B. Ensuing MS was performed on a Sciex TripleTOF 6600 (AB Sciex) operated in negative ESI mode, with ion source gas 1, 45 psi; ion source gas 2, 50 psi; curtain gas, 35 psi; temperature, 500 °C; acquisition range, m/z 50-1,000; ion-spray voltage, −4,500 V; declustering potential, −80.0 V. An information-dependent acquisition (IDA) scan was done to confirm the identity of the glycolysis intermediates, with the following conditions: collision energy, −10; acquisition time, 250 ms. For tandem MS analysis: collision energy, −30; collision energy spread, 15; ion-release delay, 30; ion-release width, 14; acquisition time, 40 ms. The IDA switching criteria were set as follows: for ions smaller than m/z 600, which exceed 200 counts per second, never exclude former target ions, exclude isotopes within 2 Da, maximum number of candidate ions: 20. Standards for quality control of F-1,6-BP and lactate were purchased from Merck Chemicals. Tandem mass spectra were curated with PeakView (version 2.2.0, AB Sciex) and manually compared with spectra in the Metlin metabolite database 63,64 . MultiQuant (version 3.0.3. AB Sciex) was used for peak integration. Peak areas were normalized to the cell count obtained after cell sorting.
Glucose-uptake assay. T cells (5 × 10 4 ) were stimulated using CD3-, CD28-or TNFR2-activating monoclonal antibodies for 24 h, as indicated. Cells were washed in DMEM without glucose and phenol red (Gibco), followed by incubation with 6-NBDG (100 μM, Invitrogen) in DMEM without glucose and phenol red for 45 min at 37 °C and 5% CO 2 . For 6-NBDG-uptake assays using freshly sorted cells, T cells were incubated directly after flow-cytometric sorting for 1 h in DMEM without glucose and phenol red, supplemented with 1% FCS. For glucose uptake, 6-NBDG was added directly to the culture for an additional 45 min. Cells were washed once in PBS with 2% FCS, followed by acquisition of the samples using a BD LSR Fortessa or BD LSR II cell analyser. DAPI (Stem Cell Technologies) or PI was used to exclude dead cells. For gating strategies, see Supplementary Fig. 1.
Confocal laser scanning microscopy. For imaging, T cells (5 × 10 4 ) were stimulated using CD3-, CD28-or TNFR2-activating monoclonal antibodies for 24 h, as indicated. Cells were stained on the plasma membrane using CD45-PE-CF594 (BD Biosciences). After cell fixation and permeabilization, GLUT1 was stained using rabbit anti-GLUT1 (Abcam) and goat anti-rabbit Alexa Fluor 488 (Invitrogen) as a secondary antibody (see Reporting Summary for additional information). Counterstaining of the nucleus was performed using DAPI (Sigma). Cells were suspended in PBS and added to 35/10-mm glass-bottom dishes (Greiner Bio-One). Images were acquired on the Andor Dragonfly 505 spinning disk confocal microscope adapted with a Leica DMi8 microscope (Oxford Instruments) at ×63 magnification. Image processing was performed using Fiji/ImageJ software (version 1.52j), and colocalization, reported as Pearson's r, was calculated on a single-cell basis using the JACoP plug-in (see Reporting Summary).
Seahorse metabolic assays. Real-time metabolic analyses were performed using a Seahorse XFe24 analyser (Agilent Technologies) 65 . For this purpose, 5 × 10 5 T cells were plated on poly-d-lysine-coated (Sigma) plates. After baseline measurement according to the manufacturer's protocol, cells were stimulated via injection port A containing CD3-, CD28-and TNFR2-activating monoclonal antibodies, and ECAR measurements were done for 15 cycles of 3 min mixing, 2 min waiting and 3 min measuring. Data were analysed using Wave software version 2.6.0.31 (Agilent Technologies). The area under the curve (AUC) of the ECAR in time following stimulation was determined by normalizing to the baseline ECAR measurement.
Statistical analysis. Data, excluding those describing transcriptomics or metabolomics data, were analysed using GraphPad Prism, version 8.1.1. Statistical analyses were performed as indicated in the figure legends. Data were log-transformed in case data were not normally distributed. Data are represented as mean ± s.e.m. A two-sided P < 0.05 was considered statistically significant.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All RNA-seq data are available in the GEO database under accession code GSE138603 and GSE138604. GSEAs were performed with GSEA software using the hallmark gene sets listed in the Molecular Signatures Database (MSigDB) available through https://www.gsea-msigdb.org/gsea/index.jsp. The proteomics data set used in this study is published by Cuadrado et al. 14 . All other data that support the findings of this study are available from the corresponding author on reasonable request.

Code availability
For RNA-sequencing, the script Itreecount was used to count the number of reads per gene. Itreecount is publicly available through https://github.com/NKI-GCF/ itreecount.