Metabolic adaptations rewire CD4+ T cells in a subset-specific manner in human critical illness with and without sepsis

Stier, Matthew T.; Sewell, Allison E.; Mwizerwa, Erin L.; Sim, Chooi Ying; Tanner, Samantha M.; Nichols, Casey M.; Durai, Heather H.; Jennings, Erin Q.; Lindau, Paul; Wilfong, Erin M.; Obeidalla, Sarah N.; Kerchberger, V. Eric; Newcomb, Dawn C.; Bastarache, Julie A.; Ware, Lorraine B.; Rathmell, Jeffrey C.

doi:10.1038/s41590-025-02390-6

Download PDF

Article
Open access
Published: 15 January 2026

Metabolic adaptations rewire CD4⁺ T cells in a subset-specific manner in human critical illness with and without sepsis

Nature Immunology (2026)Cite this article

4782 Accesses
1 Citations
22 Altmetric
Metrics details

Subjects

Abstract

Metabolic and immunologic dysfunction, including pathological CD4⁺ T cell immunosuppression, are archetypal in critical illness, but whether these factors are mechanistically linked remains incompletely defined. Here we characterized the metabolic properties of human CD4⁺ T cells from critically ill patients with and without sepsis and healthy adults. CD4⁺ T cells in critical illness showed subset-specific metabolic plasticity, with regulatory T (T_reg) cells preferentially acquiring glycolytic capacity that associated with sustained cellular fitness and worsened clinical illness. Adapted T_reg cells were more metabolically flexible and stabilized suppressive markers FOXP3 and TIGIT under mitochondrial stress. Single-cell transcriptomics suggested reactive oxygen species (ROS) and kynurenine metabolism as drivers of T_reg cell remodeling. Subsequent inhibition of ROS and kynurenine metabolism attenuated glycolytic adaptation and suppressive rewiring, respectively, in T_reg cells. These findings indicate that metabolic dysfunction was a contributor to CD4⁺ T cell remodeling in critical illness and suggest avenues to restore effective immunity.

Regulation of CD4 + T cell differentiation and function by glucose metabolism

Article 05 July 2025

Metabolic control of innate-like T cells

Article 08 September 2025

Mitochondrial dysfunction promotes the transition of precursor to terminally exhausted T cells through HIF-1α-mediated glycolytic reprogramming

Article Open access 27 October 2023

Main

Critical illness (CI) requiring intensive care unit (ICU) admission and life support due to severe organ failure affects over 5 million patients annually in the USA¹. Among the most common ICU diagnoses is sepsis, a dysregulated host response to infection causing end-organ injury². Sepsis accounts domestically for over 200,000 deaths per year and more than US$24 billion in healthcare expenditures^3,4. Current management includes antimicrobials and/or surgical source control to address the inciting infection alongside supportive care measures, including vasopressors, mechanical ventilation and renal replacement therapy, to sustain vital functions while host repair mechanisms mediate recovery. Despite decades of research, no targeted therapies exist to effectively treat the pathophysiology of sepsis⁵.

Immunologic dysfunction is a hallmark of sepsis. Early sepsis features excessive inflammation with increased levels of proinflammatory cytokines including tumor necrosis factor (TNF) and interleukin-1β (IL-1β) (referred to as a ‘cytokine storm’) that contribute to tissue injury⁶. In parallel, a compensatory but disproportionate immunosuppressive response develops and can continue beyond the acute phase. This impaired immune state predisposes patients to secondary infections, persistent organ dysfunction, hospital readmissions and death^7,8,9. CD4⁺ T cells are central to this immunosuppressive remodeling, with effector CD4⁺ T cells acquiring an ‘exhausted-like’ phenotype characterized by high PD-1 and LAG-3 expression, attenuated TNF and interferon-γ (IFNγ) secretion and increased apoptosis, whereas suppressive T_reg cells are relatively preserved^10,11,12. The severity of this cell-mediated immune impairment is underscored by the frequent detection of cytomegalovirus (CMV), Epstein–Barr virus (EBV) and herpes simplex virus (HSV) in critically ill patients¹³ at rates similar to solid organ transplant recipients receiving T cell immunosuppressants¹⁴. Although these phenotypes are consistently observed, the pathways underlying CD4⁺ T cell immunosuppression in human CI and sepsis remain incompletely understood.

Metabolic dysfunction is pervasive in CI, including sepsis¹⁵. The metabolic state of immune cells is a mechanistic determinant of their fate and function. Immunometabolism is central to the pathogenesis of obesity, autoimmunity, cancer and SARS-CoV-2 infection^16,17 and metabolic reprogramming has emerged as a promising therapeutic target in these conditions¹⁸. Previous human studies from individuals with sepsis have reported abnormalities in mitochondrial complex I and V activities and broad impairments in glycolysis and oxidative phosphorylation among bulk T lymphocytes^19,20; however, the overall role of CD4⁺ T cell immunometabolism in sepsis and other CI syndromes remains understudied, particularly stratified by CD4⁺ T cell subset.

In this study, we used biospecimens from critically ill patients to comprehensively profile the metabolic dependencies and capacities of CD4⁺ T cell subsets. We showed that reactive oxygen species (ROS)-driven glycolytic reprogramming, together with increased kynurenine metabolism, heightened metabolic plasticity in T_reg cells and promoted their resilience and suppressive function at the expense of conventional CD4⁺ T cell subsets. This enhanced glycolytic capacity in T_reg cells was additionally associated with worse clinical outcomes.

Results

ICU-derived T_reg cells have higher glycolytic capacity

We prospectively collected peripheral blood mononuclear cells (PBMCs) from 156 patients in the medical and surgical ICUs at Vanderbilt University Medical Center upon ICU admission, as well as serially, at day 2 and day 7 post-admission if available (Fig. 1a and Extended Data Table 1). Participants were adjudicated into critically ill patients who were nonseptic (CI-NS; n = 75, 57% male, median age 63 years, IQR 55–70) and critically ill patients with sepsis (CI-Sep; n = 81, 62% male, median age 65 years, IQR 56–75; Extended Data Fig. 1 and Extended Data Table 1). Nonacutely ill healthy control participants (NHCs; n = 24, 54% male, median age 54 years, IQR 42–58) were recruited from the community as controls. Complete blood counts from CI-NS and CI-Sep showed a mixed neutrophilic (NHC median 3.72 × 10³ μl⁻¹, CI-NS 7.12 × 10³ μl⁻¹ and CI-Sep 13.51 × 10³ μl⁻¹) and monocytic (NHC median 0.42 × 10³ μl⁻¹, CI-NS 0.57 × 10³ μl⁻¹ and CI-Sep 0.67 × 10³ μl⁻¹) response with marked lymphopenia (NHC median 1.92 × 10³ μl⁻¹, CI-NS 0.99 × 10³ μl⁻¹ and CI-Sep 0.83 × 10³ μl⁻¹) (Fig. 1b,c), as reported²¹, a characteristic hematologic pattern in critically ill patients reflecting heightened innate immune activation and loss of circulating lymphocytes.

**Fig. 1: CD4⁺ T cells exhibit subset-specific metabolic adaptations in the context of critical illness with and without sepsis.**

To decipher the immunometabolic landscape of CD4⁺ T cells in CI-NS and CI-Sep, we used a puromycin-incorporation assay in 9 NHC, 30 CI-NS and 32 CI-Sep samples collected 48–72 h (day 2) post-ICU admission. CD4⁺ T cell subsets, including CD45RA⁺CCR7⁺ naive T (T_N) cells, CD45RA⁻CCR7⁻ effector memory T (T_EM) cells, CD45RA⁻CCR7⁺ central memory T (T_CM) cells, CD25⁺FOXP3⁺ T_reg cells, CXCR3⁺CCR6⁻ type 1 helper T (T_H1) cells and CCR4⁺CCR6⁺ T_H17 cells, had similar dependence on glucose, fatty acid oxidation and glutaminolysis compared to NHCs (Fig. 1d and Extended Data Fig. 2a,b), whereas mitochondrial dependence and glycolytic capacity broadly differed between CI and NHCs (Fig. 1e,f). Compared to NHCs, CI T_reg cells showed the greatest increase in glycolytic capacity (median CI-NS +15%, CI-Sep +34%), whereas T_EM (CI-NS +2%, CI-Sep +12%), T_H1 (CI-NS +4%, CI-Sep +9%) and T_H17 (CI-NS +8%, CI-Sep +6%) had the least increase (Fig. 1f). These differences were higher in CI-Sep compared with CI-NS (Fig. 1f). There was broadly increased metabolic output across all CD4⁺ T cell subsets in CI compared with NHC (Fig. 1g). T_reg cell glycolytic capacity remained persistently raised up to day 7 of hospitalization (Fig. 1h and Extended Data Fig. 2c). Two metabolically distinct programs marked by sensitivity (Puro^lo) or resistance (Puro^hi) to the mitochondrial ATPase inhibitor oligomycin were detected in T_reg cells from NHCs, whereas CI-Sep T_reg cells were considerably enriched in Puro^hi cells (Fig. 1i,j). Collectively, although their baseline dependence on glucose was unchanged, CI T_reg cells showed reduced mitochondrial dependence and higher overall glycolytic capacity, indicating metabolic flexibility as part of a putative stress-adaptive program.

While lymphopenia was evident in both CI-NS and CI-Sep, the degree of attrition varied across CD4⁺ T cell subsets (Fig. 1k). At day 2 post-ICU admission, CI-Sep patients showed reduced frequencies of conventional CD4⁺ T (T_conv) cell) subsets relative to NHCs, including T_N cells (median 9% versus 16%), T_EM cells (5% versus 13%), T_H1 cells (2% versus 6%) and T_H17 cells (0.3% versus 0.6%). By contrast, the frequency of T_reg cells was preserved in CI-Sep (0.8%) compared with NHC (0.9%) (Fig. 1k). Estimated absolute cell counts showed a similar tendency in a subset of 4 NHCs, 13 CI-NS and 18 CI-Sep (Extended Data Fig. 2d), aligning with reports that stable or increased T_reg cell frequencies in sepsis reflect a loss of T_conv cell subsets¹⁰. T_reg cell proliferation was similar in CI compared to NHCs (Extended Data Fig. 2e). Increased acquisition of glycolytic capacity, represented as the change in CI-Sep versus NHC mean glycolytic capacity for each CD4⁺ T cell subset, directly associated with CD4⁺ T cell subset resistance to attrition (Fig. 1l and Extended Data Fig. 2f), consistent with metabolic plasticity as a determinant of cellular persistence in CI and sepsis that ultimately favors T_reg cells over CD4⁺ T_conv cells.

Profiling of glycolytic and oxidative features of CD4⁺ T cell subsets indicated that glucose transporter 1 (GLUT1) and the glycolytic enzymes hexokinase 1 (HK1) and lactate dehydrogenase A (LDHA) were broadly increased in CI-NS and CI-Sep (Fig. 2a-c), whereas GAPDH was not substantially changed compared to the NHC CD4⁺ T cell subset counterparts (Extended Data Fig. 3a). Phosphofructokinase (PFKP) was modestly decreased in CI-Sep T_reg cells compared with NHC T_reg cells (Extended Data Fig. 3b). Additionally, the mitochondrial electron carrier cytochrome c, which is central to oxidative phosphorylation, was similarly expressed between CI and NHCs across all CD4⁺ T cell subsets (Fig. 2d). Analysis of mitochondrial attributes showed that CD4 T_N cells had higher mitochondrial mass, mitochondrial membrane potential (MMP) and ratio of MMP:mass in CI-Sep compared with NHCs (Fig. 2e-i). CI-Sep T_EM cells were enriched for a population of cells with high mitochondrial mass, but negligible MMP (Mass^hiMMP^lo) (Extended Data Fig. 3c), consistent with severe mitochondrial dysfunctional and/or cell death and aligning with the high rate of attrition of CD4⁺ T_EM cells in CI. Overall, CI drove broad metabolic remodeling across CD4⁺ T cell subsets, with T_reg cells showing the most pronounced glycolytic augmentation and resistance to attrition.

**Fig. 2: Glycolytic and mitochondrial characteristics of CD4⁺ T cell subsets in CI with and without sepsis.**

T_reg cells retain suppressive features with stress in CI

To assess the functional capacities of CD4⁺FOXP3⁻ T_conv and CD4⁺FOXP3⁺T_reg cells in CI, PBMCs isolated from 10 NHCs, 14 CI-NS and 13 CI-Sep were stimulated with antibodies against CD3, CD28 plus CD2 (CD3/CD28/CD2 antibodies) to stimulate the T cell antigen receptor for 24 h and analyzed by flow cytometry. The frequency of TNF⁺, IL-17A⁺ and IL-4⁺CD4⁺ T_conv cells was higher, whereas the frequency of IFNγ⁺CD4⁺ T_conv cells was similar, in CI-Sep compared to NHCs (Fig. 3a,b and Extended Data Fig. 4a); however, the mean fluorescence intensity (MFI) of TNF among TNF⁺CD4⁺ T_conv cells was lower in CI-Sep compared with NHCs (Fig. 3b), consistent with an exhausted-like phenotype. IL-2⁺CD4⁺ T_conv cells were also more frequent in CI-NS and CI-Sep compared with healthy controls (Fig. 3b), potentially reflecting autocrine and/or paracrine survival signaling. CD4⁺ T_conv cells in CI-NS and CI-Sep exhibited increased frequency and MFI of activation (CD25) and checkpoint (PD-1 and LAG-3) proteins compared with CD4⁺ T_conv cells in NHCs (Fig. 3c), indicating a mixed phenotype of activation and impairment consistent with their cytokine profile.

**Fig. 3: Mitochondrial stress triggers altered functionality in CD4⁺ T cells.**

T_reg cells also exhibited functional and phenotypic differences in CI-NS and CI-Sep compared with NHCs. CI-NS and CI-Sep had a higher frequency of PD-1⁺ and LAG-3⁺ T_reg cells, markers associated with suppressive effector T_reg cells in humans, compared with NHCs^22,23 (Fig. 3d,e). In addition, CD39 and TIGIT are markers of highly suppressive T_reg cells that inhibit T_conv cells by several mechanisms including participating in the hydrolysis of proinflammatory extracellular ATP²⁴ and by promoting secretion of (FGL2²⁵, respectively. Both CD39 and TIGIT had sustained expression across NHC, CI-NS and CI-Sep T_reg cells compared with NHCs (Fig. 3d,e). The frequency and MFI of IL-10, as well as expression of proinflammatory cytokines like IFNγ and IL-17A were also comparable in NHC, CI-NS and CI-Sep T_reg cells (Fig. 3e and Extended Data Fig. 4b). CI-Sep and CI-NS T_reg cells had increased expression of CD25, which is associated with suppressive function²⁶, and similar expression of the transcription factor FOXP3 compared with NHC T_reg cells (Fig. 3e). Together, these expression patterns point to an overall more suppressive phenotype of CI T_reg cells relative to NHC T_reg cells.

Increased metabolic plasticity in subsets of CD4⁺ T cells may enable adaptation to mitochondrial stress²⁷, which is systemic in CI. To investigate this interplay, we stimulated PBMCs from 10 NHC, 14 CI-NS and 13 CI-Sep with or without a sublethal dose (10 nM) of the mitochondrial electron transport complex V inhibitor oligomycin for 24 h to partially impair mitochondrial energetics and simulate ICU-associated mitochondrial stress. Expression of FOXP3, which directly correlates with T_reg cell suppressor activity²⁸, decreased in NHC T_reg cells (−16%), but was relatively preserved in CI-NS (−9%) and CI-Sep T_reg cells (−4%) incubated with oligomycin (Fig. 3f,g). The frequency of highly suppressive TIGIT⁺ T_reg cells increased in CI-NS (+21%) and CI-Sep (+23%) but not in NHC (−4%) when stimulated and exposed to oligomycin compared with stimulation alone, although TIGIT MFI in TIGIT⁺ T_reg cells was modestly and equivalently reduced in NHC, CI-NS and CI-Sep T_reg cells under the same conditions (Fig. 3f,h and Extended Data Fig. 4c). Expression of CD25, CD39 and IL-10 changed variably in T_reg cells incubated with oligomycin, but the changes did not differ between NHC, CI-NS and CI-Sep (Fig. 3g,h and Extended Data Fig. 4c,d). The frequency of LAG-3⁺ T_reg cells was reduced in CI-NS and CI-Sep, but not in NHC stimulated and oligomycin-incubated PBMCs, compared with stimulation alone (Extended Data Fig. 4e). Thus, the more metabolically plastic T_reg cells in CI had increased resistance to mitochondrial stress and maintained expression of FOXP3 and TIGIT.

The frequency of TNF⁺, IFNγ⁺, IL-4⁺ and IL-17A⁺ T_conv cells decreased to varying degrees in stimulated oligomycin-exposed PBMCs compared with stimulation alone, but the magnitude of reduction was comparable between NHC, CI-NS and CI-Sep (Fig. 3i). There were modest differences in the MFI of IL-17A and TNF in CD4 T_conv cells in CI-Sep versus NHCs after oligomycin treatment (Extended Data Fig. 4f). By contrast, expression of CD25, PD-1 and LAG-3 were more strongly reduced in oligomycin-treated CD4 T_conv cells in CI-NS and CI-Sep compared with NHCs (Extended Data Fig. 4g). Expression of the pro-survival cytokine IL-2 was impaired in oligomycin-treated CD4 T_conv cell from CI-NS and CI-Sep compared with NHCs (Fig. 3i), underscoring a metabolically enforced vulnerability that may predispose to attrition in CI. In summary, CI T_reg cells resisted mitochondrial stress and thereby maintained a suppressive phenotype, whereas CI T_conv exhibited persistent vulnerability to mitochondrial impairment with loss of effector function.

CI globally remodels CD4⁺ T cell transcriptional states

To investigate the mechanisms of CD4⁺ T cell metabolic remodeling, we performed single-cell RNA sequencing (scRNA-seq) on PBMCs from 9 NHC, 19 CI-NS and 19 CI-Sep donors. This yielded a dataset consisting of 644,147 cells, which were annotated using canonical transcriptional markers as 9 major cell lineages and 27 distinct cell populations (Fig. 4a,b and Extended Data Fig. 5). The proportion of total CD14⁺ and CD16⁺ monocytes was increased and that of CD4⁺ T lymphocytes was reduced in CI compared with NHCs (Fig. 4c). Among the CD4⁺ T cell subsets, we observed a reduction or tendency for reduction in the proportion of CD4⁺ T_N cells and CD4⁺ T_EM cells and a relative preservation of T_reg cell proportion in CI compared with healthy controls (Fig. 4d), broadly consistent with the flow cytometry data.

**Fig. 4: Single-cell transcriptomic map of CI.**

Pseudobulked differential gene expression analysis with DESeq2 to evaluate sample-level differences indicated that most transcriptional changes were unique to CI-Sep, or shared between CI-NS and CI-Sep (Fig. 5a and Supplementary Table 1). Accordingly, we focused subsequent analyses on CI-Sep compared with NHCs. All CI-Sep CD4⁺ T cell subsets had increased expression of inflammatory regulators (SOCS1 or SOCS3) among the top differentially expressed genes (Fig. 5b). CI-Sep CD4⁺ T_EM cells showed increased expression of activation-related (NFKBIZ and IL22) and negative feedback regulation (DUSP5) transcripts, whereas CI-Sep T_reg cells exhibited increased expression of a glucocorticoid-responsive transcript (FKBP5) (Fig. 5b). Over-representation analysis found enrichment of cellular stress pathways across all CI-Sep CD4⁺ T cell subsets compared with NHCs (Fig. 5c and Supplementary Table 2). CI-Sep CD4⁺ T_N cells showed transcriptional evidence of cellular activation (GOBP_LYMPOCYTE_ACTIVATION, GOBP_REGULATION_OF_CATABOLIC_PROCESS) (Fig. 5c), consistent with known bystander effects in sepsis^29,30, whereas CI-Sep T_reg cells showed disproportionate enrichment of pathways related to glucose catabolism (GOBP_GLUCOSE_CATABOLIC_PROCESS) as well as NADH regeneration (GOBP_NADH_REGENERATION) (Fig. 5c), which supports cellular redox balance.

**Fig. 5: Transcriptomic features of CD4⁺ T cell subsets in CI.**

To further understand the transcriptional states of the CD4⁺ T cell subsets, we examined the expression patterns of individual genes canonically associated with core effector and regulatory programs in CD4⁺ T_N, CD4⁺ T_EM and T_reg cells. Activation markers (CD69, ICOS and TFRC) were increased across all CD4⁺ T subsets in CI-Sep compared to NHCs (Fig. 5d). Some transcripts increased (IFNG, IL17A and IL22) and others decreased (IL2 and TNF) in CD4⁺ T_EM cells in CI-Sep compared to NHCs (Fig. 5d). Mediators associated with wound healing and tissue repair (TGFB1 and AREG) were increased in CI-Sep CD4⁺ T_N and T_EM cells compared to NHCs (Fig. 5d), suggesting a role in tissue homeostasis beyond antimicrobial defense. CI-Sep T_reg cells had increased expression of IL10, increased expression of transcripts encoding checkpoints markers (CTLA4, PDCD1 and LAG3), and persistently high expression of TIGIT and ENTPD1 compared to NHC T_reg cells (Fig. 5d). CI-Sep T_reg cells also showed comparable expression of FOXP3 relative to NHC T_reg cells (Fig. 5d). STAT5A was increased in CI-Sep CD4⁺ T_N cells, CD4⁺ T_EM cells and T_reg cells compared to NHCs (Fig. 5d), likely reflecting enhanced IL-2 and IL-7 signaling tone. CI-Sep CD4⁺ T_EM cells and T_reg cells exhibited broadly remodeled chemokine and homing receptor expression (for example decreased CD4⁺ T_EM cell CXCR3 and increased T_reg cell CCR5 in CI-Sep relative to NHCs) (Fig. 5d), suggesting altered tissue recirculation. Finally, CD4⁺ T_N, CD4⁺ T_EM and T_reg cells in CI-Sep expressed increased pro-apoptotic (BAX and BAK1) and anti-apoptotic (BCL2 and MCL2) transcripts compared to the NHC counterparts (Fig. 5d), consistent with severe cellular stress and a compensatory survival program.

We orthogonally validated the over-representation analysis approach with gene set enrichment analysis (GSEA) using the Kyoto Encyclopedia of Genes and Genomes (KEGG) library, which broadly encompasses cellular processes and includes a robust group of metabolic gene sets. Very few pathways were significantly downregulated, limited to those related to spliceosome, herpes simplex virus 1 infection and metabolism of xenobiotics by CYP450 in CI-Sep CD4⁺ T_EM cells relative to NHC CD4⁺ T_EM cells (Fig. 5e and Supplementary Table 3). CD4⁺ T_N cells were marked by enrichment of pathways related to bacterial infection, apoptosis and inflammatory signaling (Fig. 5e). JAK-STAT signaling was significantly enriched in CI-Sep CD4⁺ T_EM and T_reg cells compared to NHCs, with additional enrichment of leukocyte transendothelial migration and microRNA pathways in CI-Sep T_reg cells compared to NHCs (Fig. 5e). All CD4⁺ T cell subsets showed significant enrichment of HIF-1 signaling in CI-Sep compared to NHCs (Fig. 5f), consistent with widespread cellular hypoxia, a context in which metabolic plasticity may be critical to sustain cellular fitness and function. Glycolysis was enriched in CI-Sep CD4⁺ T_N cells and T_reg cells with minimal change in CD4⁺ T_EM compared to their NHC counterparts (Fig. 5f). A trend toward oxidative phosphorylation enrichment was observed only in CI-Sep T_reg cells relative to NHC T_reg cells (Fig. 5f).

Evaluation of individual metabolic gene expression profiles in single cells indicated all CI-Sep CD4⁺ T cell subsets showed increased expression of genes involved in glucose transport (SLC2A3, encoding GLUT3), glycolytic processing (HK1, PFKP and PKM) and lactate production (LDHA) compared to NHCs (Fig. 5g). Tricarboxylic acid cycle and oxidative phosphorylation transcript abundance changes in CI-Sep CD4⁺ T_N, CD4⁺ T_EM and T_reg cells were more modest and heterogenous compared to NHCs (Fig. 5g). Hypoxia-related genes (HIF1A, SLC16A3 and DDIT4) were consistently higher across CI-Sep CD4⁺ T_N, CD4⁺ T_EM and T_reg cells compared to NHCs, as were several other core metabolic regulators including in fatty acid oxidation (CPT1A) and one-carbon metabolism (MTHFD2) (Fig. 5g). No single gene seemed to mediate the CD4⁺ T cell subset-specific difference in glycolytic capacity, though several transcripts (such as SLC2A3 and PFKP) were numerically most enriched in CI-Sep T_reg cells compared to CI-Sep CD4⁺ T_N and T_EM cells (Fig. 5g). Overall, CI and sepsis globally rewired CD4⁺ T cell transcriptional states toward cellular activation and functional impairments, inflammatory stress and metabolic adaptation, with the most pronounced enhancement of glycolytic features found in T_reg cells.

Sepsis status subtly alters CD4⁺ T cell transcription

Next, pseudobulked differential expression analysis of the scRNA-seq dataset revealed few genes with robust statistically significant differences between CI-Sep (n = 19 donors) and CI-NS (n = 19 donors), although immune checkpoint markers were increased in CI-Sep CD4⁺ T_EM (HAVCR2) and T_reg (LAG3) cells compared to CI-NS (Extended Data Fig. 6a and Supplementary Table 1). At single-cell level, CI-NS CD4⁺ T_EM cells had a tendency toward a T_H1 cell gene expression profile (IFNG, TBX21, EOMES and CXCR3), whereas CI-Sep CD4⁺ T_EM cells had a T_H17-biased signature (IL17A, IL22, RORC and CCR6) (Extended Data Fig. 6b). In addition, the number of IL2 transcripts in CD4⁺ T_N and T_EM cells was reduced in CI-Sep compared to CI-NS (Extended Data Fig. 6c). To better explore these subtle differences, we performed GSEA using the KEGG pathway library. The most enriched pathways in CI-Sep relative to CI-NS, including HIF-1 signaling, oxidative phosphorylation and glycolysis, were restricted to T_reg cells (Extended Data Fig. 6d and Supplementary Table 3). By contrast, numerous pathways were downregulated across all CD4⁺ T cell subsets in CI-Sep compared to CI-NS, including autoinflammatory gene sets with overlapping leading-edge genes belonging to the HLA family (Extended Data Fig. 6d,e). HLA transcripts, which have been linked to CD4⁺ T cell effector cytokine production, proliferation and survival^31,32, had a slight tendency for reduction in expression at the single-cell and sample level in CI-Sep CD4⁺ T cells compared to CI-NS CD4⁺ T cells (Extended Data Fig. 6f,g), aligning with reports of reduced HLA expression in T cells from trauma patients who subsequently develop sepsis³³. Broadly, CI-NS and CI-Sep CD4⁺ T cells generally clustered together based on their largely shared transcriptome and remained distinct from NHCs (Extended Data Fig. 6h). Together, sepsis status exerted only a modest impact on the overall CD4⁺ T cell transcriptome in CI.

Mitochondrial ROS spurs CI T_reg cell glycolytic capacity

Mitochondrial oxidative stress leads to the accumulation of mitochondrial ROS, which can promote HIF-1α and increase the expression of glycolytic genes³⁴. GSEA of the scRNA-seq dataset identified significantly increased ROS pathway activity in CD4⁺ T_N cells and T_reg cells from CI-Sep relative to NHCs (Fig. 6a and Supplementary Table 3), indicating mitochondrial ROS could be a driver of glycolytic remodeling in CI. Single-cell analysis indicated many (SOD2, TXN, MAP3K5 and ROMO1), although not all (GPX4), canonical ROS-associated genes were broadly increased in CI-Sep CD4⁺ T cells compared to NHCs (Fig. 6b). Flow cytometry showed increased mitochondrial ROS in CI-Sep CD4⁺ T cells compared to NHCs (Fig. 6c,d), whereas treatment of NHC (n = 6), CI-NS (n = 7) and CI-Sep (n = 8) PBMCs with MitoTEMPO, a mitochondrially targeted superoxide scavenger, for 24 h caused a large attenuation of the glycolytic remodeling in CI T_reg cells (pooled from CI-NS and CI-Sep) compared to vehicle-treated CI T_reg cells (mean glycolytic capacity 36% versus 49%, respectively) (Fig. 6e). A similar effect with MitoTEMPO treatment was not observed in CI CD4⁺ T_N cells, and only modestly detected in CI CD4 T_EM cells, compared to vehicle-treated CI T_N and T_EM cells, respectively (Fig. 6e). No appreciable effects of 24 h of MitoTEMPO treatment compared to vehicle treatment were observed on FOXP3 stabilization and TIGIT upregulation in CI T_reg cells or on expression of TNF in CI CD4⁺ T_conv cells in stimulated PBMCs exposed to mitochondrial impairment with low-dose oligomycin (Fig. 6f,g). These findings implicated mitochondrial ROS as a mechanistic contributor to the acquisition of glycolytic capacity but not regulation of FOXP3 or TIGIT expression in T_reg cells from patients who are CI.

**Fig. 6: Mitochondrial ROS promotes CD4⁺ T_reg cell glycolytic acquisition in CI.**

Kynurenine metabolism supports suppressive CI T_reg cells

To evaluate the factors that contributed to the stabilization of the suppressive phenotype, such as FOXP3 and TIGIT expression in CI-Sep T_reg cells, we used computational flux balance analysis (FBA). Across a curated set of 22 metabolic reaction families from the global human metabolism reconstruction (Recon2) encompassing pathways for major macronutrient metabolism and those relevant directly to T cells, CI-Sep CD4⁺ T_reg cells showed disproportionate enrichment in predicted flux through several pathways, including tryptophan metabolism, methionine and cysteine metabolism, glutathione metabolism, urea cycle and valine, leucine and isoleucine metabolism compared to CI-Sep T_N and T_EM and all NHC CD4⁺ T cell subsets (Fig. 7a and Supplementary Table 4). Tryptophan is metabolized by IDO1/2 or TDO2 into N-formyl-l-kynurenine, which is hydrolyzed to kynurenine (Fig. 7b). CD4⁺ T cells do not express IDO/TDO enzymes, but can uptake N-formyl-l-kynurenine and kynurenine through LAT1 (encoded by SLC7A5)³⁵, which contribute to de novo nicotinamide adenine dinucleotide (NAD) synthesis, acetyl-CoA production and a variety of immunomodulatory intermediates^{36,37,38,39,40}, including some that can activate the aryl hydrocarbon receptor (AHR) and/or promote T_reg cell suppressive capacity in healthy human T_reg cells^41,42. Increased plasma levels of kynurenine are associated with increased morbidity and mortality in patients with sepsis^43,44. Pairwise comparisons of FBA results showed that across CI-Sep T_reg cells versus NHC T_reg cells, CI-Sep CD4⁺ T_N cells or CI-Sep CD4⁺ T_EM cells, kynurenine-associated reactions were among the most substantially altered in the tryptophan metabolic network (Fig. 7c–e). There was a tendency for increased expression of genes for N-formyl-l-kynurenine and kynurenine import (SLC7A5), N-formyl-l-kynurenine hydrolysis (AFMID), kynurenine transamination (KYAT1 and KYAT3) and quinolinic acid catabolism for de novo NAD synthesis (QPRT) in CI-Sep T_reg cells compared to NHC T_reg cells, CI-Sep CD4⁺ T_N cells or CI-Sep CD4⁺ T_EM cells (Fig. 7f,g). Based on the intrinsic fluorescent properties of kynurenine, we measured kynurenine uptake in CD4⁺ T cell subsets by flow cytometry³⁵. LAT1-mediated kynurenine uptake was increased in all CI-Sep CD4⁺ T cell subsets, including CI-Sep T_reg cells, compared to NHC (Fig. 7h and Extended Data Fig. 7a). Consistent with this, CI-Sep T_reg cells exhibited an increased transcriptional signature of AHR activity compared to NHC T_reg cells⁴⁵ (Fig. 7i).

**Fig. 7: Kynurenine metabolism promotes CD4⁺ T_reg suppressive remodeling in CI patients.**

To determine whether kynurenine exposure alone was sufficient to augment T_reg cell glycolytic capacity, we treated NHC PBMCs for 24 h with 10 μM kynurenine, a dose within the plasma concentration range in sepsis⁴³. Exogenous kynurenine modestly but consistently increased the glycolytic capacity of NHC T_reg cells, an effect not observed in CD4⁺ T_N or T_EM cells (Fig. 7j). Conversely, short (3 h) or extended (24 h) treatment of CI-Sep PBMCs with the kynurenine-3-monooxygenase (KMO) inhibitor GSK180 did not alter the glycolytic capacity in any CD4⁺ T cell subset (Fig. 7k and Extended Data Fig. 7b,c), suggesting that central kynurenine intermediates were not contributing directly to the glycolytic capacity. This neutral impact was not due to a deficit of kynurenine in human plasma-like medium (HPLM), which we measured at near physiological levels (Extended Data Fig. 7d). In contrast to kynurenine, supplementation of NHC and CI-Sep PBMCs for 24 h with leucine, another LAT1 substrate, did not alter the glycolytic capacity of CD4⁺ T cell subsets including CD4⁺ T_reg cells, suggesting that the effect of kynurenine on T_reg cell glycolytic capacity is not a general property of all LAT1 transported amino acids (Extended Data Fig. 7e). Kynurenine metabolism can bypass KMO through anthranilic acid, which feeds further downstream reactions, including de novo NAD synthesis that could influence glycolysis (Fig. 7b). Treatment of NHC and CI-Sep PBMCs with the QPRT inhibitor phthalic acid for 24 h resulted in a modest reduction in glycolytic capacity in CI-Sep T_reg cells compared to vehicle-treated CI-Sep T_reg cells (Fig. 7l). As a control, inhibition of LDHA, a canonical enzyme in NAD/NADH recycling that sustains glycolysis, in NHC and CI-Sep PBMCs for 24 h produced a large reduction in glycolytic capacity broadly in CD4⁺ T_N, T_EM and T_reg cells from NHCs and CI-Sep (Extended Data Fig. 7f). These observations suggested that while kynurenine metabolism was altered disproportionately in CI-Sep T_reg cells compared to CI-Sep T_N and T_EM cells, it had only a minimal role in the glycolytic reprogramming of CI-Sep T_reg cells, potentially via de novo NAD generation.

Kynurenine metabolism promotes the formation of suppressive T_reg cells in healthy individuals^41,42, but direct evidence of its role in human systemic inflammatory diseases is limited. To test the role of kynurenine in sepsis, we stimulated NHC and CI-Sep PBMCs with CD3/CD28/CD2 antibodies for 24 h with low-dose (10 nM) oligomycin in the presence or absence of KMO inhibitor GSK180. Compared to vehicle-treated CI-Sep T_reg cells, treatment with GSK180 led to partial reversal of both FOXP3 stabilization (−6% with vehicle versus −10% with GSK180) and augmented TIGIT expression (20% with vehicle versus 10% with GSK180) in CI-Sep T_reg cells (Fig. 7m). No effect on FOXP3 or TIGIT expression was seen in NHC T_reg cells treated with GSK180 compared to NHC T_reg cells treated with vehicle (Fig. 7m). GSK180 further reduced the frequency of PD-1⁺ and LAG-3⁺CD4⁺ T_conv cells in both NHC and CI-Sep with GSK180 treatment compared to vehicle (Extended Data Fig. 7g), and had no effect on the frequency of CI-Sep TNF⁺, IFNγ⁺ or IL-2⁺ CD4⁺ T_conv cells, though GSK180 treatment led to a modestly greater reduction in the frequency of CI-Sep IL-17A⁺CD4⁺ T_conv cells compared to vehicle treatment (Fig. 7n and Extended Data Fig. 7h). Thus, impairment of kynurenine metabolism at the level of KMO selectively attenuated the suppressive features in CI-Sep T_reg cells without substantially compromising or exaggerating CI-Sep CD4⁺ T_conv cell effector function, highlighting a potential axis for rebalancing immune suppression and effector responses in sepsis.

We also applied FBA to CD4⁺ T cell subsets from CI-NS patients and found that the flux changes observed in CI-Sep CD4⁺ T cells were broadly conserved across the same major 22 Recon2 reaction families in CI-NS CD4⁺ T cells (Extended Data Fig. 7i and Supplementary Table 4). CI-NS T_reg cells had higher tryptophan metabolic flux than NHC T_reg cells, CI-NS CD4⁺ T_N or T_EM cells, though CI-NS T_reg cell tryptophan metabolic flux was less than CI-Sep T_reg cells (Extended Data Fig. 7j). Comparison of global effect sizes of metabolic flux across all reactions indicated a strong correlation between CI-Sep versus NHC and CI-NS versus NHC T_reg cells, with consistently larger effect sizes in CI-Sep T_reg cells (Extended Data Fig. 7k). Collectively, these data show that enhanced kynurenine metabolism promotes the stabilized suppressive phenotype of CI T_reg cells.

T_reg cell glycolytic capacity links to poor ICU metrics

Next, we analyzed the associations between T_reg cell glycolytic capacity and ICU metrics of overall disease severity, individual organ failures and clinical outcomes in pooled CI-NS (n = 30) and CI-Sep (n = 30) samples to gain adequate power. Glycolytic capacity in CI T_reg cells positively correlated with illness severity, including Sequential Organ Failure Assessment (SOFA) and Acute Physiology and Chronic Health Evaluation II (APACHE II) scores (Fig. 8a). No associations of CI T_reg cell glycolytic capacity were seen with sex or age (Fig. 8b,c). While CI T_reg cell glycolytic capacity was not correlated with peak plasma lactate, there was a trend between lower blood pH and higher T_reg cell glycolytic capacity (Fig. 8d,e). Patients with higher vasopressor requirements (with all vasopressor doses converted to a standardized norepinephrine equivalent dose), which is an indicator of more severe shock, exhibited higher T_reg cell glycolytic capacity compared to patients on lower or no vasopressors (Fig. 8f). There was no association between CI T_reg cell glycolytic capacity and severity of lung injury as measured by alveolar gas-exchange metrics (SpO₂/FiO₂ or PaO₂/FiO₂) (Fig. 8g,h).

**Fig. 8: Association of CI T_reg cell glycolytic capacity with ICU clinical metrics.**

Primary illness severity and the acquisition of secondary complications can prolong hospital length of stay or lead to in-hospital mortality. In this CI cohort, 16 individuals died during the hospitalization and 44 survived to hospital discharge. We grouped the 44 survivors into cohorts of short and long hospital lengths of stay (defined as below and above the median hospital length of stay, respectively) (Fig. 8i). Patients with a long hospital length of stay or in-hospital death had significantly higher T_reg cell glycolytic capacity than those with a short hospital length of stay (Fig. 8j), with similar effects observed when long length of stay and in-hospital death were examined individually (Fig. 8k). Together, these results linked CI T_reg cell metabolic remodeling to greater illness severity, more profound shock and acidemia, prolonged hospital length of stay and mortality in CI.

Discussion

Here, we found that CI CD4⁺ T cells underwent subset-specific metabolic rewiring characterized by persistence of T_reg cells that showed enhanced glycolytic capacity and stabilization of suppressive features and attrition of less metabolically plastic CD4⁺ T_conv cells that showed functional impairments. Parallel changes in mitochondrial ROS and kynurenine metabolism in T_reg cells promoted glycolytic remodeling and suppressive adaptations, respectively. The magnitude of T_reg cell metabolic reprogramming was associated with greater illness severity and adverse clinical outcomes.

Our findings highlighted the need to model disease-specific factors, like metabolic stress, which was evident transcriptomically with robust hypoxic and cellular stress signatures across all CI CD4⁺ T cell. The association between adaptive glycolytic remodeling in CI T_reg cells and severe shock, but not impaired gas exchange in the lungs, suggested that these stress features may be imprinted from T cell transit through inadequately perfused, dysoxic tissues rather than arising from intrinsic respiratory failure. By modeling aspects of this physiological context through partial mitochondrial impairment ex vivo, we unmasked key CI T_reg cell phenotypes, including stabilization of FOXP3 and upregulation of TIGIT.

Kynurenine metabolism promoted immunosuppressive features in CI T_reg cells. While its effects were partial (reflecting the complex physiological changes in sepsis) our findings harmonize with previous clinical studies that linked elevated plasma kynurenine to poor clinical outcomes^43,44 and extend previous findings, showing that kynurenine promotes T_reg cell programs in healthy human T cells^41,42, to human inflammatory disease. Tryptophan and kynurenine metabolism are druggable⁴⁶, and our data suggest KMO inhibition may support the restoration of immune balance by attenuating excessive T_reg cell immunosuppressive features without inducing CD4⁺ T_conv cell hyperactivation. In contrast to other candidate immunomodulatory strategies, kynurenine has both immune and nonimmune effects. Kynurenine is an arterial vasodilator⁴⁷ and kynurenine-derived metabolites are both neurotoxic⁴⁸ and tissue toxic^49,50, which may contribute to severe shock and end-organ injury, respectively. Consistent with this, high plasma kynurenine levels associate with ICU delirium⁵¹. Preclinical models support inhibition of KMO as a strategy to reduce acute kidney injury in a mouse model of ischemia/reperfusion⁵² and multiorgan injury in rodent models of acute pancreatitis^50,53. Together, these data support kynurenine metabolism among preclinical targets as a potential multifaceted therapeutic axis in CI, though further integrated studies will be necessary to prove utility.

This study has several limitations. The generalizability of these findings to other CI populations, including trauma and burn, or to less severely ill patients with infection outside of the ICU, remains unclear. Most analyses were performed at a single time point 2 days post-ICU admission. Limited blood volumes and a focus on rare cell subsets, like T_reg cells, prevented the use of certain assays like stable isotope tracing and extracellular flux analysis, which require larger cell numbers. We deferred expanding T_reg cells, as such protocols risk substantial metabolic confounding. We validated some of the transcriptomic findings experimentally (ROS and kynurenine), while others remain exploratory and hypothesis generating, particularly where effect sizes are modest and transcript abundances are low. In addition, effect sizes varied across experimental findings, reflecting both sepsis heterogeneity and its multifactorial nature, where no single pathway likely fully accounts for the observed metabolic remodeling.

Despite these constraints, our study used human biospecimens exclusively, included a diverse cohort of nonseptic ICU controls and generated a large scRNA-seq dataset from PBMCs in non-COVID CI. By focusing on ICU patients (with a median SOFA score of 9 in CI-Sep), these data complement previous large-scale human sepsis scRNA-seq studies that enrolled broader, less severely ill cohorts (for example, median SOFA 2–4 (ref. ⁵⁴) and 5 (ref. ⁵⁵)). While CI-Sep generally exhibited larger effect sizes than CI-NS, the directionality of changes was consistent between the two, suggesting overlapping mechanisms of metabolic dysfunction in CI.

Our results indicated that metabolic remodeling favored T_reg cell persistence and suppressive function (a shift broadly implicated in patient morbidity and mortality)⁵⁶. These findings provide mechanistic insight into CD4⁺ T cell dysregulation in human CI and underscore the value of context-specific, cell subset-resolved analyses. More broadly, they highlight that newly acquired metabolic capacities, not just metabolic deficiencies such as impaired mitochondrial function^19,57,58, may shape immune cell fate in CI and sepsis. These findings inform our understanding of immune dysfunction in human CI and suggest precision-based candidates for restoring immune homeostasis.

Methods

Human participants, clinical metrics and sepsis adjudication

All study protocols were approved by the Vanderbilt Institutional Review Board (IRB) (institutional protocol approval numbers 211462 and 191562). CI adults (≥18 years of age) were enrolled from the medical and surgical ICUs at Vanderbilt University Medical Center from 2022 to 2024 as part of The Sepsis ClinicAl Resource And Biorepository (SCARAB). Exclusion criteria included active exsanguination without hemostatic control or hemoglobin ≤7.5 g dl⁻¹, ICU stay >48 h or cardiac arrest before enrollment, moribund state with imminent death, receipt of extracorporeal membrane oxygenation at the time of enrollment, pregnancy, medical trainees, in police custody, ICU admission solely for frequency of nursing care, or admission for uncomplicated diabetic ketoacidosis, drug overdose or alcohol withdrawal. For this analysis, we additionally excluded individuals receiving T cell-directed immunosuppression (such as tacrolimus, azathioprine and mycophenolate), immunomodulatory antineoplastic agents (such as pembrolizumab, ipilimumab and rituximab) and those with severe leukopenia with insufficient PBMCs for analysis, active hematologic malignancy or previous allogenic stem cell transplant.

Written informed consent was obtained from patients or surrogates. If informed consent could not be obtained owing to lack of patient capacity or inability to reach a listed surrogate, then patients were enrolled under an IRB-approved waiver of consent for minimal-risk research. Nonacutely ill healthy controls were recruited from bedside surrogates of enrolled ICU patients and provided written informed consent. Healthy controls were required to be ≥18 years of age; exclusion criteria included: pregnancy, presence of an acute infection, severe anemia, receipt of T cell-directed immunosuppression or receiving any immunomodulatory antineoplastic agents, presence of a hematologic malignancy, or previous allogeneic stem cell transplant. ICU participants were not compensated; healthy controls received USD40 for participation.

Peripheral blood was collected from patients in the ICU within 24 h of ICU admission (day 0) and then serially at days 2 and 7 after study enrollment. Unless otherwise stated, all experiments were performed with day-2 samples. Approximately 7–11 ml of blood were collected at each time point in citrate and EDTA vacutainer tubes, which were pooled for PBMC isolation. Clinical data, including demographics, comorbidities, vital signs, laboratory data, imaging reports, medications and flowsheet data were abstracted from the electronic medical record into a HIPAA-compliant REDCap database. We abstracted International Classification of Disease, Clinical Modification 10 (ICD-10-CM) diagnoses codes entered into the electronic medical record during the course of each patient’s hospital admission to identify comorbidities and calculate the weighted Elixhauser–van Walraven Comorbidity Index^59,60. Healthy control participants contributed up to 50 ml of peripheral blood at a single time point; only age, sex and race/ethnicity were recorded in a HIPAA-compliant REDCap database.

Sepsis status was adjudicated post-hospitalization by a practicing physician intensivist (Extended Data Fig. 1). The presence of an infection was established based on positive bacterial cultures, infection-equivalent findings such as a visibly contaminated abdomen during laparotomy or through integrative clinical assessment in the absence of the former. End-organ dysfunction was assessed using Sepsis-3 criteria, defined as a SOFA score increase of ≥2 from the patient’s baseline². Patients meeting both infection and organ-injury criteria as a result of that infection were classified as having sepsis (CI-Sep); all others were designated nonsepsis (CI-NS).

Peripheral blood mononuclear cell isolation and storage

Fresh peripheral blood was diluted 1:2 with PBS and layered over 15 ml of Lymphoprep density gradient medium (STEMCELL Tech., cat. no. 18061) in SepMate-50 tubes (STEMCELL Tech., cat. no.85450). After density gradient centrifugation (1,200g for 15 min at 20 °C), the PBMC-containing upper layer was pelleted (400g for 10 min at 20 °C), RBC-lysed for 7 min with ACK buffer (Gibco, cat. no. A1049201), washed (300g for 10 min at 20 °C), filtered through a 70-μm strainer and washed a final time (300g for 10 min at 20 °C). Isolated PBMCs were resuspended in Bambanker Cell Freezing Medium (GC LYMPHOTEC, cat. no. BB02) at 2 million PBMCs per ml, aliquoted and slow frozen in a CoolCell FTS30 freezing container (Corning, cat. no. 432006) at −80 °C. PBMC aliquots were transferred to liquid nitrogen vapor phase storage within 72 h.

Cryorecovery of PBMCs

PBMCs were thawed by rapid rewarming at 37 °C for 3 min and immediately diluted 1:5 in HPLM (Gibco, cat. no. A4899101) supplemented with 10% dialyzed fetal bovine serum (dFBS) (Sigma-Aldrich, cat. no. F0392) and 100 U ml⁻¹ of penicillin plus 100 μg ml⁻¹ of streptomycin (P/S) (Gibco, cat. no. 15140122). Cells were centrifuged at 300g for 5 min at 20 °C, washed twice in the same medium and filtered through a 35-μm cell strainer cap during the final wash to remove debris. PBMCs were resuspended in HPLM with 10% dFBS and P/S for immediate use in downstream experiments.

Flow cytometry

Cells were distributed in 96-well V-bottom microplates (Corning, cat. no. 3894), pelleted (300g for 5 min at 4 °C) and washed twice in cold PBS. All cell staining was conducted in 50-μl volumes. Cells were stained with Ghose Dye Red 780 Fixable Viability Dye (Cell Signaling, cat. no. 18452, diluted 1:2,000) and Human TruStain FcX solution for Fc receptor blocking (BioLegend, cat. no. 422302, diluted 1:40) in PBS for 30 min at 4 °C. Cells were washed with flow staining buffer (cold PBS with 2% FBS) and stained for surface markers (antibodies listed in Extended Data Table 2) in flow staining buffer for 30 min at 4 °C followed by two additional washes. In some experiments, intracellular markers were evaluated. Cells were fixed and permeabilized using 100 μl of Fix/Perm reagent (Tonbo Biosciences, cat. no. TNB-0607-KIT) for 30 min at room temperature followed by two washes in permeabilization buffer (Tonbo Biosciences, cat. no. TNB-1213-L150). Cells were stained for intracellular markers (antibodies in Extended Data Table 2) in permeabilization buffer for 30 min at 4 °C followed by two washes in permeabilization buffer. Data were acquired on a Miltenyi MACSQuant Analyzer 16 flow cytometer and exported as MQD files, which were analyzed in FlowJo v.10.9.0 (Becton Dickson & Company). Representative gating is shown (Extended Data Fig. 8).

Puromycin-incorporation assay

We measured metabolic dependencies and capacities using a flow cytometric puromycin-incorporation assay adapted from the SCENITH method⁶¹. After cryothaw, PBMCs were rested for 3 h cells at 37 °C in HPLM with 10% dFBS and P/S to allow for metabolic recovery. For each sample, single metabolic inhibitors were added in separate wells for 30 min at 37 °C: glucose-dependent pathway inhibitor 2-deoxy-d-glucose (2-DG) (Cayman Chemical, cat. no. 14325, 100 mM), mitochondrial ATP synthase inhibitor oligomycin A (Cayman Chemical, cat. no. 11342, 1.5 μM), glutaminase 1 (GLS1) glutaminolysis inhibitor CB-839 (MedChemExpress, cat. no. HY-12248, 10 μM) or carnitine palmitoyltransferase-1 (CPT-1) fatty acid oxidation inhibitor Etomoxir (Sigma-Aldrich, cat. no. E1905, 5 μM). Control wells with vehicle or a full inhibitor cocktail (ATP extinction control) were included. Cells were pulsed with puromycin (Sigma-Aldrich, cat. no. P8833, 10 μM) for 40 min at 37 °C, washed in cold PBS and prepared for flow cytometry with intracellular detection of puromycin by antibody staining as described above. Metabolic dependencies and capacities were calculated from the geometric MFI of puromycin as originally reported⁶¹. Where indicated, cells were pretreated for 3 h or 24 h at 37 °C in HPLM with 10% dFBS and P/S with l-kynurenine (MedChemExpress, cat. no. HY-104026, 10 μM), l-leucine (MedChemExpress, cat. no. HY-N0486500, 500 μM), GSK180 (MedChemExpress, cat. no. HY-112179, 10 μM), phthalic acid (MedChemExpress, cat. no. HY-I0508, 100 μM), FX-11 (Millipore-Sigma, cat. no. 427218, 20 μM) or MitoTEMPO (MedChemExpress, cat. no. HY-112879, 10 μM) before the puromycin-incorporation assay.

Mitochondrial mass, membrane potential and ROS

Mitochondrial mass, mitochondrial membrane potential and mitochondrial ROS were measured using MitoTracker Green FM (Invitrogen, cat. no. M46750), TMRE (Invitrogen, cat. no. T669) and MitoSOX Green superoxide indicator (Invitrogen, cat. no. M36006), respectively. Cryorecovered PBMCs were resuspended in HPLM with 10% dFBS and P/S, rested for 30 min at 37 °C and treated with either MitoTracker Green FM (25 nM), TMRE (40 nM), or MitoSOX Green (500 nM) for 30 min at 37 °C. Cells were washed twice with cold flow staining buffer and prepared for flow cytometry as described above.

Ex vivo restimulation for CD4⁺ T cell function

Cryorecovered PBMCs were resuspended in HPLM with 10% dFBS and P/S and stimulated with ImmunoCult Human CD3/CD28/CD2 T Cell Activator (STEMCELL Tech., cat. no. 10970, 1:20 dilution) for 24 h at 37 °C. Brefeldin A (GolgiPlug, BD, cat. no. 555029, 1:1,000 dilution) was added for the final 5 h to enable intracellular cytokine detection. Cells were analyzed by flow cytometry as described above. Where indicated, cells were treated with low-dose oligomycin (Cayman Chemical, cat. no. 11342, 10 nM), GSK180 (MedChemExpress, cat. no. HY-112179, 10 μM) or MitoTEMPO (MedChemExpress, cat. no. HY-112879, 10 μM) for the full 24 h stimulation.

Kynurenine uptake assay

Intracellular kynurenine uptake was measured as previously described³⁵ with modifications. Cryorecovered PBMCs were incubated in HPLM with 10% dFBS and P/S for 1 h at 37 °C. Cells were washed twice with warm PBS and stained with Fc-block, viability dye and cell surface antibodies in PBS for 30 min at room temperature. Cells were then washed once with prewarmed 37 °C HBSS and resuspended in 37 °C HBSS containing l-kynurenine (MedChemExpress, cat. no. HY-104026, 100 μM) for 5 min. Uptake was quenched with 100 μl IC Fixation Buffer (Invitrogen, cat. no. 00-8333) for 30 min at room temperature, washed twice with flow staining buffer and analyzed by flow cytometry. l-kynurenine was detected by its intrinsic spectral properties using a 405-nm excitation (violet laser) and a 450/50-nm emission filter (V1 channel, MACSQuant Analyzer 16). Where indicated, LAT1 transport inhibitor BCH (MedChemExpress, cat. no. HY-108540, 10 mM) was added at time of cell plating or a saturating quantity of LAT1 substrate l-leucine (MedChemExpress, cat. no. HY-N0486, 5 mM) was added at the time of l-kynurenine treatment as controls for receptor specificity. Where indicated, HBSS with l-kynurenine chilled to 4 °C was added to cells on ice for the 5-min pulse to discriminate l-kynurenine uptake versus cell surface binding.

LC–MS/MS quantification of kynurenine in cell culture medium

To determine the pre-existing availability of l-kynurenine in the media preparations used in experiments where modulations included adding exogenous l-kynurenine or a kynurenine-3-monooxygenase inhibitor (KMOi), we performed quantitative liquid chromatography with tandem mass spectrometry (LC–MS/MS). l-kynurenine levels were tested in fresh HPLM, fresh HPLM plus 10% dFBS and P/S, and HPLM plus 10% dFBS and P/S stored for 30 days at 4 °C. Then, 1 ml of −80 °C 80:20 methanol:water was added to the 330 µl of the medium sample in a 1.7-ml tube and vortexed vigorously. Then, 20 nmol of internal standard (¹³C-1-Lactate, Millipore-Sigma) was spiked into each sample. Cells were then placed in −80 °C freezer to extract for 15 min. After extraction, any insoluble debris was pelleted at 16,000g for 10 min at 4 °C. The supernatant containing extracted metabolites was then transferred to a new 1.7-ml tube. Extracts were then dried under N₂. Dried samples were resuspended in 50 µl 3:2 mobile buffer A: mobile buffer B (see below) and transferred to 1.7-ml tubes. Samples were then centrifuged at 16,000g for 10 min at 4 °C to remove any insoluble debris. A standard curve was generated using 20 nmol of internal standard (¹³C-1-Lactate) in samples containing either 0, 3.75, 7.5, 15 or 30 nmol of a natural isotope kynurenine standard. Then, 18 µl of the sample was then chromatographed with a Shimadzu LC system equipped with a 100 × 2.1 mm, 3.5-μm particle diameter XBridge Amide column (Waters). Mobile phase A: 20 mM NH₄OAc, 20 mM NH₄OH and 5% acetonitrile in H₂O, pH 9.45 (pH with NH₄OH); mobile phase B: 100% acetonitrile. With a flow rate of 0.45 ml min⁻¹ the following gradient was used: 2.0 min, 95% B; 3.0 min, 85% B; 5.0 min, 85% B; 6.0 min, 80% B; 8.0 min, 80% B; 9.0 min, 75% B; 10 min, 75% B; 11 min, 70% B; 12 min, 70% B; 13 min, 50% B; 15 min, 50% B; 16 min 0% B; 17.5 min, 0% B; and 18 min, 95% B. The column was equilibrated for 3 min at 95% B between each sample. Scheduled multiple reaction monitoring (MRM) was conducted in negative and positive mode with a detection window of 120 s using an AB SCIEX 6500 QTRAP with the following analyte parameters: m/z 209 → 94 (positive mode, retention time: 6.3 min) for l-kynurenine; m/z 90 → 43 (negative mode, retention time of 6.0 min) for ¹³C-1-lactate internal standard. All analytes were quantified via LC–MS/MS using the standard curve.

Single-cell library preparation and RNA sequencing

Cryopreserved PBMCs were rapidly thawed at 37 °C for 3 min, diluted dropwise into 7 ml of room temperature HPLM with 10% dFBS and centrifuged at 300g for 5 min. Cells were washed once with HPLM with 10% dFBS and once in cold PBS with 0.04% bovine serum albumin (BSA), with the final wash passed through a 35-μm cell strainer (Falcon, cat. no. 352235). Cells were resuspended in cold PBS with 0.04% BSA at 1,000 cells per μl. Single-cell RNA-seq libraries were generated using the Chromium Next GEM Single Cell 5’ HT Kit v2 (10x Genomics) targeting 20,000 cells per sample. Libraries were sequenced on an Illumina NovaSeq 6000 (S4 flow cell, PE150) to a target depth of 50,000 reads per cell. Raw data were processed with Cell Ranger (v.7.1.0) using the GRCh38-2020-A genome assembly with the intron inclusion option enabled.

ScRNA-seq preprocessing, quality control, integration and modeling

Raw feature barcode matrices from Cell Ranger were processed through a custom Nextflow workflow executed within Docker containers on AWS HealthOmics. Ambient RNA and technical artifacts were removed using CellBender remove-background (v.0.3.0)⁶². Learning rates (0.0000005–0.00005) and training epochs (150–300) were iteratively tuned; defaults were used for all other parameters.

CellBender outputs were filtered using Seurat (v.4.3.0)⁶³ to exclude cells with >20% mitochondrial content or <400 or >5,000 detected features. Doublets were identified and removed using scDblFinder (v.1.13.10)⁶⁴ using temporary SCTransform normalization and shared nearest neighbor clustering (2,000 highly variable genes (HVGs), 30 principal components and resolution 0.5), which were discarded after doublet removal. Sample-level metadata were added, and cleaned matrices were converted to AnnData using the sceasy package (v.0.0.7)⁶⁵.

Filtered per-sample data were merged, normalized (target_sum = 1 × 10⁴) and log-transformed using Scanpy (v.1.9.5)⁶⁶. The top 5,000 HVGs were selected using the ‘seurat_v3’ method. Data were integrated and modeled using scVI-tools (v.1.0.4)⁶⁷. The scVI model was trained using n_hidden = 128, n_latent = 30, n_layers = 2, negative binomial likelihood (gene_likelihood = ‘nb’) with gene level dispersion, dropout rate of 0.1, AdamW optimizer and a learning rate of 0.001. Training proceeded for up to 400 epochs with early stopping. Modeling incorporated categorical covariates (Biologic_Sex), continuous covariates (percent.mt, nCount_RNA, and percent.ribosomal) and batch correction (batch_key).

scRNA-seq clustering, visualization and annotation

Scanpy (v.1.9.5) was used to create a k-nearest neighbors (kNN) graph (scanpy.pp.neighbors) using the scVI latent space (use_rep = ‘X_scVI’, n_neighbors = 30). Clustering was performed with the Leiden algorithm (scanpy.tl.leiden, resolution = 1). DEGs for each cluster were identified using scVI’s differential_expression function with denoised expression values. Cell types were annotated using canonical marker genes with the Azimuth Human PBMC core reference map as the primary guide⁶⁸, supplemented by literature-defined markers and expert consensus. Clusters lacking clear marker dominance were iteratively subclustered (resolutions between 0.2 and 2.0) until high-confidence annotations could be assigned. Clusters exhibiting two distinct, nonoverlapping gene expression profiles were excluded as potential doublets. A small B cell cluster unique to sample 10069-MS-0033 expressing canonical markers of chronic lymphocytic leukemia (ROR1, FMOD and CD5) and identical heavy- and light-chain V-segments (IGHV3-66 and IGKV1D-39) was excluded as a probable subclinical neoplasm. Final annotations were provided at two levels: celltype_coarse for major lineages and celltype_fine for granular subpopulations.

Cell proportionality

Cell lineage proportions were quantified using Scanpro (v.0.2)⁶⁹ applied to NHC, CI-NS and CI-Sep samples with both celltype_coarse and celltype_fine cell lineage annotations.

Differential gene expression and pathway analysis

Pseudobulk profiles were generated from single-cell data using the Python implementation of the decoupleR package (v.1.6.0)⁷⁰ by summing counts per cell type across within each sample, retaining profiles from ≥20 cells and with ≥1,000 total counts. Differential expression was performed using PyDESeq2 (v.1.40.2)⁷¹ with low-expression genes removed (filter_by_expr function, min_count=3, min_total_count = 25). Visualization of per-sample gene expression was performed using pseudobulk counts normalized to CPM and log-transformed. Visualization of single-cell gene expression was performed using Scanpy-normalized and log-transformed single-cell counts. Normalized, scaled pseudobulk gene expression profiles were used to select the top 2,000 HVGs (Seurat v3 method) followed by hierarchical clustering using Ward’s linkage based on Euclidean distances.

Over-representation analysis of significantly upregulated genes (FDR-adjusted P < 0.05) was performed with decoupleR using the GO:BP collection. GSEA⁷² was performed on pseudobulk differential gene expression results (Wald statistic) using GSEApy (v.1.1.4)⁷³ with 1,000 permutations and the KEGG 2021 Human and Hallmark 2020 libraries, using a fixed random seed. NES and FDR-adjusted P values were used to score pathways. A binary matrix indicating leading-edge gene membership across significantly enriched gene sets (FDR-adjusted P < 0.05) was constructed and pairwise Jaccard similarity between pathways was calculated based on shared leading-edge genes. Pathway activity for a human pan-tissue AHR gene signature⁴⁵ was quantified using the AUCell method⁷⁴ implemented in decoupleR.

Flux balance analysis

FBA was conducted using the Compass package (v.0.9.10.2)⁷⁵. For each condition and cell subset, single-cell transcriptomes were normalized to CPM (scanpy.pp.normalize_total) and aggregated into tab-delimited input files. Compass was run with the IBM CPLEX optimization engine using the Recon2 consensus metabolic reconstruction⁷⁶ and default parameters (penalty-diffusion knn, lambda 0, num-neighbors 30) with microclusters of 25 cells. Reaction penalty scores were negative log-transformed to yield values representing reaction flux potential and filtered to remove low-confidence or invariant reaction. For bidirectional reactions, all reactions were modeled but only positive fluxes were retained for visualization. Reaction activity was standardized as z-scores or calculated as Cohen’s d effect sizes for visualization and comparison between conditions.

Statistics and reproducibility

Statistical analyses were performed in Python (v.3.9) using scipy.stats (v.1.11.4), scikit-posthocs (v.0.9.0) and statsmodels (v.0.14.1). Two-sided Mann–Whitney U-tests were used for two-group comparisons and Kruskal–Wallis tests with Dunn’s post hoc testing for multigroup comparisons. Spearman correlations were used for associations with LOWESS smoothing (fraction = 0.8) with 1,000 bootstrap resamples applied where indicated. Patient summary statistics (median, IQR and P values) were generated using tableone (v.0.9.1).

For scRNA-seq analyses, logit-transformed proportion testing used an empirical Bayes ANOVA (scanpro); differential expression used the Wald test with Benjamini–Hochberg correction (DESeq2); over-representation analysis used a one-tailed Fisher’s exact test with Benjamini–Hochberg correction; gene-set enrichment used a permutation-based Kolmogorov–Smirnov statistic with Benjamini–Hochberg correction (pyGSEA); and Compass flux comparisons used two-sided Mann–Whitney U-tests with Benjamini–Hochberg correction. Effect sizes were represented as Cohen’s d values derived from the pooled s.d. Bar graphs show means with error bars representing s.e.m. Boxplots follow standard conventions (median line, IQR box boundaries and whiskers extending to the minimum and maximum values within 1.5 × IQR).

No statistical method was used to predetermine sample size but our sample sizes are similar to those reported in previous publications^54,55. For flow cytometric assays reporting mean intensity values, including the puromycin-incorporation assay, cell populations with <25 detected cells were excluded because a mean could not reliably be estimated. One scRNA-seq sample (10069-MS-0036) was excluded before analysis after grossly failing quality control metrics. Some clinical variables were unavailable for specific patients in the electronic medical record and were therefore not included. The investigators were not blind to the sample group assignments while completing the analysis. Most data were generated from ≥2 independent experimental replicates and shown as pooled data, with limited analyses performed in a single replicate (Figs. 2a,c,d, 6f,g and 7k,l and Extended Data Figs. 3b and 7b–f).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

Single-cell RNA-seq data supporting this study have been deposited in the Gene Expression Omnibus (GEO) under accession no. GSE290679 with FASTQ files available in the Database of Genotypes and Phenotypes under accession nos. phs004258.v1.p1 (ICU patients) and phs004377.v1.p1 (healthy donors). Source data are provided with the paper, except certain clinical data that could be potentially uniquely identifying. Further information and requests for data should be directed to, and will be fulfilled by, the corresponding authors J.C.R. and M.T.S. Source data are provided with this paper.

Code availability

Scripts associated with the data analysis in this manuscript have been deposited at https://github.com/StierLab/2025-CD4T-sepsis.

References

Barrett, M. L., Smith, M. W., Elixhauser, A., Honigman, L. S. & Pines, J. M. Utilization of Intensive Care Services, 2011, HCUP Statistical Brief #185 (Agency for Healthcare Research and Quality, 2014).
Singer, M. et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA 315, 801–810 (2016).
Article PubMed PubMed Central CAS Google Scholar
Rudd, K. E. et al. Global, regional, and national sepsis incidence and mortality, 1990–2017: analysis for the Global Burden of Disease Study. Lancet 395, 200–211 (2020).
Article PubMed PubMed Central Google Scholar
Torio, C. M. & Moore, B. J. National Inpatient Hospital Costs: The Most Expensive Conditions by Payer, 2013. HCUP Statistical Brief #204 (HCUP, 2016).
Marshall, J. C. & Leligdowicz, A. Gaps and opportunities in sepsis translational research. eBioMedicine 86, 104387 (2022).
Article PubMed PubMed Central CAS Google Scholar
Wiersinga, W. J. & Poll, T. van der. Immunopathophysiology of human sepsis. eBioMedicine 86, 104363 (2022).
Article PubMed PubMed Central CAS Google Scholar
Torres, L. K., Pickkers, P. & Poll, T. Sepsis-induced immunosuppression. Annu. Rev. Physiol. 84, 1–25 (2021).
Google Scholar
Jones, T. K. et al. Post-acute care use and hospital readmission after sepsis. Ann. Am. Thorac. Soc. 12, 904–913 (2015).
Article PubMed PubMed Central Google Scholar
Davenport, E. E. et al. Genomic landscape of the individual host response and outcomes in sepsis: a prospective cohort study. Lancet Respir. Med. 4, 259–271 (2016).
Article PubMed PubMed Central Google Scholar
Venet, F. et al. Increased percentage of CD4⁺CD25⁺ regulatory T cells during septic shock is due to the decrease of CD4⁺CD25⁻ lymphocytes. Crit. Care Med. 32, 2329–2331 (2004).
Article PubMed Google Scholar
Niu, B. et al. Different expression characteristics of LAG3 and PD-1 in Sepsis and their synergistic effect on T cell exhaustion: a new strategy for immune checkpoint blockade. Front. Immunol. 10, 1888 (2019).
Article PubMed PubMed Central CAS Google Scholar
Guignant, C. et al. Programmed death-1 levels correlate with increased mortality, nosocomial infection and immune dysfunctions in septic shock patients. Crit. Care 15, R99 (2011).
Article PubMed PubMed Central Google Scholar
Ong, D. S. Y. et al. Epidemiology of multiple herpes viremia in previously immunocompetent patients with septic shock. Clin. Infect. Dis. 64, 1204–1210 (2017).
Article PubMed CAS Google Scholar
Styczynski, J. Who is the patient at risk of CMV recurrence: a review of the current scientific evidence with a focus on hematopoietic cell transplantation. Infect. Dis. Ther. 7, 1–16 (2018).
Article PubMed Google Scholar
Willmann, K. & Moita, L. F. Physiologic disruption and metabolic reprogramming in infection and sepsis. Cell Metab 36, 927–946 (2024).
Article PubMed CAS Google Scholar
Makowski, L., Chaib, M. & Rathmell, J. C. Immunometabolism: from basic mechanisms to translation. Immunol. Rev. 295, 5–14 (2020).
Article PubMed PubMed Central CAS Google Scholar
Karagiannis, F. et al. Impaired ketogenesis ties metabolism to T cell dysfunction in COVID-19. Nature 609, 801–807 (2022).
Article PubMed PubMed Central CAS Google Scholar
Stine, Z. E., Schug, Z. T., Salvino, J. M. & Dang, C. V. Targeting cancer metabolism in the era of precision oncology. Nat. Rev. Drug Discov. 21, 141–162 (2022).
Article PubMed CAS Google Scholar
Venet, F. et al. IL-7 restores T lymphocyte immunometabolic failure in septic shock patients through mTOR activation. J. Immunol. 199, 1606–1615 (2017).
Article PubMed CAS Google Scholar
Merz, T. M. et al. Mitochondrial function of immune cells in septic shock: a prospective observational cohort study. PLoS ONE 12, e0178946 (2017).
Article PubMed PubMed Central Google Scholar
Agnello, L. et al. The value of a complete blood count (CBC) for sepsis diagnosis and prognosis. Diagnostics 11, 1881 (2021).
Article PubMed PubMed Central CAS Google Scholar
Kamada, T. et al. PD-1⁺ regulatory T cells amplified by PD-1 blockade promote hyperprogression of cancer. Proc. Natl Acad. Sci. USA 116, 9999–10008 (2019).
Article PubMed PubMed Central CAS Google Scholar
Camisaschi, C. et al. LAG-3 expression defines a subset of CD4⁺CD25^highFoxp3⁺ regulatory T cells that are expanded at tumor sites. J. Immunol. 184, 6545–6551 (2010).
Article PubMed CAS Google Scholar
Borsellino, G. et al. Expression of ectonucleotidase CD39 by Foxp3⁺ Treg cells: hydrolysis of extracellular ATP and immune suppression. Blood 110, 1225–1232 (2007).
Article PubMed CAS Google Scholar
Joller, N. et al. Treg cells expressing the coinhibitory molecule TIGIT selectively inhibit proinflammatory Th1 and Th17 cell responses. Immunity 40, 569–581 (2014).
Article PubMed PubMed Central CAS Google Scholar
Bonelli, M. et al. Phenotypic and functional analysis of CD4⁺CD25⁻Foxp3⁺ T cells in patients with systemic lupus erythematosus. J. Immunol. 182, 1689–1695 (2009).
Article PubMed CAS Google Scholar
Slack, M., Wang, T. & Wang, R. T cell metabolic reprogramming and plasticity. Mol. Immunol. 68, 507–512 (2015).
Article PubMed PubMed Central CAS Google Scholar
Wan, Y. Y. & Flavell, R. A. Regulatory T-cell functions are subverted and converted owing to attenuated Foxp3 expression. Nature 445, 766–770 (2007).
Article PubMed CAS Google Scholar
Yosri, M., Dokhan, M., Aboagye, E., Moussawy, M. A. & Abdelsamed, H. A. Mechanisms governing bystander activation of T cells. Front. Immunol. 15, 1465889 (2024).
Article PubMed PubMed Central CAS Google Scholar
Lee, H., Jeong, S. & Shin, E.-C. Significance of bystander T cell activation in microbial infection. Nat. Immunol. 23, 13–22 (2022).
Article PubMed CAS Google Scholar
Takao, S., Ishikawa, T., Yamashita, K. & Uchiyama, T. The rapid induction of HLA-E is essential for the survival of antigen-activated naive CD4 T cells from attack by NK cells. J. Immunol. 185, 6031–6040 (2010).
Article PubMed CAS Google Scholar
Tippalagama, R. et al. HLA-DR marks recently divided antigen-specific effector CD4 T cells in active tuberculosis patients. J. Immunol. 207, 523–533 (2021).
Article PubMed CAS Google Scholar
Ditschkowski, M. et al. HLA-DR expression and soluble HLA-DR levels in septic patients after trauma. Ann. Surg. 229, 246–254 (1999).
Article PubMed PubMed Central CAS Google Scholar
Movafagh, S., Crook, S. & Vo, K. Regulation of hypoxia-inducible factor-1a by reactive oxygen species: new developments in an old debate. J. Cell. Biochem. 116, 696–703 (2015).
Article PubMed CAS Google Scholar
Sinclair, L. V., Neyens, D., Ramsay, G., Taylor, P. M. & Cantrell, D. A. Single cell analysis of kynurenine and system L amino acid transport in T cells. Nat. Commun. 9, 1981 (2018).
Article PubMed PubMed Central Google Scholar
Stone, T. W. & Williams, R. O. Modulation of T cells by tryptophan metabolites in the kynurenine pathway. Trends Pharmacol. Sci. 44, 442–456 (2023).
Article PubMed CAS Google Scholar
Michaudel, C. et al. Rewiring the altered tryptophan metabolism as a novel therapeutic strategy in inflammatory bowel diseases. Gut 72, 1296–1307 (2023).
Article PubMed CAS Google Scholar
Badawy, A. A.-B. & Dawood, S. Molecular insights into the interaction of tryptophan metabolites with the human aryl hydrocarbon receptor in silico: tryptophan as antagonist and no direct involvement of kynurenine. Front. Biosci. Landmark 29, 333 (2024).
Article CAS Google Scholar
Seok, S.-H. et al. Trace derivatives of kynurenine potently activate the aryl hydrocarbon receptor (AHR). J. Biol. Chem. 293, 1994–2005 (2018).
Article PubMed CAS Google Scholar
Carreño, M. et al. Immunomodulatory actions of a kynurenine-derived endogenous electrophile. Sci. Adv. 8, eabm9138 (2022).
Article PubMed PubMed Central Google Scholar
Hippen, K. L. et al. In vitro induction of human regulatory T cells using conditions of low tryptophan plus kynurenines. Am. J. Transplant. 17, 3098–3113 (2017).
Article PubMed PubMed Central CAS Google Scholar
Gandhi, R. et al. Activation of the aryl hydrocarbon receptor induces human type 1 regulatory T cell-like and Foxp3⁺ regulatory T cells. Nat. Immunol. 11, 846–853 (2010).
Article PubMed PubMed Central CAS Google Scholar
Darcy, C. J. et al. An observational cohort study of the kynurenine to tryptophan ratio in sepsis: association with impaired immune and microvascular function. PLoS ONE 6, e21185 (2011).
Article PubMed PubMed Central CAS Google Scholar
Ferrario, M. et al. Mortality prediction in patients with severe septic shock: a pilot study using a target metabolomics approach. Sci. Rep. 6, 20391 (2016).
Article PubMed PubMed Central CAS Google Scholar
Sadik, A. et al. IL4I1 is a metabolic immune checkpoint that activates the ahr and promotes tumor progression. Cell 182, 1252–1270.e34 (2020).
Article PubMed CAS Google Scholar
Modoux, M., Rolhion, N., Mani, S. & Sokol, H. Tryptophan metabolism as a pharmacological target. Trends Pharmacol. Sci. 42, 60–73 (2021).
Article PubMed CAS Google Scholar
Wang, Y. et al. Kynurenine is an endothelium-derived relaxing factor produced during inflammation. Nat. Med. 16, 279–285 (2010).
Article PubMed PubMed Central CAS Google Scholar
Zhang, S., Collier, M. E. W., Heyes, D. J., Giorgini, F. & Scrutton, N. S. Advantages of brain penetrating inhibitors of kynurenine-3-monooxygenase for treatment of neurodegenerative diseases. Arch. Biochem. Biophys. 697, 108702 (2021).
Article PubMed CAS Google Scholar
Wang, Q. et al. Activation of NAD(P)H oxidase by tryptophan-derived 3-hydroxykynurenine accelerates endothelial apoptosis and dysfunction in vivo. Circ. Res. 114, 480–492 (2014).
Article PubMed CAS Google Scholar
Hayes, A. J. et al. Kynurenine monooxygenase regulates inflammation during critical illness and recovery in experimental acute pancreatitis. Cell Rep 42, 112763 (2023).
Article PubMed CAS Google Scholar
Wilson, J. R. A. et al. The association of the kynurenine pathway of tryptophan metabolism with acute brain dysfunction during critical illness. Crit. Care Med. 40, 835–841 (2012).
Article Google Scholar
Zheng, X. et al. Kynurenine 3-monooxygenase is a critical regulator of renal ischemia–reperfusion injury. Exp. Mol. Med. 51, 1–14 (2019).
Article PubMed PubMed Central CAS Google Scholar
Mole, D. J. et al. Kynurenine-3-monooxygenase inhibition prevents multiple organ failure in rodent models of acute pancreatitis. Nat. Med. 22, 202–209 (2016).
Article PubMed PubMed Central CAS Google Scholar
Reyes, M. et al. An immune-cell signature of bacterial sepsis. Nat. Med. 26, 333–340 (2020).
Article PubMed PubMed Central CAS Google Scholar
Kwok, A. J. et al. Neutrophils and emergency granulopoiesis drive immune suppression and an extreme response endotype during sepsis. Nat. Immunol. 24, 767–779 (2023).
Article PubMed CAS Google Scholar
Huang, S. et al. Decoding the potential role of regulatory T cells in sepsis-induced immunosuppression. Eur. J. Immunol. 54, e2350730 (2024).
Article PubMed Google Scholar
Cheng, S.-C. et al. Broad defects in the energy metabolism of leukocytes underlie immunoparalysis in sepsis. Nat. Immunol. 17, 406–413 (2016).
Article PubMed CAS Google Scholar
McBride, M. A. et al. The metabolic basis of immune dysfunction following sepsis and trauma. Front. Immunol. 11, 1043 (2020).
Article PubMed PubMed Central CAS Google Scholar
Quan, H. et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med. Care 43, 1130–1139 (2005).
Article PubMed Google Scholar
Walraven et al. A Modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. Med. Care 47, 626–633 (2009).
Article PubMed Google Scholar
Argüello, R. J. et al. SCENITH: a flow cytometry-based method to functionally profile energy metabolism with single-cell resolution. Cell Metab. 32, 1063–1075.e7 (2020).
Article PubMed PubMed Central Google Scholar
Fleming, S. J. et al. Unsupervised removal of systematic background noise from droplet-based single-cell experiments using CellBender. Nat. Methods 20, 1323–1335 (2023).
Article PubMed CAS Google Scholar
Satija, R., Farrell, J. A., Gennert, D., Schier, A. F. & Regev, A. Spatial reconstruction of single-cell gene expression data. Nat. Biotechnol. 33, 495–502 (2015).
Article PubMed PubMed Central CAS Google Scholar
Germain, P.-L., Lun, A., Meixide, C. G., Macnair, W. & Robinson, M. D. Doublet identification in single-cell sequencing data using scDblFinder. F1000Research 10, 979 (2021).
Article PubMed Google Scholar
Cakir, B. et al. Comparison of visualization tools for single-cell RNAseq data. NAR Genom. Bioinform. 2, lqaa052 (2020).
Article PubMed PubMed Central Google Scholar
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Article PubMed PubMed Central Google Scholar
Lopez, R., Regier, J., Cole, M. B., Jordan, M. I. & Yosef, N. Deep generative modeling for single-cell transcriptomics. Nat. Methods 15, 1053–1058 (2018).
Article PubMed PubMed Central CAS Google Scholar
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587.e29 (2021).
Article PubMed PubMed Central CAS Google Scholar
Alayoubi, Y., Bentsen, M. & Looso, M. Scanpro is a tool for robust proportion analysis of single-cell resolution data. Sci. Rep. 14, 15581 (2024).
Article PubMed PubMed Central CAS Google Scholar
Badia-i-Mompel, P. et al. decoupleR: ensemble of computational methods to infer biological activities from omics data. Bioinform. Adv. 2, vbac016 (2022).
Article PubMed PubMed Central Google Scholar
Muzellec, B., Teleńczuk, M., Cabeli, V. & Andreux, M. PyDESeq2: a Python package for bulk RNA-seq differential expression analysis. Bioinformatics 39, btad547 (2023).
Article PubMed PubMed Central CAS Google Scholar
Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).
Article PubMed PubMed Central CAS Google Scholar
Fang, Z., Liu, X. & Peltz, G. GSEApy: a comprehensive package for performing gene set enrichment analysis in Python. Bioinformatics 39, btac757 (2022).
Article PubMed Central Google Scholar
Aibar, S. et al. SCENIC: single-cell regulatory network inference and clustering. Nat. Methods 14, 1083–1086 (2017).
Article PubMed PubMed Central CAS Google Scholar
Wagner, A. et al. Metabolic modeling of single Th17 cells reveals regulators of autoimmunity. Cell 184, 4168–4185.e21 (2021).
Article PubMed PubMed Central CAS Google Scholar
Thiele, I. et al. A community-driven global reconstruction of human metabolism. Nat. Biotechnol. 31, 419–425 (2013).
Article PubMed CAS Google Scholar

Download references

Acknowledgements

We thank the patients for their participation and invaluable donations of the human tissues used in this study. We are grateful for the contributions of all clinical staff that interfaced with our research team to support this work. We appreciate the helpful discussions with S. Peebles, T. Blackwell, A. Hemnes, H. Serezani, J. Kropski, D. Mogilenko and the members of the laboratories of J. Rathmell, J. Bastarache and L. Ware. We are grateful for the assistance of A. Jones and staff in the VANTAGE core. This work was supported by the National Institutes of Health (NIH) R01 DK105550 (J.C.R.), R01 HL136664 (J.C.R. and D.C.N.), R01 HL118979 (J.C.R.), R01 AI153167 (J.C.R.), T32 HL094296 (M.T.S.), T32 CA009582 (E.Q.J.), T32 AR059039 (P.L.), R33 GM144915 (J.A.B. and L.B.W.), HL150783 (J.A.B.), AG075341 (J.A.B.), R01 HL158906 (L.B.W.), R01 HL164937 (L.B.W.), K08 AR080808 (E.M.W), KL2 TR002245 (E.M.W.), T32 GM007347 (C.M.N.), T32 GM152284 (C.M.N.) and the Vanderbilt Center for Immunobiology Human Immunology Discovery Initiative. The Vanderbilt Technologies for Advanced Genomics Core (VANTAGE) is supported by the Vanderbilt CTSA (5UL1 RR024975-03), the Vanderbilt Ingram Cancer Center (P30 CA68485), the Vanderbilt Vision Center (P30 EY08126) and NIH/NCRR (G20 RR030956). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Author information

Authors and Affiliations

Division of Allergy, Pulmonary and Critical Care Medicine, Department of Medicine, Vanderbilt University Medical Center, Nashville, TN, USA
Matthew T. Stier, Erin L. Mwizerwa, Chooi Ying Sim, Samantha M. Tanner, Erin M. Wilfong, Sarah N. Obeidalla, V. Eric Kerchberger, Dawn C. Newcomb, Julie A. Bastarache & Lorraine B. Ware
Vanderbilt Center for Immunobiology, Vanderbilt University Medical Center, Nashville, TN, USA
Matthew T. Stier, Erin M. Wilfong, Dawn C. Newcomb & Jeffrey C. Rathmell
Department of Pathology, Microbiology and Immunology, Vanderbilt University Medical Center, Nashville, TN, USA
Allison E. Sewell, Casey M. Nichols, Heather H. Durai, Dawn C. Newcomb, Julie A. Bastarache, Lorraine B. Ware & Jeffrey C. Rathmell
Division of Hematology and Oncology, Department of Medicine, Vanderbilt University Medical Center, Nashville, TN, USA
Erin Q. Jennings
Department of Medicine, Vanderbilt University Medical Center, Nashville, TN, USA
Paul Lindau
Division of Rheumatology and Immunology, Department of Medicine, Vanderbilt University Medical Center, Nashville, TN, USA
Erin M. Wilfong
Department of Biomedical Informatics, Vanderbilt University Medical Center, Nashville, TN, USA
V. Eric Kerchberger
Department of Cell and Developmental Biology, Vanderbilt University, Nashville, TN, USA
Julie A. Bastarache
Ben May Department for Cancer Research, Ludwig Center at the University of Chicago, Chicago, IL, USA
Jeffrey C. Rathmell

Authors

Matthew T. Stier
View author publications
Search author on:PubMed Google Scholar
Allison E. Sewell
View author publications
Search author on:PubMed Google Scholar
Erin L. Mwizerwa
View author publications
Search author on:PubMed Google Scholar
Chooi Ying Sim
View author publications
Search author on:PubMed Google Scholar
Samantha M. Tanner
View author publications
Search author on:PubMed Google Scholar
Casey M. Nichols
View author publications
Search author on:PubMed Google Scholar
Heather H. Durai
View author publications
Search author on:PubMed Google Scholar
Erin Q. Jennings
View author publications
Search author on:PubMed Google Scholar
Paul Lindau
View author publications
Search author on:PubMed Google Scholar
Erin M. Wilfong
View author publications
Search author on:PubMed Google Scholar
Sarah N. Obeidalla
View author publications
Search author on:PubMed Google Scholar
V. Eric Kerchberger
View author publications
Search author on:PubMed Google Scholar
Dawn C. Newcomb
View author publications
Search author on:PubMed Google Scholar
Julie A. Bastarache
View author publications
Search author on:PubMed Google Scholar
Lorraine B. Ware
View author publications
Search author on:PubMed Google Scholar
Jeffrey C. Rathmell
View author publications
Search author on:PubMed Google Scholar

Contributions

M.T.S. and J.C.R. designed the research and prepared the manuscript with contributions from all other authors. M.T.S., A.E.S., C.M.N., E.Q.J., S.N.O. and V.E.K. performed the research and analyzed data. P.L. provided bioinformatics technical support. E.M.W. provided technical support with methods development. D.C.N. provided essential intellectual support in experimental design and planning. J.A.B. and L.B.W. provided essential expertise and clinical samples. H.H.D. provided key technical support. E.L.M., C.Y.S., S.M.T., C.M.N., M.T.S. and A.E.S. collected and/or processed clinical samples.

Corresponding authors

Correspondence to Matthew T. Stier or Jeffrey C. Rathmell.

Ethics declarations

Competing interests

J.C.R. has held stock equity in Sitryx and in the past 3 years has received unrelated research support, travel and honoraria from Sitryx, Caribou, Nirogy, Kadmon, Calithera, Tempest, Merck, Mitobridge and Pfizer. J.C.R. is divested of these interests at the present. E.M.W. has consulting agreements with Boehringer-Ingelheim, AstraZeneca, Allogene and Arcellx. L.B.W. owns stock in Virtuoso Surgical and in the past 3 years has received unrelated research support from Bluejay Diagnostics and consulting fees from Healios, Novartis, Arrowhead and Akebia. The remaining authors declare no competing interests.

Peer review

Peer review information

Nature Immunology thanks the anonymous reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Ioana Staicu, in collaboration with the Nature Immunology team

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Sepsis adjudication algorithm.

Adjudication algorithm used to assign sepsis status to enrolled subjects. UA = urinalysis; WBC = white blood cells; UTI = urinary tract infection; CSF = cerebrospinal fluid; BAL = bronchoalveolar lavage; PMN = polymorphonuclear cells.

Extended Data Fig. 2 Additional features of CD4⁺ T Cells in CI.

a,b, Glutaminolysis dependence (a) and fatty acid oxidation (FAO) dependence (b) in CD4⁺ T_N, T_EM, T_CM, T_reg, T_H1, and T_H17 cells determined by puromycin-incorporation assay at day 2 post-ICU admission in NHC (n = 9), CI-NS (T_N, T_EM, T_CM, and T_reg cells, n = 30; T_H1 and T_H17 cells, n = 29) and CI-Sep (glutaminolysis dependence in T_N, T_EM, and T_CM cells, n = 31; FAO dependence in T_N, T_EM, and T_CM cells, n = 32; glutaminolysis dependence T_reg cells, n = 29; FAO dependence T_reg cells, n = 30; glutaminolysis dependence T_H1 cells, n = 30; FAO dependence T_H1 cells, n = 31; glutaminolysis dependence T_H17 cells, n = 23; FAO dependence T_H17 cells, n = 24). Displayed as percentage of the maximum puromycin incorporation without metabolic inhibitors. c, Complete-case analysis of glycolytic capacity in T_reg cells from patients with PBMC at all timepoints (day 0, 2 and 7) post-ICU admission for CI-NS (n = 11) and CI-Sep (n = 9). Estimated normal range is shaded. d, Cell counts of total CD4⁺ T, T_N, T_EM, T_CM, T_reg, T_H1, and T_H17 cells per μL blood in NHC (n = 4), CI-NS (n = 13) and CI-Sep (n = 18) at day 2 post-ICU admission. e, Flow cytometric quantification of the frequency of Ki67⁺ cells among T_reg cells in NHC (n = 9), CI-NS (n = 30) or CI-Sep (n = 30) T_reg cells at day 2 post-ICU admission. f, Scatter plot showing two-sided Spearman rank correlation of the percentage change in the frequency and population mean change in glycolytic capacity of CD4⁺ T_N, T_CM, T_reg, T_H1, and T_H17 cells in CI-Sep compared to NHC, excluding the T_EM cell subset to prevent pseudoreplication from overlap with nested T_H1 and T_H17 cell subsets. Statistical analysis for a,b,d,e was performed using the Kruskal–Wallis test with Dunn’s post-hoc testing. Data for c are presented as mean ± SEM. Boxplots show the median with boxes bounded by interquartile range (IQR, 25–75th percentile) and whiskers extending to the minimum and maximum values within 1.5 x IQR. Each data point represents an individual donor. K–W = Kruskal–Wallis.

Source data

Extended Data Fig. 3 Additional glycolytic and mitochondrial characteristics of CD4⁺ T cell subsets in CI.

a,b, Representative flow cytometry histograms (left) and quantification (right) of expression of glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (a) and phosphofructokinase (PFKP) (b) in CD4⁺ T_N, T_EM, T_CM, and T_reg cells from day 2 post-ICU admission. Shown as fold change of the MFI in CI-NS (GAPDH, n = 14; PFKP, n = 13) and CI-Sep (GAPDH, n = 19; PFKP, n = 14) relative to NHC (n = 11). c, Proportion of cells with high mitochondrial mass and low mitochondrial membrane potential (MMP) (Mass^hi MMP^lo) among CD4⁺ T_N, T_EM, T_CM, and T_reg cells from NHC (n = 10), CI-NS (n = 16), and CI-Sep (n = 22) at day 2 post-ICU admission. Statistical analysis for a-c was performed using the Kruskal–Wallis test followed by Dunn’s post-hoc testing. Data are presented as mean ± SEM. Each data point represents an individual donor. FMO = fluorescence minus one control; MFI = mean fluorescence intensity (geometric).

Source data

Extended Data Fig. 4 Additional functionalities of CD4⁺ T cells exposed to mitochondrial stress.

a, Flow cytometric quantification of IL-4 expression in NHC (n = 10), CI-NS (n = 14), and CI-Sep (n = 13) CD4⁺FOXP3⁻ T_conv cells from PBMC isolated day 2 post-ICU admission stimulated with CD3/CD28/CD2 Ab for 24 h. Shown as the frequency of IL-4⁺ cells among T_conv cells (left) and the MFI of IL-4 among IL-4⁺ T_conv cells (right). b, Flow cytometric quantification of IFNγ and IL-17A expression as in (a) in NHC (n = 10), CI-NS (n = 14), and CI-Sep (n = 13) CD4⁺FOXP3⁺ T_reg cells. c, Flow cytometric quantification of CD39 and FOXP3 expression in NHC (n = 10), CI-NS (n = 14), and CI-Sep (n = 13) CD4⁺FOXP3⁺ T_reg cells from PBMC isolated day 2 post-ICU admission stimulated with CD3 + CD28 + CD2 Ab for 24 h with or without mitochondrial impairment with low-dose oligomycin (10 nM). Shown as the percent change in MFI among T_reg cells positive for each marker from stimulated, oligomycin-treated PBMC relative to stimulated, untreated PBMC. d, Flow cytometric quantification of IL-10, IFNγ, and IL-17A expression as in (c) in NHC (n = 10), CI-NS (IL-10⁺, IFNγ⁺, or IL-17A⁺ cells and IFNγ and IL-17A MFI, n = 14; IL-10 MFI, n = 13), and CI-Sep (n = 13) CD4⁺FOXP3⁺ T_reg cells. e, Flow cytometric quantification of PD-1 and LAG-3 expression as in (c) in NHC (n = 10), CI-NS (n = 14), and CI-Sep (n = 13) CD4⁺FOXP3⁺ T_reg cells. f, Flow cytometric quantification of TNF, IFNγ, IL-4, IL-17A, and IL-2 expression as in (c) in NHC (n = 10), CI-NS (n = 14), and CI-Sep (n = 13) CD4⁺FOXP3⁻ T_conv cells. g, Flow cytometric quantification of CD25, PD-1, and LAG-3 expression as in (c) in NHC (n = 10), CI-NS (n = 14), and CI-Sep (n = 13) CD4⁺FOXP3⁻ T_conv cells. Statistical analysis was performed using the Kruskal–Wallis test followed by Dunn’s post-hoc testing. Data are presented as mean ± SEM. Each data point represents an individual donor. K–W = Kruskal–Wallis; MFI = mean fluorescence intensity (geometric).

Source data

Extended Data Fig. 5 scRNA-seq quality control metrics.

a, Number of unique genes (nFeature_RNA) per cell after artifact and ambient RNA removal with CellBender. Dashed lines represent subsequent filtering at 400 and 5000 features. b, Percentage of mitochondrial reads (percent.mito) per cell after artifact and ambient RNA removal with CellBender. Dashed line represents subsequent filtering threshold of 20% mitochondrial reads. c, Frequency of doublets removed with scDblFinder. Each sample is listed along the x-axis.

Extended Data Fig. 6 Transcriptomic comparison of CI-NS versus CI-Sep CD4⁺ T cells.

a, Volcano plots showing differential gene expression by DESeq2 in pseudobulked scRNA-seq samples of CI-Sep (n = 19) versus CI-NS (n = 19) CD4⁺ T_N, T_EM, and T_reg cells. Dashed lines denote a fold change > 1.5 and a p_adj < 0.05. b, Dot plot showing single-cell normalized, log-transformed mean expression of selected CD4⁺ T_H genes in NHC (n = 9), CI-NS (n = 19), and CI-Sep (n = 19) CD4⁺ T_N, T_EM, and T_reg cells. c, Normalized scRNA-seq pseudobulked IL2 expression in NHC (n = 9), CI-NS (n = 19, and CI-Sep (n = 19) CD4⁺ T_N and T_EM cells. d, GSEA showing the top enriched and de-enriched gene sets by normalized enrichment scores (NES) for KEGG pathways in CI-Sep (n = 19) versus CI-NS (n = 19) CD4⁺ T_N, T_EM, and T_reg cells. e, Hierarchical clustering and similarity matrix of KEGG pathways (p_adj < 0.05) as in (d) in CI-Sep (n = 19) versus CI-NS (n = 19) CD4⁺ T_N, T_EM, and T_reg cells based on Jaccard similarity index of their leading-edge genes. Clustered pathways of interest are annotated in pink with gene overlap reported. f, Dot plot showing single-cell normalized, log-transformed mean expression of HLA genes in CI-NS (n = 19) and CI-Sep (n = 19) CD4⁺ T_N, T_EM, and T_reg cells. g, Volcano plots showing differential gene expression as in (a) of HLA genes in CI-Sep (n = 19) versus CI-NS (n = 19) CD4⁺ T_EM cells. Dashed lines denote a fold change > 1.5 and a p_adj < 0.05. h, Hierarchical clustering of NHC (n = 9), CI-NS (n = 19), and CI-Sep (n = 19) CD4⁺ T_N, T_EM, and T_reg cells. Statistical analysis for a,g was performed using the Wald test and Benjamini–Hochberg FDR correction for multiple testing; for d using permutation testing of the enrichment score (weighted Kolmogorov–Smirnov statistic) with Benjamini–Hochberg FDR correction for multiple testing. Boxplots show the median with boxes bounded by interquartile range (IQR, 25–75th percentile) and whiskers extending to the minimum and maximum values within 1.5 x IQR. Each data point for c represents an individual donor.

Extended Data Fig. 7 Additional effects of kynurenine on CD4⁺ T cells from CI-Sep patients.

a, Flow cytometry histograms of kynurenine (KYN) uptake following 5 minute pulse of NHC and CI-Sep CD4⁺ T_reg cells with kynurenine (100 μM), without pre-treatment or pretreated with SLC7A5 inhibitor BCH (10 mM) or L-leucine (LEU, 5 mM) or incubated at 4 °C prior to adding kynurenine. b,c, Glycolytic capacity by puromycin-incorporation assay in CI-Sep CD4⁺ T_N, T_EM, and T_reg cells from paired PBMC cultures treated with vehicle or KMO inhibitor GSK180 (KMOi, 10 μM) for 3 h (n = 7) (b) or 24 h (n = 6) (c). Displayed as percentage of the maximum puromycin incorporation without metabolic inhibitors. d, LC–MS/MS showing kynurenine concentration in HPLM with or without dialyzed FBS (dFBS), fresh or stored at 4 °C for 30 days. e, Glycolytic capacity as in (b) in NHC (n = 4) and CI-Sep (n = 4) CD4⁺ T_N, T_EM, and T_reg cells from paired PBMC cultures treated with vehicle or L-Leucine (500 μM) for 24 h. f, Glycolytic capacity as in (b) in NHC (n = 5) and CI-Sep (n = 7) CD4⁺ T_N, T_EM, and T_reg cells from paired PBMC cultures treated with vehicle or LDHA inhibitor FX-11 (LDHi, 20 μM) for 24 h. g,h Flow cytometric quantification of PD-1 and LAG-3 expression (g) or IFNγ, IL-17A, and IL-2 cytokine expression (h) in NHC (n = 10) and CI-Sep (n = 12) CD4⁺FOXP3⁻ T_conv cells from PBMC isolated 2 days post-ICU admission stimulated with CD3/CD28/CD2 Ab for 24 h with or without low-dose oligomycin (10 nM) and treated with vehicle or KMO inhibitor GSK180 (10 μM). Shown as percent change in the frequency of PD-1⁺, LAG-3⁺, IFNγ⁺, IL-17A⁺, or IL-2⁺ cells among T_conv cells from stimulated, oligomycin-treated PBMC with vehicle or KMO inhibitor GSK180 relative to stimulated, untreated PBMC. i, FBA as a heatmap showing in silico flux potential (z-scored) for individual metabolic reactions in scRNA-seq data from NHC and CI-NS CD4⁺ T_N, T_EM, and T_reg cells. j, FBA pairwise comparisons of Cohen’s d effect size of tryptophan metabolism reactions among NHC, CI-NS, and CI-Sep T_N, T_EM, and T_reg cells. k, Scatterplot showing FBA results as in (i) for CI-NS T_reg cells versus NHC T_reg cells and CI-Sep T_reg cells versus NHC T_reg cells reported as Cohen’s d values. A linear regression line modeling the association between all reactions is shown in red. Statistical analysis for b,c,g,h was performed using with two-sided Wilcoxon signed-rank test for paired sample analysis; for e,f, using two-tailed permutation testing for paired samples; for j, using two-tailed Mann–Whitney U testing with Benjamini–Hochberg FDR correction; for k, using a two-sided t-test of the regression slope. Data for b,c,e-h are presented as mean ± SEM. Each data point in b,c,e-h represents a biologic or in d represents a technical replicate.

Source data

Extended Data Fig. 8 Representative flow cytometry gating strategies.

a, Gating strategy for the identification of FSC^loSSC^loCD3⁺CD4⁺ T_N cells (FOXP3⁻CD45RA⁺CCR7⁺), T_EM cells (FOXP3⁻CD45RA⁻CCR7⁻), T_CM cells (FOXP3⁻CD45RA⁻CCR7⁺), T_H1 cells (FOXP3⁻CXCR3⁺CCR4⁻CCR6⁻), T_H17 cells (FOXP3⁻CXCR3⁻CCR4⁺CCR6⁺), and T_reg cells (FOXP3⁺CD25⁺) using surface and intracellular staining. Representative anti-puromycin staining with and without metabolic inhibitors for NHC CD4⁺ T_N and T_EM cells is shown. b, Gating strategy for the identification of unstimulated FSC^loSSC^loCD3⁺CD4⁺ T_conv cells (CD25^loCD127^hi) and T_reg cells (CD25^hiCD127^lo) using surface staining only. c, Gating strategy for the identification of T_conv cells (FSC^loSSC^loCD4⁺FOXP3⁻) and T_reg cells (FSC^loSSC^loCD4⁺FOXP3⁺) in CD3 + CD28 + CD2 Ab stimulation experiments.

Extended Data Table 1 Baseline characteristics and incident ICU clinical data for study participants

Full size table

Extended Data Table 2 Flow cytometry antibodies

Full size table

Supplementary information

Reporting Summary

Supplementary Table 1

Pseudobulk DESeq2 results.

Supplementary Table 2

Over-representation analysis results.

Supplementary Table 3

GSEA results.

Supplementary Table 4

Compass FBA results.

Source data

Source Data Fig. 1

Source data for figures.

Source Data Fig. 2

Source data for figures.

Source Data Fig. 3

Source data for figures.

Source Data Fig. 4

Source data for figures.

Source Data Fig. 6

Source data for figures.

Source Data Fig. 7

Source data for figures.

Source Data Extended Data Fig. 2

Source data for figures.

Source Data Extended Data Fig. 3

Source data for figures.

Source Data Extended Data Fig. 4

Source data for figures.

Source Data Extended Data Fig. 7

Source data for figures.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Stier, M.T., Sewell, A.E., Mwizerwa, E.L. et al. Metabolic adaptations rewire CD4⁺ T cells in a subset-specific manner in human critical illness with and without sepsis. Nat Immunol (2026). https://doi.org/10.1038/s41590-025-02390-6

Download citation

Received: 11 February 2025
Accepted: 04 December 2025
Published: 15 January 2026
Version of record: 15 January 2026
DOI: https://doi.org/10.1038/s41590-025-02390-6

Subjects

Abstract

Similar content being viewed by others

Main

Results

ICU-derived Treg cells have higher glycolytic capacity

Treg cells retain suppressive features with stress in CI

CI globally remodels CD4+ T cell transcriptional states

Sepsis status subtly alters CD4+ T cell transcription

Mitochondrial ROS spurs CI Treg cell glycolytic capacity

Kynurenine metabolism supports suppressive CI Treg cells

Treg cell glycolytic capacity links to poor ICU metrics

Discussion

Methods

Human participants, clinical metrics and sepsis adjudication

Peripheral blood mononuclear cell isolation and storage

Cryorecovery of PBMCs

Flow cytometry

Puromycin-incorporation assay

Mitochondrial mass, membrane potential and ROS

Ex vivo restimulation for CD4+ T cell function

Kynurenine uptake assay

LC–MS/MS quantification of kynurenine in cell culture medium

Single-cell library preparation and RNA sequencing

ScRNA-seq preprocessing, quality control, integration and modeling

scRNA-seq clustering, visualization and annotation

Cell proportionality

Differential gene expression and pathway analysis

Flux balance analysis

Statistics and reproducibility

Reporting summary

Data availability

Code availability

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding authors

Ethics declarations

Competing interests

Peer review

Peer review information

Additional information

Extended data

Supplementary information

Source data

Rights and permissions

About this article

Cite this article

Share this article

Search

Quick links

ICU-derived T_reg cells have higher glycolytic capacity

T_reg cells retain suppressive features with stress in CI

CI globally remodels CD4⁺ T cell transcriptional states

Sepsis status subtly alters CD4⁺ T cell transcription

Mitochondrial ROS spurs CI T_reg cell glycolytic capacity

Kynurenine metabolism supports suppressive CI T_reg cells

T_reg cell glycolytic capacity links to poor ICU metrics

Ex vivo restimulation for CD4⁺ T cell function