FAS-controlled T cells drive lymphoproliferation through glycolysis without effector differentiation

FAS-controlled T cells (FCT) represent a recently defined, physiological T cell subset stringently controlled by FAS-mediated apoptosis (1). FCT are highly proliferative and characterized by a distinct phenotype and transcriptome differing from other known T cell differentiation subsets (1). They can be found among mature CD4, CD8, and T cell receptor (TCR)αβ⁺CD4⁻CD8⁻ double-negative T cells (DNT) and are defined via flow cytometry by a CD38^highCD45RA^highCD45R B220⁺ phenotype (1). In healthy individuals, FCT make up <1% of TCRαβ⁺ T cells, whereas they massively accumulate, mostly as DNT, in patients with autoimmune lymphoproliferative syndrome (ALPS), caused by genetic defects in the FAS signaling pathway (2, 3, 4, 5). Thus, the majority of ALPS-DNT are FC-DNT (1). It was previously shown that human FC-DNT share TCR CDR3β sequences with phenotypically identical FC CD4 and FC CD8, suggesting that these FC CD4/8 can transition to FC-DNT by loss of co-receptor expression (1, 6). TCR CDR3β sequence comparisons suggested a predominant CD8 origin of DNT in ALPS (7).

Lymphoproliferative manifestations in ALPS patients are caused by uncontrolled proliferation of activated FCT cells, which cannot be eliminated via FAS-induced apoptosis. The highly activated state is reflected by their high CD38, HLA-DR, and Ki-67 expression (1). Strong in vivo proliferation leading to lymphadenopathy and splenomegaly is also a feature of CD8 T cells responding to acute Epstein–Barr virus (EBV) infection. In ALPS patients, the driving signals of FCT proliferation are currently poorly understood. Specific infectious triggers of lymphoproliferative manifestations in ALPS patients have not been reported (8, 9, 10). Notably, Fas-deficient mice develop lymphadenopathy with accumulation of B220⁺ DNT under germ-free conditions (11). Increased CD3ζ phosphorylation was demonstrated in B220⁺ DNT of Fas-deficient mice, indicating active TCR signaling (12, 13). In addition, elevated FCT frequencies in patients with CTLA-4 deficiency (1) or STAT3-activating mutations support a role for CD28 and STAT3-activating cytokines in FCT activation/expansion. Finally, FCT show active mTOR signaling (1, 14), which can be triggered by TCR, co-stimulatory molecules, and/or cytokines. Consistently, FCT are highly susceptible to mTOR inhibition and rapidly decline under rapamycin treatment, which represents a targeted therapy for ALPS patients (15). In summary, FCT are highly active, proliferating cells—both in healthy individuals and in FAS-deficient patients—and exhibit signs of TCR activation. These characteristics closely resemble those of virus-activated CD8 T cells.

TCR activation and co-stimulation initiate sets of signaling cascades that promote T cell exit from quiescence (16). The mTOR pathway is crucial for early metabolic changes associated with T cell activation and differentiation (17, 18, 19). A crucial metabolic change is an increase in glucose metabolism. CD28-mediated signaling controls the expression of the plasma membrane glucose transporter, Glut1 (20). In addition to increasing glucose import, activated T cells engage aerobic glycolysis to generate ATP and metabolic intermediates required for cell growth. This glycolytic switch has been shown mainly in murine effector T cells under controlled infectious conditions. Enhanced glycolysis not only fuels proliferation but directly promotes IFNγ translation and thus effector function (21). Indeed, T cells activated without glucose have defective effector function (22). If T cells are continuously stimulated, as in chronic viral infection, they can become exhausted, characterized by attenuated glycolytic metabolism, mitochondrial dysfunction, and impaired functionality (23, 24). Effector T cells also increase their mitochondrial mass but show fissed rather than the fused mitochondria characteristic for memory T cells, which rely on oxidative phosphorylation (OXPHOS) as their major source of energy (25). Effector T cells generate reactive oxygen species (ROS) via the mitochondrial respiratory chain or dedicated ROS-producing enzymes such as NADPH oxidases (26). These ROS have important signaling function but in excess can cause oxidative stress and cellular damage (25, 27, 28).

To understand the metabolic processes underlying abnormal lymphoproliferation in ALPS, we studied the metabolic profile of the highly active FCT cells using metabolic assays, RNA-sequencing analysis, as well as theoretical modeling via genome-scale metabolic reconstruction and constraint-based modeling (29, 30). As a benchmark for strongly in vivo–activated effector T cells, we compared the FCT profile to that of T cells activated during primary EBV infection. Resting T cells from healthy individuals served as controls. The identified similarities and differences provide insights into the metabolic processes fueling lymphoproliferation and may in the future open novel treatment options targeting the metabolic basis of lymphoproliferative diseases.

To study the metabolic basis of T cell–driven lymphoproliferation in ALPS, we focused on TCRαβ⁺CD4⁻CD8⁻ DNT (ALPS-DNT), which in many untreated ALPS patients consist of >90% FC-DNT. As a benchmark for highly in vivo–activated T cells, we compared them to CD8 T cells of individuals with lymphoproliferation driven by acute primary EBV infection (EBV-CD8) in the absence of a known genetic predisposition. As expected for bona fide CD8 effector cells, human CD8 responding to EBV were highly activated, proliferative, and expressed CD38, HLA-DR, EOMES, and Ki67 (Fig. 1 A and Table S1) (31). Bulk RNA-sequencing analysis of sorted T cell populations revealed highly expressed cell cycle–related genes targeted by the transcription factor (TF) E2F in ALPS-DNTs compared to CD8 T cells of healthy individuals (healthy donor [HD]-CD8) (Fig. 1 B). Interestingly, expression levels of activation and proliferation markers as well as cell division–associated genes were similarly elevated albeit not identical in ALPS-DNT compared to EBV-CD8, (Fig. 1, A and B). Transcripts associated with mTOR signaling were also increased in both cell populations (Fig. 1 C). Consistently, flow cytometry revealed increased expression of the mTOR downstream targets HIF1α and GLUT1 in ALPS-DNT and less pronounced in EBV-CD8 (Fig. 1, D and E) when compared to HD-CD8. GLUT1 expression was not only increased intracellularly but also on the cell surface (Fig. S1 A). To evaluate the capacity of ALPS-DNT, EBV-CD8, and HD-CD8 for biomass production in silico, we performed flux balance analysis (FBA) with RNA-sequencing data as input using the generic human biomass reaction from Recon3D (29), which models the stoichiometric demands of major biomass precursors, including amino acids, nucleotides, lipids, and cofactors (Fig. S1 B and Materials and methods section). ALPS-DNT and EBV-CD8 models showed greatly increased biomass production capacity in silico compared to HD-CD8 (Fig. 1 F) after specifying metabolic constraints according to the cell-specific transcriptomic profiles (30). In conclusion, ALPS-DNT showed a similar degree of activation, mTOR signaling, biomass production capacity, and proliferation compared to acutely EBV-stimulated CD8 T cells.

Consistent with increased expression of the glucose transporter GLUT1, enhanced glucose uptake was documented by increased 2-[N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl) amino]-2-deoxy-D-glucose (2-NBDG) fluorescence of both ALPS-DNT and EBV-CD8 (Fig. 2 A). Functional Seahorse analysis of sorted cell populations showed increased extracellular acidification rate (ECAR) as a proxy for baseline glycolysis and increased glycolytic reserve in both ALPS-DNT and EBV-CD8 compared to HD-CD8 (Fig. 2 B and Fig. S2 A). Consistent with increased glucose metabolism, RNA-sequencing revealed comparably increased expression of glycolytic enzymes, including GAPDH, TPI1, α-enolase, and PKM (Fig. S2 B). In contrast, baseline oxygen consumption rate (OCR) was not significantly increased despite upregulation of OXPHOS-related genes (Fig. S2, C and D), leading to a relative shift to glycolytic metabolism as indicated by a lower OCR/ECAR ratio in ALPS-DNT and EBV-CD8 compared to resting HD-CD8 (Fig. 2 C). Notably, both cell populations showed reduced spare respiratory capacity (Fig. 2 D), as previously described for acute effector T cells (32, 33). Mirroring the seahorse analysis, the in silico investigations of the three T cell populations revealed stronger dependence of ATP production on glycolysis in EBV-CD8 and ALPS-DNT versus HD-CD8, with in silico ATP production only marginally affected under simulated anoxic conditions for EBV-CD8 and ALPS-DNT (Fig. S2 E). Furthermore, in silico analysis indicated markedly increased L-lactic acid production capacity and hexokinase fluxes in ALPS-DNT in comparison to EBV-CD8, which in return showed higher glycolytic capacity than HD-CD8 (Fig. S2 F). Using in silico blockade of the hexokinase reaction led to a decrease in in silico ATP production in ALPS-DNT models but not in EBV-CD8 and HD-CD8 models, indicating higher metabolic flexibility in EBV-CD8 cells than in ALPS-DNT cells (Fig. S2 E). Furthermore, we performed in silico inhibitor modeling for fatty acid oxidation (FAO) (blocking CPT1), glutaminolysis (blocking glutamine uptake and the mitochondrial glutaminase, respectively), and the pentose phosphate pathway (blocking G6PD). Intriguingly, ALPS-DNT models were no longer able to produce biomass upon blocking glutamine uptake, while also showing reduced biomass production with mitochondrial glutaminase blocked (Fig. S2 G), highlighting a possibly high glutamine dependency. In contrast, blocking FAO had only minor effects on the ALPS-DNT in silico biomass production, while the HD-CD8 models could no longer support biomass production with blocked CPT1 (Fig. S2 G). However, the EBV-CD8 models were unaffected by any in silico inhibitor modeling. Taken together, our in silico inhibitor modeling supports the hypothesis that FCT rely on distinct metabolic pathways, including glutamine-dependent metabolism, whereas EBV-CD8 cells appear metabolically more flexible and less sensitive to perturbation of these pathways. Consequently, we hypothesized that this strong dependence on glycolytic metabolism could be targeted to reduce FCT survival. Since ALPS-DNT rapidly undergo apoptosis ex vivo (34 and Fig. S3 A), we first established culture conditions that improve ALPS-DNT survival and enable metabolic intervention. Co-culturing peripheral blood mononuclear cells (PBMC) with allogenic monocyte-derived dendritic cells (DC) in the presence of IL-10 and IL-21 (“Stim” in Fig. 2, E and F) improved 24-h survival of CD38^highCD45RA^high ALPS-DNT and CD38⁺ EBV-CD8, as determined by flow cytometric quantification of viable cells. We then performed overnight survival assays in the presence of the glycolysis-blocking agent 2-deoxy-D-glucose (2DG) or the mTORC1 inhibitor rapamycin. While both subsets showed reduced survival upon rapamycin addition, cytokine/monocyte-derived DC-sustained ALPS-DNT exhibited lower survival than EBV-CD8 cells upon inhibition of glucose metabolism (Fig. 2, E and F; and Fig. S3 B). In contrast, total CD4 and CD8 cells of healthy individuals were not affected by rapamycin or glycolysis inhibition (Fig. S3 C). Of note, the effect of rapamycin on ALPS-DNT cells most likely involves cell survival, as these cells are non-proliferative in vitro. However, an additional effect on proliferation is likely, as illustrated for anti-CD3/28 in vitro–activated CD4 or CD8 T cells (Fig. S3 D). In conclusion, ALPS-DNT cells share a metabolic profile similar to that of EBV-stimulated CD8⁺ T cells, predominantly driven by glycolysis.

Mitochondria are critical contributors to cellular energy generation and macromolecule biosynthesis. Moreover, mitochondrial membrane potential (MMP) and ROS impact signaling, stemness, and viability/longevity (35). EBV-CD8 showed an increase in mitochondrial mass as determined by MitoTracker Green staining (Fig. 3 A). Mitochondrial mass of ALPS-DNT was even higher (Fig. 3 A). Interestingly, MMP and cellular ROS as determined by tetramethylrhodamine methyl ester (TMRM) and CellROX staining were highest and more homogenously elevated in ALPS-DNT compared to EBV-CD8 and HD-CD8 (Fig. 3, B–E). Importantly, these analyses were performed directly ex vivo within 1–2 h after blood collection and immediate PBMC isolation. MMP and ROS dropped over time ex vivo, concomitant with the induction of cell death (Fig. S3, D and E), suggesting that the high ROS levels ex vivo reflect the in vivo state and are not a consequence of impaired viability. As a possible cause of high MMP and ROS, we reasoned that altered mitochondrial ultrastructure and function could lead to impaired proton utilization (increased MMP) and electron leakage (increased ROS). Electron microscopy revealed that ALPS-DNT had fissed, round-shaped mitochondria as expected for effector T cells (Fig. 3 F) but with significantly increased width of the cristae and reduced numbers of cristae per mitochondrion (Fig. S3 F). Interestingly, when investigating condition-specific pathways in silico, the ALPS-DNT model identified heme oxygenase 1 (HMOX1)–mediated heme degradation leading to carbon monoxide production, a pathway linked to oxidative stress response (Fig. 3 G), which was not seen in HD-CD8 and EBV-CD8 models. This aligns with the experimental finding of elevated ROS in ALPS-DNT. Exploring further differences in ROS control and ROS response, the ALPS in silico models had lower glutathione synthetase fluxes in comparison to EBV-CD8 models as indicated by random flux sampling (Fig. 3 H), indicating potential limitations in effective ROS scavenging. To further explore the potential of ROS scavenging in silico, we induced 2% mitochondrial superoxide production per consumed mmol oxygen (see Materials and methods) and investigated superoxide detoxification pathways by random sampling (Fig. 3 I). In line with the experimental observations, ALPS-DNT models had the highest rate of superoxide leakage caused by a decreased superoxide detoxification by ascorbic acid (Fig. 3, I and J). As the models recover oxidized ascorbic acid via NADPH-dependent glutaredoxin activity, we tested then the maximum production potential of NADPH for each cell-specific model, finding that ALPS-DNT models had far lower NADPH production capacity (7.71 mmol/gDW/h) than EBV-CD8 (14.499 mmol/gDW/h) and HD-CD8 models (31.538 mmol/gDW/h). Collectively, FCT displayed an unusual increase in MMP and ROS accumulation, while the in silico models hint at limitations in core ROS scavenging pathways.

We previously showed that FCT are prominent in ALPS patients but also exist in healthy individuals at low frequency with the same phenotype and the same clear-cut difference from other conventional T cell populations, including other DNT (1). The differentiation and transcriptional program of FCT are not restricted to DNT but can be found in small subsets of CD4 and CD8 T cells. For this reason, we also analyzed glucose uptake, glucose dependency, mitochondrial mass, MMP, and ROS production in FC-CD4, -CD8, and -DNT of healthy individuals. We used flow cytometry–based methods instead of Seahorse extracellular flux assays due to the small cell numbers. While in ALPS patients the majority of DNT carry the FCT signature, cells with this signature are rare in HD, below 1% among CD4 and CD8 and below 40% among DNT. We therefore specifically gated on CD38^highCD45RA^high (FC) versus non-CD38^highCD45RA^high “conventional” T cell populations for this analysis (Fig. 4 A). Interestingly, we found similarly elevated glucose uptake, increased mitochondrial mass, hyperpolarized mitochondria, and high ROS in all CD4, CD8, and DNT FCT subsets of HD as seen in their expanded FAS-deficient counterparts of ALPS patients (Fig. 4, B–E). FC-DNT of healthy individuals also showed impaired viability ex vivo and reduced survival upon glucose inhibition (Fig. 4 F and Fig. S4 A), but cell death as a result of 2-DG treatment was not as pronounced as in ALPS-DNT. In conclusion, the metabolic profile of FCT is intrinsic to this distinct T cell differentiation state and neither restricted to the DNT state nor to FAS deficiency.

DNT, and more precisely FC-DNT, rapidly decline in ALPS patients treated with the mTOR inhibitor rapamycin (15), which impacts many cellular processes, including glycolysis (18, 36). Since FCT survival and/or expansion is glucose metabolism dependent (Fig. 2, E and F), we hypothesized that the anti-glycolytic effect of rapamycin likely contributes to its efficacy. To address whether rapamycin reverts the metabolism of FCT to less glycolysis, we compared the glycolytic activity of ALPS-DNT sorted from one ALPS patient before and after 3 wk of clinically effective rapamycin treatment. DNT had declined from 9% to 6% of CD3 T cells. Interestingly, ECAR was no longer elevated in ALPS-DNT under rapamycin (Fig. 5 A). To address if rapamycin altered other metabolic features of FCT, we performed additional flow cytometry–based metabolic studies of DNT from ALPS patients before and under rapamycin. Rapamycin treatment reduced the frequency of FC-DNT among total ALPS-DNT (Fig. 5, B and C), but it did not change other metabolic features of the remaining FC-DNT, which continued to show elevated glucose uptake, mitochondrial mass, MMP, and ROS (Fig. 5 D). Similarly, in immunologically healthy individuals who received rapamycin for other diseases (Fig. S4, B and C), rapamycin reduced FC-DNT frequencies but did not change these investigated metabolic features. Thus, decreased glycolysis does not appear to be linked to reduced glucose/2NBDG uptake. In summary, mTOR inhibition impacts FCT survival and aerobic glycolysis but does not revert other hallmark metabolic features associated with FCT cell differentiation.

TCR activation in combination with CD28 co-stimulation enhances mTOR signaling and induces metabolic remodeling, including an increase in glycolysis, which has been directly linked to effector differentiation and function. While cell cycle–related TFs were similarly expressed by ALPS-DNT and EBV-CD8, only the latter showed high transcriptional levels of the TFs TBX21, ZEB2, and IRF8, which are typically expressed by CD8 effector T cells (Fig. 6 A). However, the expression of other effector and exhaustion-associated TFs, including EOMES, BATF, TOX/TOX2, and IRF2, was even higher in FCT than in EBV-CD8 (37, 38, 39, 40). Notably, FCT highly expressed TFs, including MYB, POU2AF1, PRDM8, and NFATC1, which in this combination clearly delineates them from other T cell populations. Strikingly, while IFNG transcript levels were elevated in FCT (Fig. S4 D), they were not able to produce IFNγ after stimulation with PMA/ionomycin (Fig. 6 B) ex vivo, and expression of perforin and granzyme B was almost absent (Fig. 6 B) (shown also in 1). In contrast, they produced high amounts of IL-10 and granzyme K (Fig. 6 B). This functional signature was shared by FC-CD4 and FC-CD8 (Fig. S4 E). Interestingly, exhausted CD8 T cells, best described in murine models of chronic LCMV infection, are also characterized by a dysfunctional effector state. To further position FCT between effector and exhausted CD8 T cells, we performed a gene set enrichment analysis using our human RNA-sequencing data and murine reference data of day 8 acute LCMV CD8 effector versus day 30 chronic (clone 13) LCMV-exhausted CD8 T cells (41). Both EBV-CD8 and FCT showed an enrichment of genes associated with acute effector rather than exhausted CD8 T cells (Fig. 6 C). Hence, FCT differentiation is associated with an effector-like glycolytic switch and high proliferative activity but remains uncoupled from classical effector differentiation. The FCT state is characterized by a distinct TF profile along with a more undifferentiated state regarding classical effector functions.

In this study, we dissect the metabolic programs that drive pathological T cell proliferation in ALPS—a disease in which highly proliferative, FCT evade apoptosis. We compared these FCT to CD8 T cells in acute EBV infection, a model of physiological, though sometimes excessive, antigen-driven T cell expansion. While both conditions are marked by T cell–mediated lymphadenopathy and splenomegaly, the identity of the disease-driving T cells and the nature of their activating triggers differ. By comparing these two conditions, we identified overlapping and distinct metabolic features with potential relevance for therapeutic intervention. By assessing genome-scale metabolic models reflective of the distinct metabolic wiring of the investigated T cell populations, we offer a resource to complement experimental and clinical research with in silico analyses capable of estimating metabolic pathway activities or flux states under various perturbations and conditions. These in silico analyses are useful to either support existing experimental findings or generate new hypotheses that need to be validated through further experiments.

Unexpectedly, we found that FCT engage a metabolic program as strongly glycolytic as that of acutely activated, virus-specific CD8 T cells during primary EBV infection. This is remarkable, given that EBV-driven T cells are often considered the benchmark for rapid proliferation and effector function—optimally tuned to mount a vigorous antiviral response. Yet, despite lacking a known antigenic stimulus, FCT showed equally high proliferative activity, elevated activation markers, active mTOR pathway signaling, and markedly increased glucose metabolism. Clinical observations have hinted at this glycolytic bias, as ALPS patients often show enhanced 18F-fluorodeoxyglucose Positron emission tomography (PET) positivity in enlarged lymph nodes (42, 43). Similarly, in Fas-deficient MRL/lpr mice, DNT exhibit higher baseline glycolysis than CD4 or CD8 T cells (44, 45). However, without comparison to a well-defined in vivo–activated T cell subset, the magnitude of glycolytic activity in FCT remained unclear. Here, we demonstrate that human FCT are as highly glycolytic as EBV-stimulated CD8 T cells. These findings suggest that uncontrolled T cell expansion relies on a shared, augmented glycolytic state. In vitro, FCT displayed a strong decrease in survival upon glucose metabolism inhibition, which was mirrored in the in silico analysis. This effect may not only reflect glucose dependency but possibly also decreased capacity of ALPS-DNT to utilize alternative metabolic pathways, an intrinsically higher susceptibility of ALPS-DNT to cell death and/or reduced functionality of glucose deprived moDCs in our experimental setting.

One of our main observations is that, despite their bioenergetic similarity to classical effector T cells, FCT fail to execute canonical effector differentiation programs. Whereas EBV-responsive CD8 acquire potent cytotoxic function, FCT, including FC-CD8 and their FC-DNT progeny, do not produce IFNγ, granzyme B, or perforin and thus remain functionally uncoupled from the classical CD8 effector cytokine and cytolytic mediator production. In murine effector T cells, glycolytic engagement of Ldha and Gapdh allows IFNγ transcription and translation, respectively (46, 47), suggesting a direct link between metabolism and effector function. How this mechanism relates to the proliferation–effector uncoupling in FCT remains to be clarified. Our results indicate that the usual association of mTOR activation, glycolysis, and effector differentiation does not apply across all proliferative CD8-derived subsets. In FCT, mTOR and glycolysis support a distinct differentiation program with an altered functional profile. Moreover, the combination of increased MYB and absent T-BET expression could play a role in preventing classical effector differentiation in FCT (48). Notably, increased glycolysis upon activation can also be linked to IL-4 or IL-17 instead of IFNγ production in the respective CD4 T helper subsets. Thus, not only metabolic cues but also T cell–intrinsic and –extrinsic factors, including mode of antigen encounter and cytokines, determine differentiation and effector profile. Interestingly, regulatory T cells that also highly express IL-10 exhibit only a partial metabolic shift to glycolysis upon activation—though not as robust as conventional effector T cells. FAO remains their dominant metabolic signature and is essential for lineage stability and their core regulatory functions (49, 50).

We also identify metabolic features unique to FCT, including markedly elevated MMP and ROS, absent in EBV-CD8. Consistent with reports of enlarged, rounded mitochondria in DNTs from MRL/lpr mice (45), human FCT displayed increased mitochondrial fission and altered cristae structure, with potential consequences for mitochondrial function (25). Nevertheless, OXPHOS was largely intact as indicated by comparable basal OCR. Previous studies in MRL/lpr mice even showed enhanced OXPHOS in DNT compared to CD4 and CD8 T cells (45). In our study, human FCT dying in vitro showed a rapid decrease of both the MMP and ROS levels, indicating that mitochondria represent a major source of ROS in FCT. When the MMP is high, electrons tend to back up in complexes I and III, increasing the probability of electron leakage to molecular oxygen, generating ROS. In addition, insufficient antioxidant capacity as suggested by our in silico analyses may contribute to the elevated cellular ROS but requires experimental validation (51, 52). ROS are known to promote activation and expansion of antigen-specific T cells (28) and act as signaling intermediates (53). Among ROS-responsive elements, HMOX1 and NFATC1 were particularly upregulated in FCT. However, excessive ROS can be cytotoxic, possibly contributing to the rapid ex vivo death of FCT and potentially to their short lifespan in vivo. In line with this, high MMP was shown to correlate with short-lived, non–self-renewing phenotypes, while low MMP is associated with stemness and longevity (35). Assuming a detrimental role of ROS, it would be interesting to explore the effects of ROS scavenging. However, mitochondrial ROS scavenging did not alter DNT frequency or lymphoproliferation in a mouse model of Fas deficiency (45), suggesting either the inefficiency of the scavenger used or the insufficiency of ROS scavenging in restoring viability in FCT. Chronic antigen-experienced exhausted CD8 T cells also exhibit elevated ROS and NFATC1, and FCT share some features with these cells, including upregulation of TOX, PD-1, CTLA-4, and LAG-3 (1, 54). However, FCT differ strikingly by their complete lack of T-BET, high expression of MYB and POU2AF1, sustained proliferation, and high IL-10 production—features not typical of exhausted cells. Moreover, FCT are highly glycolytic with increased MMP, whereas exhausted T cells generally show reduced glycolysis and depolarized mitochondria (24).

The highly glycolytic signature in combination with hyperpolarized mitochondria defines a distinct T cell differentiation state found in subsets of CD4, CD8, and DNT in both ALPS patients and healthy individuals. This suggests that FCT represent a physiologically relevant subset characterized by an intrinsic mTOR-driven glucose-dependent anabolic program. The signals that initiate this state and drive lymphoproliferation remain unknown but may include self-antigens. Our data suggest that these signals induce anabolic metabolism and proliferation but are not sufficient or processed in a different signaling context that fails to promote classical effector differentiation. Instead, they initiate a separate differentiation trajectory supported by a unique TF profile. In healthy individuals, FCT are tightly regulated and rapidly eliminated via FAS-mediated apoptosis. Hence, their life cycle appears to be designed for maximal, transient proliferative bursts at the expense of stemness and longevity. In ALPS, loss of the apoptotic FAS checkpoint permits uncontrolled expansion of these metabolically primed but functionally altered T cells, driving pathological lymphoproliferation without significant inflammation.

The glucose dependency of FCT survival could offer a rationale for targeted metabolic therapy. We show that glycolysis blockade with 2DG rapamycin strongly impairs FCT survival. This observation is clinically relevant, as rapamycin is already used therapeutically for ALPS, consistent with the observation that FCT are “addicted” to anabolic, mTOR-driven metabolism. Interestingly, unlike certain exhausted CD8 subsets (24), FCT partially retain an altered metabolic profile under rapamycin treatment, suggesting that rapamycin primarily limits survival rather than fully reprogramming FCT metabolism. Future studies should aim to define the stimuli that initiate and sustain FCT proliferation and explore additional metabolic interventions for targeted immunomodulation in ALPS and related lymphoproliferative disorders.

We included 24 ALPS-FAS/sFAS patients and 13 patients without known genetic predisposition, who presented with lymphadenopathy and splenomegaly in the context of serologically confirmed acute primary EBV infection. Table S1 summarizes genetic and clinical characteristics, including age and gender. Moreover, five immunologically healthy individuals before and after rapamycin treatment (mainly for vascular malformations) were analyzed. Finally, age-matched healthy individuals were analyzed in parallel. Experiments were conducted primarily using fresh blood samples of untreated individuals, which were processed on the day of blood sampling to accurately capture the in vivo cellular state. All participants gave informed consent. The study was conducted according to the Declaration of Helsinki and approved by Ethics Committees of the University of Freiburg (protocol numbers: 409/16, 282/11, 610/15, and 147/15).

Flow cytometry was performed with PBMC. The following antibodies were used: CD3 PerCP (clone SK7; BD Biosciences; RRID: AB_400513), CD8 BV510 (clone RPA-T8; BD Horizon; RRID: AB_314120), CD8 BUV395 (clone RPA-T8; BioLegend; RRID: AB_893425), CD8 PC7 (clone SFCI21Thy2D3; Beckman Coulter), CD8 Krome Orange (PN B00067; Beckman Coulter), TCRαβ FITC (clone T10B9.1A-31; BD Biosciences), TCRαβ BV421 (clone IP26; BioLegend), TCRαβ PE (clone IP26A; Beckman Coulter), CD45RA PC7 (clone HI100; BD Biosciences), CD45RA BV510 (clone HI100; BioLegend), HLA-DR PC7 (clone L243/G46-6; BD Pharmingen), HLA-DR FITC (clone G46-6; BD Biosciences), Ki-67 PC7 (clone B56; BD Biosciences), Ki-67 FITC (clone MIB-1; DAKO), CD4 APC-Cy7 (clone RPA-T4; BioLegend), CD4 PB (clone 13B8.2; Beckman Coulter), CD4 PE (clone 13B8.2; Beckman Coulter), CD4 FITC (clone RPA-T4; BD Pharmingen), CD4 APC (clone SK3; BD Pharmingen), CD38 PerCP (clone HIT2; BioLegend), CD38 APC (clone HIT2; BioLegend), CD38 BV421 (clone HB-7; BioLegend), T-BET PE (clone eBio4B10; eBioscience), EOMES APC (clone WD1928; eBioscience), IFN-γ FITC (clone B27; BD Biosciences), IL-10 PE (clone JES3-9D7; BioLegend), Perforin FITC (clone delta G9; Ancell), Granzyme B BV421 (clone GB11; BD Biosciences), Granzyme A PC7 (clone CB9; BioLegend), Granzyme K FITC (clone GM26E7; BioLegend), GLUT1 AlexaFluor647 (clone EPR3915; Abcam), HIF1α PE (clone 546-16; BioLegend), Annexin V FITC (BD Pharmingen), Annexin V BV605 (BD Horizon), and Annexin V APC (BD Pharmingen). Intracellular GLUT1 and HIF1α stainings were performed using the BD Cytofix/Cytoperm Kit. GLUT1 surface staining was performed with GLUT1-G125 (GLUT1 cell surface RBD ligand Kit, METAFORA biosystems). Perforin, granzyme B, Ki-67, and Eomes stainings were performed using the Human Regulatory T cell Staining Kit (eBioscience). PBMC were incubated with 2-NBDG-FITC (Life Technologies), MitoTracker Green (Thermo Fisher Scientific), TMRM (Thermo Fisher Scientific), and CellRox (Invitrogen) for 15 min at 37°C. To measure cytokine production, 4 × 10⁶ PBMC were stimulated with PMA (Sigma-Aldrich, 3.5 µg/ml) and ionomycin (Sigma-Aldrich, 2 µM) for 5 h and then analyzed using the Cytofix/Cytoperm Plus Kit and Golgi Plug (BD). Data were analyzed using FlowJo 7.2.5 or 10.4 software (Tree Star). Cell populations were either magnetically enriched using the Double-negative T cell Isolation Kit or the CD8 Isolation Kit (both Miltenyi Biotec) or sorted with MoFlo Astrios (Beckman Coulter) or FACS Aria Fusion (Becton Dickinson) cell sorters immediately after PBMC isolation on the day of blood collection.

ECAR and OCR were measured using the Seahorse XFe 96 bioanalyzer (Agilent) as previously described (55). 2 × 10⁵ MACS magnetic-activated cell sorting (MACS) enriched or sorted T cells per well (3 wells per sample) were spun onto poly-D-lysine–coated seahorse 96-well plates and preincubated at 37°C for a minimum of 45 min in the absence of CO₂. OCR and ECAR were measured under basal conditions and after the addition of the following drugs: 1 mM oligomycin, 1.5 mM flurorcarbonyl cynade phenylhydrazon, and 100 nM rotenone + 1 mM antimycin A (all Sigma).

For the survival assay 2 × 10⁵ PBMC per well were seeded into 96-well plates and stimulated either with medium alone or with allogenic monocyte-derived DC (moDC) (ratio 1:1, generated using the moDC Generation toolbox from Miltenyi Biotec) plus IL-21 (100 ng/ml, Miltenyi Biotec) and IL-10 (100 ng/ml, PreproTech) for 18 h at 37°C. As metabolic intervention, 2-DG (Sigma-Aldrich) or rapamycin (Selleckchem) was added at different concentrations. Survival was measured by flow cytometry, and annexin V staining was performed using the Annexin V binding buffer (BD Pharmingen). Survival normalized to day 0 was calculated as (cell count day 1/cell count day 0)*100. Survival normalized to the Stim condition was calculated as (cell count day 1*100/cell count day 1 Stim). For the proliferation assay, PBMCs were labeled with CFSE (Molecular Probes, C-1157), seeded into 96-well plates, and stimulated either with medium alone or with plate-bound CD3 OKT3 (14-0037; eBioscience) and soluble CD28 (555725; BD) in the presence of increasing concentrations of rapamycin (Selleckchem). Proliferation was measured after 5 days via flow cytometry.

DNT (CD3⁺TCRαβ⁺CD4⁻CD8⁻CD38⁺CD45RA⁺) of three ALPS-FAS patients, CD8 T cells (CD3⁺CD8⁺CD28⁺CD57⁻) of three HD (peripheral blood buffy coats), and CD8 T cell (CD3⁺CD8⁺CD38⁺) of three individuals with acute EBV infection (two peripheral blood samples and one bone marrow blood sample) were sorted. RNA-sequencing steps included RNA isolation (Qiagen RNeasy microkit), RNA-quality analysis (2100 Bioanalyzer, Agilent Technologies), creation of barcoded RNA-sequencing libraries (from 100 ng total RNA with Ovation Human RNA-Seq system, Nugen), and depletion of rRNAs (insert-dependent adaptor cleavage method, Nugen). Alignment and quantification were performed as in (1). Downstream analyses were performed using R 4.4.0 (R Core Team 2015). Low expressed genes (total raw counts across the 15 samples ≤5) were filtered out, and differential expression analysis was performed with the DESeq package v.1.44 (RRID:SCR_000154). Significantly regulated genes were selected based on their adjusted P value (Benjamini and Hochberg) <0.05. All genes presented in the figures of this manuscript meet this significance threshold. Gene set enrichment was performed using clusterProfiler package v.4.12.6 (RRID:SCR_016884) with MSigDB gene sets v7.0. Adjusted P value <0.05 was considered as significant. Data are available under GEO Accession viewer (GSE154929 and GSE319376).

A graphical summary of the workflow used for the analysis is provided in Fig. S1. We processed transcriptomic data from three T cell types (ALPS-DNT, EBV-CD8, and resting HD-CD8), each with three biological replicates, by generating mean expression profiles and mapping them to the Recon3D metabolic model (29). Using these profiles and a core set of 25 reactions, curated from a whole-body model (56) to include reactions expected to be active in CD8 T cells, as input, we reconstructed context-specific models via the fastCore algorithm (30) within the COBRA Toolbox v3.0 (57). These models were subsequently constrained with a simulated physiological T cell medium and further refined using the eFlux algorithm (30) based on gene expression data. Note that the only difference between the three cell type models is constituted by the input transcriptome data. All other aspects of the modeling, including the simulated medium, in silico experimental procedures, and settings within the fastCore algorithm, are identical across the three conditions.

A series of metabolic flux analyses were then performed, including FBA to predict maximum biomass under unconstrained, physiologically constrained, and eFlux-constrained conditions and flux variability analysis (FVA) to delineate the feasible solution space at optimal biomass. We also assessed the relative contributions of glycolysis and OXPHOS to ATP production, simulating anoxia by blocking oxygen uptake and impairing glycolysis by blocking hexokinase reaction. We further simulated FAO inhibition by blocking all reactions facilitated by CPT1, PPP (Pentose Phosphate Pathway) inhibition by blocking the enzyme G6PD, and blocking glutaminolysis by either blocking glutamine uptake or the mitochondrial glutaminolysis. Finally, the optGpSampler (58) algorithm generated 10,000 flux samples per model to characterize probable flux distributions, and we repeated the random sampling using a second sampling algorithm (Riemannian Hamiltonian Monte Carlo [RHMC] sampling [https://doi.org/10.48550/arXiv.2202.01908]) for sensitivity analysis. Results in key reactions were comparable across the two sampling algorithms. Key reactions were visualized using kernel density estimation and box plots. To test in silico the oxidative stress response, we induced 2% superoxide production per consumed mmol oxygen in the mitochondrial electron chain in all three models. Then, we investigated superoxide leakage and superoxide detoxification pathways via random sampling with the RHMC sampling, setting 10,000 samples as the target. Finally, we assessed NADPH production capacity by adding a NADPH sink reaction to the models and performing FVA. Context-specific metabolic models, full FVA results, random sampling, all scripts, and the complete modeling output can be accessed via https://github.com/SysPsyHertel/CodeBase/tree/main/Scripts_Huang_Tcell.

Sorted T cells (1.2  ×  106 DNT of an ALPS patient and 2  ×  10⁶ CD3 of two healthy individuals) were fixed in 2.5% glutaraldehyde in 100 mM sodium cacodylate and washed in cacodylate buffer. After dehydration, samples were embedded in Eponate 12 resin (Ted Pella), and sections were cut. Images were acquired using a FEI Tecnai 12 Transmission electron microscope equipped with a Veleta digital camera. Brightness and contrast were adjusted using ImageJ 1.44o (NIH) (RRID:SCR_003070). The analysis was performed blindly and using ImageJ 1.44o (NIH): only entire mitochondria were included in the analysis; ratio length/width, maximal cristae width, and numbers of cristae per mitochondrion were measured.

Statistical analyses were performed with GraphPad Prism software (RRID:SCR_002798). T tests (Mann–Whitney U), Wilcoxon test or ANOVA, Kruskal–Wallis test for multiple comparisons, and mixed-effects analysis with Tukey's multiple comparisons test were applied to compare T cell marker expression or frequencies (*P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001). If a comparison was not significant, no sign or ns was added. To allow for comparisons across multiple acquisition dates of the flow cytometry data, we normalized the data to the values of the CD4 subset of the same individual.

The supplemental material includes one table and four figures. Fig. S1 shows GLUT1 surface expression and the workflow for the generation and analysis of context-specific metabolic models in silico. Figs. S2 and S3 extend data on glycolytic and OXPHOS metabolism and on survival assays. Fig. S4 adds metabolic data on FCT in HD and on the analysis of the effector profile of FCT. Table S1 shows the clinical and genetic features of the individuals involved in the study.

The data that support the findings of this study are available from the corresponding author upon reasonable request. The RNA-sequencing data are available under GEO Accession viewer (GSE154929 and GSE319376). All in silico metabolic models generated in this study and the corresponding analysis scripts can be downloaded freely from https://github.com/SysPsyHertel/CodeBase/tree/main/Scripts_Huang_Tcell.

We are extremely grateful to Karen A. Fortner and Ralph C. Budd for the valuable scientific discussions. We are grateful for the excellent technical support by Séverine Kayser. The Core Facility for Electron Microscopy (EMcore) at the University Freiburg Medical Center—IMITATE is registered with the Deutsche Forschungsgemeinschaft (DFG) under the reference number RI_00555. The graphical abstract was created in BioRender.

M.E. Maccari was supported by the IMMediate Advanced Clinician Scientist-Program, Department of Medicine II, Medical Center University of Freiburg and Faculty of Medicine, University of Freiburg, Germany, funded by the Bundesministerium für Bildung und Forschung (BMBF, Federal Ministry of Education and Research)—01EO2103. The work of R.I. Klein Geltink is supported by a Michael Smith Health Research BC Scholar Award. The work of J. Hertel and C. Huang was supported by DFG project ID 499552394 (SFB 1597/1). We acknowledge funding from the DFG within the CRC1160 (Project ID 256073931-Z02 to M. Börries), CRC/TRR167 (Project ID 259373024-Z01, M. Börries), CRC1453 (Project ID 431984000-S1, M. Börries), CRC1479 (Project ID 441891347- S1 to M. Börries), TRR 359 (Project ID 491676693-Z01, M. Börries), and FOR 5476 UcarE (Project ID 493802833-P7, M. Börries). We also acknowledge funding from the German BMBF within the Medical Informatics Funding Scheme PM4Onco–FKZ 01ZZ2322A (M. Börries) and EkoEstMed–FKZ 01ZZ2015 (G. Andrieux). M. Fischer was supported by the IMM-PACT Clinician Scientist Program, Faculty of Medicine, University of Freiburg, funded by the DFG, project number 413517907, Siegenthaler Foundation and Faculty of Medicine, University of Zurich. The work of B. Bengsch was supported by DFG project #505372148 (FOR 5560) and project #256073931 (SFB1160). O. Groß was supported by the DFG through SFB 1160 (Project ID 256073931), SFB/TRR 167 (Project ID 259373024), SFB 1425 (Project ID 422681845), SFB 1479 (Project ID 441891347), SFB/TRR 417 (Project ID 540805631), GRK 2606 (Project ID 423813989), and, under Germany’s Excellence Strategy, through Centre for Integrative Biological Signalling Studies—EXC-2189 (Project ID 390939984), as well as by the European Research Council through Starting Grant 337689, Proof-of-Concept Grant 966687, and the EU-H2020-MSCA-COFUND EURIdoc programme (101034170).

Author contributions: Maria Elena Maccari: conceptualization, data curation, formal analysis, investigation, methodology, resources, validation, visualization, and writing—original draft, review, and editing. Christoph König: data curation, formal analysis, investigation, and writing—review and editing. Geoffroy Andrieux: data curation, formal analysis, software, visualization, and writing—review and editing. Patrick Kury: formal analysis and investigation. Sarah A. Berger: investigation and methodology. Jasmin Mann: investigation and methodology. Beth Kelly: investigation. Simon Völkl: conceptualization, data curation, formal analysis, investigation, resources, validation, and writing—review and editing. Chenglong Huang: formal analysis, methodology, software, validation, visualization, an writing—original draft, review, and editing. Martin Helmstädter: data curation, investigation, and methodology. Simon Lagies: formal analysis and investigation. Marco Fischer: resources and writing—review and editing. Francesc Baixauli: resources. Oliver Gorka: investigation and resources. Olaf Groß: conceptualization and funding acquisition. Markus Hufnagel: resources and writing—review and editing. Sarah Salou: resources. Susan Farmand: writing—review and editing. Sabine Heine: resources. Volker Schuster: resources. Wolfgang Willenbacher: conceptualization, data curation, project administration, validation, and writing—review and editing. Gregor Dückers: resources. Bodo Grimbacher: resources and writing—review and editing. Klaus Warnatz: resources and writing—review and editing. Friedrich G. Kapp: resources and writing—review and editing. Myriam Lorenz: investigation. Miriam Groß: resources. Jens Wittner: formal analysis, resources, and validation. Roland Elling: resources and writing—review and editing. Luca Scorrano: resources. Iwona A. Koenig: conceptualization, methodology, and writing—review and editing. Bertram Bengsch: resources and writing—review and editing. Carsten Speckmann: data curation, resources, and writing—review and editing. Bernd Kammerer: formal analysis and investigation. Melanie Börries: formal analysis and writing—review and editing. Klaus Schwarz: investigation, supervision, and writing—review and editing. Johannes Hertel: conceptualization, formal analysis, methodology, supervision, and writing—review and editing. Erika L. Pearce: funding acquisition, resources, and supervision. Stephan Ehl: conceptualization, funding acquisition, supervision, and writing—original draft, review, and editing. Ramon I. Klein Geltink: conceptualization, data curation, formal analysis, investigation, methodology, project administration, supervision, visualization, and writing—original draft, review, and editing. Anne Rensing-Ehl: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, validation, visualization, and writing—original draft, review, and editing.