Leukoreduction system chambers (34) from apheresis platelet donors were used as source of peripheral blood mononuclear cells (PBMCs) and were obtained from the Central Institute for Blood Transfusion and Immunology situated at the Tirol Kliniken GmbH in Innsbruck, Austria. At the time of the donation, platelet donors were in a healthy state and fulfilled the general requirements for donating blood in Austria. Leukoreduction system chambers from platelet donors were provided on informed consent after finishing the apheresis procedure. PBMCs were isolated by density gradient centrifugation on a Ficoll-plaque by using Lymphocyte Separation Medium (LSM-A) (Capricorn Scientific, Ebsdorfergrund, Germany). After Ficoll gradient separation, the PBMC fraction was collected and washed twice with ice-cold DPBS (Gibco/Thermo Fisher Scientific). PBMCs were counted under a hemocytometer and the viability was assessed via Trypan blue (Sigma Aldrich, St.Louis, USA). The cells were cultured in RPMI 1640 medium containing 10% Fetal Bovine Serum (FBS), 2 mM L-glutamine and 1% penicillin-streptomycin (all from Sigma Aldrich, St.Louis, USA) and incubated overnight at 37°C and 5% CO₂ before starting the CD8⁺ T cell differentiation.

Naïve CD8⁺ T cells were sorted using a FACSAria I Sorter (BD Biosciences) by staining the isolated PBMCs with anti-CD8 FITC (BD Biosciences), anti-CD197 (CCR7) PE, anti-CD95 APC (both from BioLegend), anti-CD45RO PE-Cy7, anti-CD45RA BV421 and 7-Amino-Actinomycin D (7-AAD) (all from BD Biosciences). PBMCs were sorted into naïve T cells (T_N, CD8⁺ CD197⁺ CD95^- CD45RO^- CD45RA⁺), collected into RPMI 1640 medium containing 1% FBS and counted manually before starting the differentiation experiments.

For in vitro differentiation, naïve CD8⁺ T cells (1.5 x 10⁶ cells/mL and 1 x 10⁶ cells/mL) were distributed into four wells of a 24-well plate and activated for 8 days by adding prewashed anti-CD3/CD28 Dynabeads at a 1:1 bead-to-cell ratio (Invitrogen/Thermo Fisher Scientific, Massachusetts, USA) and human rIL-2 (30 U/mL) (Roche/Merck, Darmstadt, Germany) to the culture. To generate effector memory T cells, 1 x 10⁶ cells/mL were restimulated on day 7 by adding prewashed anti-CD3/CD28 Dynabeads at a 1:1 bead-to-cell ratio and human rIL-2 (30 U/mL) a second time to the culture. The cells were cultured in RPMI 1640 medium without glucose (Gibco/Life Technologies, Carlsbad, USA) where 11 mM D-glucose was added to 10% FBS, 2 mM L-glutamine and 1% penicillin-streptomycin (all from Sigma Aldrich, St.Louis, USA). 24 h before collection, culture medium was replaced with RPMI 1640 medium containing 11 mM of ¹³C labeled-glucose (Cambridge Isotope Laboratories, Massachusetts, USA). At indicated time points, naïve T cells (T_N, day 1), stem cell memory T cells (T_SCM, day 2), central memory T cells (T_CM, day 5) and effector memory T cells (T_EM, day 8) were collected and anti-CD3/CD28 beads were removed by using a DynaMag-2 magnet (Thermo Fisher Scientific, Massachusetts, USA) before further processing for flow cytometry, RNA sequencing and metabolomics analyses.

To analyze cell surface marker during differentiation, 1 x 10⁵ PBMCs were labeled at every time point (day 1, 2, 5 and 8) with the same panel of human antibodies as used for cell sorting and analyzed by flow cytometry on a FACS Fortessa (BD Biosciences). UltraComp eBeads (Invitrogen/Thermo Fisher Scientific, Massachusetts, USA) were used for the compensation setup. The cell type fractions were determined as the following: Naïve T cells (T_N, CD8⁺ CD197⁺ CD95^- CD45RO^- CD45RA⁺), stem cell memory T cells (T_SCM, CD8⁺ CD197⁺ CD95⁺ CD45RO^- CD45RA⁺), central memory T cells (T_CM, CD8⁺ CD197⁺ CD95⁺ CD45RO⁺ CD45RA^-) and effector memory T cells (T_EM, CD8⁺ CD197^- CD95⁺ CD45RO⁺ CD45RA^-). Analysis was performed using FlowJo software.

In vitro exhausted CD8⁺ T cells were generated as previously described (35). Briefly, CD8⁺ T cells isolated from healthy donor PBMCs were stimulated by a 1:1 ratio of anti-CD3/28 activation beads (Miltenyi) in complete human T cell medium (RPMI1640 from Sigma containing 2 mM glutamine with the addition of 1 mM pyruvate, 1% penicillin-streptomycin, 10% heat-inactivated AB+ male human serum, 50 nM beta-mercaptoethanol in presence of 150 U/mL rhIL2 (Proleukin). The day after stimulation, cells were transduced with VSVg-pseudotyped Lentivirus encoding a human T cell receptor specific for NY-ESO-1 (gift from Natalie Rufer and Michael Hebeisen) (36). To generate exhausted T cells (T_EX), TCR⁺ cells were stimulated every 3 days for four cycles with the HLA-A2⁺ T2 tumor cell line loaded with 1000 nM NY-ESO-1 SLLMWIQV peptide at a 1:3 effector to target ratio in T cell medium with 50 U/mL IL-2. For functional effector cells (T_EFF), the T cells were expanded in the same media for 9 days before a single stimulation with the peptide loaded T2 cells as described above. 3 days post last stimulation, the T cells were sorted for live TCR⁺ CD8⁺ CD19^- CD14^- CD56^- cells and resuspended in RPMI 1640 medium containing 5 mM of ¹³C labeled-glucose at 1x10⁶ cells/mL with the addition of 50 U/mL IL-2. Cells were cultured in a 24-well plate for 24 h at 37°C with 5% CO₂ before further processing for RNA sequencing and metabolomics analyses.

Naïve T cells (T_N), stem cell memory T cells (T_SCM), central memory T cells (T_CM) and effector memory T cells (T_EM) were collected, counted, and centrifuged at 1500 rpm for 10 min at room temperature. Cells were resuspended at a concentration of 8x10⁶ cells/mL in Seahorse XF Base Medium (Agilent Technologies, Santa Clara, CA) supplemented with 1 mM pyruvate, 2 mM glutamine, 10 mM glucose (all from Sigma Aldrich, St.Louis, USA), and pH was adjusted to 7.4. Seahorse XFp Cell Culture Miniplates (Agilent Technologies, Santa Clara, CA) were precoated with poly-L-lysine (Sigma) and 1x10⁵ to 4x10⁵ cells were plated into each well. Miniplates were centrifuged at 300 g for 1 min with low-brake deceleration. Seahorse XFp Cell Energy Phenotype Tests (Agilent Technologies, Santa Clara, CA) were performed according to the manufacturer’s protocol on an XFp instrument and data analysis was performed using the Seahorse XFe Wave software.

After 24 h culturing in ¹³C-containing medium, 1x10⁶ cells were collected at indicated time points and centrifuged at 2000 rpm for 1 to 3 min at 4°C. For medium extraction, 10 µL of the supernatant was added to 990 µL 80% methanol containing 2 µM of d27 myristic acid (pre-chilled to -80°C) (extraction buffer provided by the VIB Metabolomics Expertise Center). For cellular extraction, the supernatant was removed, the pellet was carefully washed with 1 mL of ice-cold 0.9% NaCl (Sigma Aldrich, St. Louis, USA) and this solution was then aspirated. Pellet was resuspended in 250 µL (T_EFF and T_EX) or 300 µL (T_N, T_SCM, T_CM and T_EM) of 80% methanol containing 2 µM of d27 myristic acid (pre-chilled to -80°C) and pulse vortexed three times for five seconds. Samples were stored at -80°C for 24 h, centrifuged at 1500 rpm for 15 min at 4°C and 250 µL of metabolite-containing supernatants of each sample were sent for analysis.

10 µL of each sample was loaded into a Dionex UltiMate 3000 LC System (Thermo Scientific Bremen, Germany) equipped with a C-18 column (Acquity UPLC -HSS T3 1. 8 µm; 2.1 x 150 mm, Waters) coupled to a Q Exactive Orbitrap mass spectrometer (Thermo Scientific) operating in negative ion mode. A step gradient was carried out using solvent A (10 mM TBA and 15 mM acetic acid) and solvent B (100% methanol). The gradient started with 5% of solvent B and 95% solvent A and remained at 5% B until 2 min post injection. A linear gradient to 37% B was carried out until 7 min and increased to 41% until 14 min. Between 14 and 26 min the gradient increased to 95% of B and remained at 95% B for 4 min. At 30 min the gradient returned to 5% B. The chromatography was stopped at 40 min. The flow was kept constant at 0.25 mL/min and the column was placed at 40°C throughout the analysis. The MS operated in full scan mode (m/z range: [70.0000-1050.0000]) using a spray voltage of 4.80 kV, capillary temperature of 300°C, sheath gas at 40.0, auxiliary gas at 10.0. The AGC target was set at 3.0E+006 using a resolution of 140000, with a maximum IT fill time of 512 ms. Data collection was performed using the Xcalibur software (Thermo Scientific). The data analyses were performed by integrating the peak areas (El-Maven - Polly - Elucidata) (37).

Quantification thresholds were defined for each batch from three blank measurements as two standard deviations above blank means. Isotopologue intensities below blank means were set to zero and isotopologue intensities above blank means but below quantification thresholds were corrected by a linear mapping between zero and the quantification threshold. Only isotopologues with at least two biological replicates above quantification thresholds were used for statistical testing. Correction for natural isotope abundance was applied with the IsoCorrectoR R package 1.16.0 (38) in high-resolution mode. Corrected data were normalized between isotopologues by mean metabolite abundance and between samples by size factors consisting of total isotopologue sums weighted by their inverse relative variances. Isotopologue abundances were further cleaned from sample groups with more than 1/3 of the respective values below quantification threshold. Metabolite abundances were calculated from the sums of isotopologue abundances and fractions of isotopologues relative to isotopologue sums were used as mass isotopomer distributions. For differential abundance analysis, metabolite abundances were assessed in a linear mixed model using the nlme R package 3.1-162 and isotopologue fractions were modeled with beta regression using the glmmTMB R package 1.1.7. Celltype was treated as fixed and donor as random effect. The multcomp R package 1.4-23 was used for statistical hypothesis testing of particular contrasts and the resulting p-values were FDR-adjusted. Statistical analyses of isotopologue abundances using a linear mixed model and of fractional metabolite labeling using beta regression yielded similar conclusions and were thus not further discussed.

Naïve T cells (T_N), stem cell memory T cells (T_SCM), central memory T cells (T_CM), effector memory T cells (T_EM), full effector function T cells (T_EFF) and exhausted T cells (T_EX) were collected, snap frozen and total RNA was isolated using the RNeasy Plus Mini Kit (Qiagen) or RNeasy Plus Micro Kit (Qiagen GmbH-Austria, Hilden, Germany), including DNAse treatment and following the manufacturer’s instructions. All Isolated total RNA samples were quality validated and submitted to library preparation following the Lexogen QuantSeq 3’mRNA protocol (Lexogen GmbH, Vienna, Austria). The resulting libraries were multiplexed and sequenced with Ion Proton technology and Ion Hi-Q chemistry (Ion Torrent, Thermo Fisher Scientific, Vienna, Austria). CUTADAPT 4.0 was used to trim sequencing reads of low-quality bases and poly-A tails with the options -q 20, -O 10, -e 0.15, -m 10 and the adapter sequence -a A100 (39). We used the nf-core rnaseq pipeline 3.6 (10.5281/zenodo.6327553) with the trimming step disabled for the alignment of reads with STAR 2.7.9a to the human genome (GRCh38) and for the quantification of reads with SALMON 1.5.2 (40). The correct pairing of samples was confirmed by grouping samples to donors of origin with NGSCheckMate 1.0.0 (41). Raw counts were imported into R 4.1.3 and DESeq2 1.38.3 was used with IHW 1.26.0 to test for differentially expressed genes (42). For analyses of biological functions, we compiled a collection of gene sets from gene ontology (GO), HALLMARK, KEGG and the metabolic databases MitoCarta 3.0 (43) and MetabolicAtlas 3.3 (44). For the differentiation samples (T_N, T_SCM, T_CM and T_EM), significantly different genes with padj < 0.05 were partitioned into groups of up- (U) and downregulation (D) according to their z-scaled expression values. The resulting gene clusters were further used for gene set overrepresentation analysis (ORA) with the ClusterProfiler R package 4.6.2 (45). For samples from the exhaustion experiments, we performed gene set enrichment analysis (GSEA) with ClusterProfiler on genes ranked by the Wald test statistic from DESeq2. Raw bulk RNA sequencing data were retrieved from the GEO database for accession numbers GSE121226 (46), GSE140430 (47), GSE147398 (48) and GSE179609 (49) and uniformly processed. The single-cell CD8⁺ T cell atlas was downloaded from https://singlecell.mdanderson.org/TCM (50). We then identified the top expressed genes in each of our subsets by one-vs-all comparisons and calculated the enrichment scores of these signatures in the normalized bulk and pseudo-bulk reference samples using gene set variation analysis (51). The regression model for mitochondrial abundance prediction was based on public data from (52) and (53).

To study the differentiation dynamics of human T cells, we employed an in vitro model of naïve T cell (T_N) activation ( Figure 1A ). Sorted CD8⁺ T_N cells from three healthy donors were stimulated with anti-CD3/CD28 beads and IL-2 to drive activation and differentiation towards stem-like (T_SCM) and central memory (T_CM) cells, followed by restimulation to promote full differentiation of effector (T_EM) cells. Based on the expression of a panel of markers, we confirmed the presence of these distinct phenotypes at different time points by flow cytometry ( Figures S1A-C ), showing that our data are consistent with a differentiation model where the majority of activated cells acquire an effector phenotype after passing through a stem-like progenitor phase, followed by a contraction phase at the end.

Based on our initial observation of surface marker dynamics during differentiation, we characterized gene expression changes on a global scale by RNA sequencing, followed by differential gene expression analysis. We compiled a collection of publicly available RNA sequencing datasets that include T_SCM or T_CM samples to demonstrate the similarity of our samples with these T cell subtypes ( Figure S2A ). In addition, we compared our subsets to a recently published single-cell atlas of tumor-infiltrating CD8⁺ T cells to confirm their relevance in vivo ( Figure S2B ). Figure S2C shows the expression levels of selected marker genes. Quiescence- (FOXO1, BTG1/2) and stemness-associated genes (TCF7) were highly expressed in naïve cells. Activation markers (ILRA, CD27, CD28) were upregulated following activation and effector proteins (IFNG, GZMB) were most strongly upregulated in T_EM cells. Among inhibitory receptors, VSIR (VISTA) was highly expressed in T_N, PDCD1 was transiently upregulated following the initial stimulation, and both PDCD1 and LAG3 were upregulated following re-stimulation in T_EM cells.

Differential gene expression analysis revealed global transcriptomic changes between these CD8⁺ T cell subsets ( Table S1 ), with the strongest effect following the initial stimulation of T_N cells. At a significance level of 0.05, 7076 genes were differentially expressed between T_N and T_SCM cells ( Figure 1B ), 2164 genes changed between the T_SCM and the T_CM subset ( Figure 1C ), and 1591 differed significantly upon restimulation and further development of T_CM into T_EM cells ( Figure 1D ). KLF2, which has been previously associated with quiescence (54), was highly expressed in T_N and downregulated upon activation in T_SCM cells. Besides classical activation markers, genes involved in cell cycle regulation (CDK1, AURKB) and nucleotide biosynthesis (TYMS, RRM2) were among the highest upregulated genes in T_SCM, along with other key differentially expressed metabolic genes like GAPDH, ENO1, FABP5, LDHA or TXNIP. Compared to T_SCM cells, T_CM cells upregulated several genes also found in T_N cells (KLF2, SLFN5) and downregulated several metabolic genes (ODC1, NAMPT). T_EM cells had high expression of genes associated with effector cells, including PDCD1, IL2, FASLG and IFNG.

As many differentially expressed genes were shared between subsets, we partitioned these genes into clusters of up (U)- and downregulation (D) and performed gene set overrepresentation analysis on the individual clusters ( Figure 1E , Table S2 ). Gene sets overrepresented only in T_N cells (cluster 1, 1-UDDD) included pathways involved in histone binding and modification, transcriptional repression and in the generic activation of immune cells. Various processes related to cell cycle control, DNA replication and cell proliferation, as well as MYC, mTORC1 and E2F signaling were highly enriched in all activated subsets (2-DUUU). This concurred with an upregulation of metabolic processes including OXPHOS, TCA cycle, mitochondria, glycolysis, fatty acid metabolism, nucleotide biosynthesis, NADH metabolism and succinate dehydrogenase activity, indicating major metabolic reprogramming due to increased enzyme expression. Two clusters were mainly comprised of transiently activated genes that depended on acute (re-) stimulation (i.e., genes highly expressed in T_SCM and T_EM cells). Therein, MYC signaling, ribosome biogenesis and metabolic pathways were even further upregulated (3-DUDU), whereas lysosomal pathways and antigen processing were downregulated (4-UDUD). Wnt signaling was a key feature of T_N and T_SCM cells (5-UUDD). Specifically enriched in T_SCM (6-DUDD) was “KEGG glyoxylate and dicarboxylate metabolism”, and downregulated genes (7-UDUU) were involved in leukocyte cell-cell adhesion, apoptosis and immune activation and differentiation. Genes shared by T_SCM and T_CM were involved in PI3K signaling and amino acid transport (8-DUUD). Actin cytoskeleton remodeling and leukocyte migration were characteristically activated in T_CM cells (10-DDUD), whereas cell cycle, Wnt signaling and histone acetylation were repressed (11-UUDU). Genes up in the T_CM and T_EM subsets (12-DDUU) included DNA deamination (nucleotide salvage), IFN-γ response, secretion and leukocyte degranulation. Apoptosis was most strongly upregulated in T_EM cells (13-DDDU), together with hypoxia and lymphocyte migration, while they were devoid of Ca²⁺ signaling (14-UUUD). The top pathways specifically upregulated in each subset are summarized in Figure S2D .

Thus, these results show that in vitro differentiation reflects the main processes of in vivo immune responses (activation, proliferation, migration, contraction) and recovered their main regulatory cascades, including Wnt and Ca²⁺ signaling. Several pathways that orchestrate metabolic reprogramming during T cell activation and differentiation, such as mTOR and MYC, and key metabolic genes (ENO1, TYMS, SDHA, SDHB, SDHC) were upregulated following T cell activation, whereas the levels of the metabolic regulator TXNIP decreased.

As the transcriptomic analyses indicated prominent changes in cellular metabolism, we went on to directly investigate metabolic pathway activities by performing ¹³C tracing experiments followed by LC-MS measurements. First, we generated a global qualitative map of tracer fluxes in central carbon metabolism ( Figure 2 ). U-¹³C-glucose was readily imported and shunted into glycolysis within 24 h of labeling in all subsets, including naïve T cells, resulting in full ¹³C labeling of metabolites in upper glycolysis. In T_N cells, however, no labeling was detected downstream of phosphoglycerate (3pg), indicating that no glycolytic flux went into the production of phosphoenolpyruvate (pep). This was in line with a pronounced transcriptomic downregulation of ENO1 mRNA in T_N cells, which we assumed to be the major driver of the enolase reaction, as ENO2 was only very lowly expressed ( Figure S3 ). Enzymes of the rate-limiting steps of glycolysis were also differentially expressed (HK1, PFK, PK). Whereas PKLR was lowly expressed in T_N cells and even further downregulated upon activation, PKM was abundant and increased further after activation, in line with its known role to promote anabolism by increasing the availability of glycolytic intermediates (55). Thus, T_N cells had a pronounced suppression of glycolytic flux and did not oxidize glucose in the TCA cycle, indicating that these cells used carbon sources other than glucose for OXPHOS. In contrast, activated cells showed a consistent pattern of m+2 incorporation into the TCA cycle, indicating that glycolytic flux was directed there primarily via acetyl-CoA. To a lesser extent, m+3 labeling was observable at a mean m+3/m+2 ratio of 0.15, which likely originates from anaplerotic flux via pyruvate carboxylase (PC), and m+4 isotopomers originating from the condensation of m+2-labeled oxaloacetate with m+2-labeled acetyl-CoA in higher-order turns of the cycle. High flux through the PC reaction has previously been described as a characteristic of in vitro activated T cells with an m+3/m+2 ratio well above 0.5, whereas in vivo ¹³C tracing showed a predominance of m+2 labeling in the TCA cycle (31). Thus, our subsets exhibit an m+3/m+2 ratio similar to that observed in vivo, in line with low PC expression.

To study the metabolic dynamics of CD8⁺ T cells during activation in more detail, we analyzed the LC-MS data to detect significant differences in both metabolite abundances ( Figures 3A–C ; Table S3 ) and labeling states ( Figures 3D–F ; Table S3 ). Compared to T_N, T_SCM cells had decreased intracellular levels of hexoses (hex), but increased levels of hexose-6-phosphates (h6p), glycolytic intermediates (dhap, g3p) and lactate (lac) ( Figure 3A ). Analysis of mass isotopomer distributions further revealed significantly increased labeling of glycolytic products, in particular, a shift from unlabelled to fully labeled pyruvate (pyr_m3) and lactate (lac_m3), and to 50% labeled alanine (ala_m3) ( Figure 3D ). As described above, the expression of glycolytic genes was increased. Thus, the suppression of glycolytic flux in naïve T cells was released upon activation. Metabolites in the beginning of the TCA cycle had 30% increased labeling derived from acetyl-CoA (citrate-m+2, aconitate-m+2), which declined to 10-20% further downstream. Surprisingly among the TCA cycle metabolites, succinate levels were increased and the fumarate/succinate ratio was decreased in T_N cells ( Figure S4A ). Expression of succinate, fumarate and malate dehydrogenases was also decreased ( Figure S5 ), implying substrate accumulation due to downregulation of succinate dehydrogenase (SDH) activity in T_N cells. As expected, T_SCM cells also upregulated several biosynthetic pathways. Aspartate, glutamate and proline exhibited the same labeling patterns as their respective precursors in the TCA cycle due to ongoing biosynthetic flux. Nucleotide, glycosylation (UDP-GlcNAc) and glucuronidation (UDP-glucuronate) substrate levels and labeling were increased and the glycine/serine ratio was higher ( Figures S4B, C ), indicating increased serine biosynthesis and demand for 1C units.

The GSSG/GSH (oxidized and reduced glutathione, respectively) ratio was higher in T_N cells compared to all other subsets ( Figure S4C ). Moreover, in T_SCM cells, most amino acid levels were decreased, whereas biosynthesis and the expression of amino acid transporters (SLC7A5, SLC3A2, SLC1A5, SLC38A5) were simultaneously upregulated. Notable exceptions among amino acids were significantly higher concentrations of proline and alanine in T_SCM compared to T_N cells. T_SCM cells also had the highest NADH/NAD⁺ ratio and increased labeling of NADH. Continuous regeneration of NAD⁺ from NADH is required to support high rates of glycolytic flux, which can be achieved in the LDH reaction together with conversion of pyruvate to lactate. As expected, we found increased LDHA expression (log2FC=2.7). However, we also found production of alanine from pyruvate, as well as increased flux into the TCA cycle, and both of these pathways can not be directly used to regain NAD⁺. Rather, catabolisation of pyruvate and subsequent oxidation of TCA cycle intermediates in mitochondria produces additional NADH from NAD⁺. NADH is eventually oxidized to NAD⁺ in OXPHOS, but then needs to be transferred across the mitochondrial membranes to be used in glycolysis. Several shuttle systems exist for this purpose. We found increased expression of genes involved in the malate-aspartate shuttle and partial upregulations of the glycerol-3-phosphate shuttle and the citrate-malate shuttle ( Figure S6 ). In addition, genes involved in the proline cycle were upregulated (PYCR1, PYCR2 and ALDH18A1). As we also found an increase in proline levels and labeling, we therefore suggest that T_SCM cells utilize the proline cycle to maintain NADH balance. In T_CM cells, proline levels were moderately reduced compared to T_SCM cells ( Figure 3B ), PYCR1 was downregulated and the glycolytic intermediate dihydroxyacetone phosphate (dhap) was decreased. Labeling of TCA cycle metabolites was higher in T_CM cells, most likely due to a decrease in both, draining cataplerotic reactions and replenishing anaplerosis ( Figure 3E ). Decreased glutamine catabolization is in line with a lower activity of MYC, a known driver of glutaminolysis. Compared to T_CM cells, the intracellular levels of most amino acids were increased in the T_EM subset ( Figure 3C ), reflecting decreased consumption, as the expression of importers was only marginally increased or even decreased ( Figure S7 ). Whereas glycolytic intermediates were increased and the labeling remained high in glycolysis, the labeled fractions of TCA cycle intermediates were decreased ( Figure 3F ). Thus, high glycolytic flux, decreased biosyntheses and decreased entry of glucose carbon into the TCA cycle were characteristic of T_EM metabolism.

In summary, the ¹³C tracing data revealed a quiescence-associated metabolism in T_N cells. Reprogramming to an anabolism-oriented phenotype was indicated by high rates of glycolysis and nucleotide biosynthesis in T_SCM cells, which also coincided with an upregulation of the proline cycle. T_CM metabolism was qualitatively similar to the T_SCM phenotype, but compatible with lower rates in biosynthetic pathways. A further increase in the reliance on glycolysis with a concomitant decrease in other pathways was the main metabolic feature of the T_EM subset.

We then performed extracellular flux analysis on a Seahorse Analyzer to relate the observed intracellular metabolic fluxes to extracellular acidification rates (ECAR) and oxygen consumption rates (OCR) ( Figures 3G, H ). Unstimulated T_N cells had only basal ECAR and OCR. T_SCM cells were the most active subset and operated at maximum capacities of both ECAR and OCR. Both T_CM and T_EM cells had low OCRs that were only slightly higher than that of T_N cells, but had higher spare respiratory capacity. ECAR was similarly decreased in T_CM cells, but upon restimulation and effector differentiation, ECAR significantly increased to rates approximately half of those of T_SCM cells in the T_EM cells. Despite these pronounced differences in metabolic rates between subsets, we found a progressive and gradual increase in the ECAR/OCR ratios in consecutive subsets ( Figure 3I ), suggesting that CD8⁺ T cells increasingly rely on glycolysis during differentiation. To further investigate these findings on a transcriptional level, we checked the expression of OXPHOS subunits ( Figure S8 ). Unexpectedly, expression of most of the respiratory complexes was not decreased or even increased in T_EM cells. Additionally, a regression model trained on public datasets with mitochondrial copy numbers available predicted a high abundance of mitochondria in T_EM cells ( Figure S9A ), indicating that the observed shift towards glycolysis was not due to decreased mitochondrial abundance or changes in mitochondrial gene expression.

Finally, we investigated whether our observations were related not only to the differentiation but also to the proper function and dysfunction of T cells. We adopted an in vitro exhaustion model (35) to generate control effector and hypofunctional exhausted T cells ( Figure 4A ). Briefly, CD8⁺ T cells from healthy human donors were transduced with a T cell receptor specific for the NY-ESO-1 cancer antigen, and the transgenic T cells were stimulated with NY-ESO-1 either acutely or repeatedly to induce full effector function (T_EFF) or a hypofunctional state of exhaustion (T_EX), respectively. Analysis of the RNA sequencing data showed 3625 differentially expressed genes between T_EX and T_EFF cells ( Figure 4B ; Table S1 ). Whereas most effector molecules (GZMB, LTA) were decreased, expression of GNLY was increased in exhausted cells. As in quiescent naïve T cells, the transcription factors KLF2, TCF7 and BTG1 were upregulated. Notably, among the genes with metabolic functions, TXNIP had the most differentially upregulated transcript in T_EX cells. We further confirmed differential regulation of these genes ( Table S4 ) in a previously published dataset (35).

Gene set enrichment analysis ( Figure 4C ; Table S5 ) found an upregulation of DNA-binding transcription factor activity, with KLF2 and TCF7 as the top contributing genes. Also upregulated were cell adhesion terms and “GO negative regulation of leukocyte mediated cytotoxicity”. A downregulation was observed for gene sets involved in the main signaling pathways involved in T cell activation, including MYC, E2F, mTORC1 signatures. Moreover, terms related to cell cycle, ribosomes, OXPHOS, glycolysis and amino acid metabolism were downregulated. Most striking, however, was the depletion of mitochondrial components. Consistently, the predicted mitochondrial abundance was significantly decreased in T_EX cells ( Figure S9B ).

We further performed ¹³C tracer studies with exhausted and control effector cells ( Table S6 ). Whereas pyruvate levels were slightly higher in T_EX compared to T_EFF cells, lactate levels were decreased and less labeled ( Figures 4D, E ). Consistent with lower glycolytic flux, LDHA and key glycolytic enzymes were expressed at lower levels in T_EX cells. Most amino acids had decreased intracellular concentrations, particularly alanine and proline. Genes of the proline cycle were strongly downregulated (PYCR1, PYCR2, PYCR3 and ALDH18A1) while the NAD⁺/NADH ratio remained high, suggesting that T_EX cells could easily maintain redox balance even under low lactate production. Alanine production is usually coupled to glycolytic flux and its labeling was likewise decreased. Presumably, T_EX cells need less nutrients for protein synthesis and growth, so lower amino acid levels would hint towards decreased uptake. Indeed, we found that several of the main amino acid transporters were downregulated in T_EX cells, including SLC7A5, SLC1A5 and SLC38A5 ( Figure S7 ). In the TCA cycle of T_EX cells, succinate was slightly but significantly increased and the fumarate/succinate ratio was decreased ( Figure S4F ), indicating lower SDH activity. Expression of SDH was moderately decreased, as well as MDH2 and, most pronouncedly, FH. In addition, the TCA-cycle-derived itaconate showed a small but significant increase in T_EX cells.

Together, these data indicate that exhausted cells had reduced effector functions and partially activated signaling pathways associated with quiescent cells. T_EX cells were less metabolically active than T_EFF cells, had lower glycolytic flux and were depleted in intracellular amino acids due to restricted uptake. In addition, changes in the TCA cycle were associated with lower SDH activity, succinate and itaconate accumulation and mitochondrial abundance was decreased.