PGE2 inhibits TIL expansion by disrupting IL-2 signalling and mitochondrial function – Nature

Tumour samples

Tumour samples were collected from individuals with melanoma, non–small cell lung cancer, ovarian cancer, and breast cancer undergoing surgical treatment between October 2016 and August 2023 at the Centre Hospitalier Universitaire Vaudois (CHUV), Lausanne, Switzerland, under a specific protocol TIL-ME study with the number 247/13. The subsequent samples were collected by using the Pre-IT protocol (2016-02094).

Informed consent was obtained from any patients undergoing surgery at the CHUV. Patients were approached and requested to consent to donating their samples for translational research if the samples were not required for clinical pathological evaluation. There is no tissue selection based on patient sex, gender, history, age, previous treatments and thus no potential selection bias exists. The population characteristics were blinded to researchers.

For the melanoma cohort, we re-analysed results already published from a phase 1 trial of ACT with TILs in patients with melanoma (ClinicalTrials.gov ID NCT03475134)^11,12. For correlation of PGE₂ in the supernatant and TIL expansion, we collected supernatant of TIL cultures from patients enrolled in a phase I trial of ACT with TILs in solid tumours (CHUV-DO-0018-NeoTIL-2019; ClinicalTrials.gov ID NCT04643574).

The reported work was carried out in conformity with the Helsinki Declaration, and the protocol was authorized by the ethics committee of the canton of Vaud (Switzerland). Prior to the collection of study materials, all patients provided written informed consent. Tumour samples were processed and stored as previously described⁴⁷.

Mouse experiments

All mice were housed in a conventional animal facility of University of Lausanne and kept in individually ventilated cages, between 19–23 °C with 45–65% humidity and a 12 h dark/light cycle. All studies were approved by the Veterinary Authority of the Canton of Vaud and performed in accordance with Swiss ethical guidelines. No statistical methods were used to predetermine sample size. Sample sizes for in vivo assays were determined empirically based on previous work. For Winn assay, mice were randomly allocated to the different treatment groups based on weight of the mice while for the mouse adoptive cell therapy tumour control experiment, mice were randomized based on tumour size. Mouse experiments were performed blind to experimental conditions.

Tumour processing and TIL expansion

For conventional pre-REP TIL expansion, freshly received solid tumour specimens were minced into 1 2 mm² fragments and plated in a 48-well plate in 500 μl RPMI + 10% FBS + 1% PS containing 6,000 IU ml⁻¹ IL-2 (Proleukin)⁴⁷. Medium was refreshed twice a week and TILs splitted when confluent. In some cases, PGE₂ at 1 μM, 10 μM EP2 (TG4-155, Cayman) and EP4 (ONO-AE3–208, ONO Pharmaceuticals) inhibitors, 11.7 μM ketorolac (TORA-DOL, 30 mg ml⁻¹) were added once to the cultures at day 0. TG4-155 is a potent, brain-permeant and selective EP2 receptor antagonist with an inhibition constant (K_i) of 9.9 nM and 550- to 4,750-fold selectivity for EP2 over EP1, EP3 and EP4⁴⁸. ONO-AE3-208 is an orally active EP4-selective antagonist (K_i of 1.3 nM for EP4, 30 nM for EP3 and more than 10 μM for the prostanoid receptors EP1 and EP2)⁴⁹.

Non-adherent plates were used to avoid fibroblast overgrowth in response to PGE₂, which would affect TIL expansion. After 14–28 days of pre-REP TIL expansion, expanded TILs were counted, used for in vitro assays, or further expanded in REP phase for 14 days by using 40 Gy irradiated peripheral blood mononuclear cells (PBMCs) as feeder cells, 30 ng ml⁻¹ OKT3 and 3,000 IU ml⁻¹ IL-2.

Blood processing and generation of RA T cells

PBMCs from leukapheresis of healthy donors were isolated by Ficoll gradient. Peripheral blood lymphocytes (PBLs) were enriched from PBMC using a pan T cell isolation kit (Miltenyi, 130-096-535) according to the manufacturer’s instructions. To generate RA T cells, PBLs were stimulated with CD3/CD28 beads (Dynabeads, 11132D) at 1:1 ratio for 10 days, changing beads every 3–4 days for a total of 3 stimulations in low-dose 60 IU ml−1 IL-2⁵⁰. ‘Unstimulated’ T cells were cultured two days with 60 IU ml⁻¹ IL-2.

Protocol for the generation of mouse T cells overexpressing PGC1α and mouse RA T cells

OT-1-CD45.1 mice were obtained from and maintained in a conventional animal facility at the University of Lausanne following institutional guidelines. This study was approved by the Veterinary Authority of the Canton of Vaud (under license 6387) and performed in accordance with Swiss ethical guidelines. All mice (female) were housed in a conventional animal facility of University of Lausanne and kept in individually ventilated cages, between 19–23 °C with 45–65% humidity and a 12 h dark/light cycle. For mouse T cell PGC1α overexpression experiment, CD8⁺ OT-1 T cells were isolated from the spleens of OT-1 mice using CD8⁺ T cell Isolated Kit (MojoSortTM, 480044). CD8⁺ T cells (1 × 10⁶ cells per ml) were seeded into 24-well plates with a volume of 2 ml per well. Unstimulated CD8⁺ T cells were activated by treatment with anti-CD3 (2 μg ml⁻¹, Invitrogen, 16-0031-86), anti-CD28 (1 μg ml⁻¹, Invitrogen, 16-0281-86), and IL-2 (10 ng ml⁻¹, Pepro Tech, AF-200-02-1000) concurrently for 72 h. Cells were then transduced with scramble or PGC1α-overexpressing retroviral construct⁵¹. Cells were stimulated as per protocol.

CD8⁺ OT-1 T cells were purified from the spleen of OT-1-CD45.1 mouse using a negative selection using the EasySep Mouse T cell Isolation kit (Stemcell Ref 19851).

Purified OT-1 CD8⁺ T cells were cultured in complete T cell medium: RPMI 1640 Glutamax supplemented with 10% heat-inactivated FBS, 100 UI ml⁻¹ penicillin, 100 μg ml⁻¹ streptomycin, 1 mM Hepes, 10 mM non-essential amino acids and 50 μM β-mercaptoethanol. Unstimulated, OT-1 T cells were cultured in T cell medium containing human IL-7/IL-15 at 5 ng ml⁻¹. The medium was refreshed every 2 days. For repeatedly stimulation, OT-1 T cells were first stimulated with anti-CD3/CD28 beads (2 beads per cell) in the presence of 5 UI ml⁻¹ of IL-2 during the first 3 days of culture and with IL-7/IL-15 at 5 ng ml⁻¹ for 2 extra days. At day 5 after culture initiation, anti-CD3/CD28 beads were removed, and cells were re-stimulated at a concentration of 1 × 10⁶ cells per ml in the presence of 10 ng ml⁻¹ SIINFEKL OVA peptide. OVA peptide stimulation was repeated on day 6 and 7. At day 8, unstimulated cells and RA T cells were collected for downstream experiments.

Winn-type assay

NOD SCID common gamma KO mice were obtained from and maintained in a conventional animal facility at the University of Lausanne following institutional guidelines and kept in individually ventilated cages between 19–23 degrees with 45–65% humidity and a 12 h dark/light cycle. This study was approved by the Veterinary Authority of the Canton of Vaud (under license 3623b) and performed in accordance with Swiss ethical guidelines. All animals (male mice) were used at ages of 11 weeks. Control or COXi-expanded TILs were mixed with fresh autologous melanoma tumour cells in a 1:1 (T cell:tumour) ratio. A total of 1 × 10⁶ total cells (5 × 10⁵ tumour cells+ 5 × 10⁵ TILs) were subcutaneously injected into the right flank of each mouse (11-week-old) in 100 µl of PBS. Mice were monitored three times per week, and tumour volumes were calculated using the formula: V  =  (L × W²)/2. Mice were euthanized once tumours reached 1,000 mm³, or, according to regulation, if they became distressed, moribund or the tumour became necrotic. As negative controls, cancer cells (5 × 10⁵ cells) were injected with PBS.

NSG in vivo tumour homing

IL-2 NOG mice (Taconic Biosciences) were maintained in a conventional animal facility at the University of Lausanne following institutional guideline and kept in individually ventilated cages, between 19–23 °C with 45–65% humidity and a 12 h dark/light cycle. This study was approved by the Veterinary Authority of the Canton of Vaud (under license 3746) and performed in accordance with Swiss ethical guidelines. Six- to nine-week-old female mice were anaesthetized with isoflurane and subcutaneously injected with 1 × 10⁶ human melanoma cells. At day 14, autologous 9 × 10⁶ control and COXi TILs were injected in the retro-orbital vein. At day 14 post-ACT, mice were euthanized by CO₂ inhalation. Tumours were collected and dissociated in RPMI 1640 GlutaMAX with 0.3 PZ activity units per ml of collagenase and 30 IU ml⁻¹ of pulmozyme (Roche) for 1 h at 37 °C, 5% CO₂ on an orbital shaker. After digestion, tumour specimen suspension was filtered through a 100-µM nylon cell strainer and washed with PBS. Dissociated cells were then labelled for flow cytometry analysis.

Cell culture and in vitro assays

Cells were maintained at 37 °C in complete medium R-10: RPMI 1640 supplemented with 2 mM l-glutamine, and 100 μg ml⁻¹ penicillin and 100 U ml⁻¹ streptomycin and 10% (v/v) heat-inactivated FBS or 8% human serum for T cells and TILs respectively. All T cell cultures were performed under low-dose 60 IU ml⁻¹ IL-2 if not otherwise specified. For in vitro treatment of cells, MHY1485 (100 nM, Sigma-Aldrich, SML0810), everolimus (50 nM, Sigma-Aldrich, SML2282), 10 mM NAC, 100 μM vitamin E, 50 μM BAPTA-AM or 100 μM Rp-8-CPT cAMP analogue were added to the cultures. For metabolic experiments and metabolomics, Human Plasma-Like Medium (HPLM, A4899101, Gibco) was used. Cell count was performed using the AccuChip Kit with the ADAM-MC (NanoEntek) automated counter using propidium iodide to identify healthy cells per the manufacturer instruction.

Autologous tumour cell lines for tumour recognition assay were established from primary tumours by the Center of Experimental Therapies at CHUV. All cell lines were tested and negative for Mycoplasma.

Cell viability assay

Fifty thousand TILs were seeded on 96-well plates and treated with PGE₂ (1 µM). MCC950 (pyroptosis), Fst1 (ferroptosis), zVAD-FMK (apoptosis) and necrostatin 1S (necroptosis) inhibitors were added along with PGE₂ at the indicated doses. Cell viability was assessed 72 h after the treatment using Cell Counting Kit 8 (WST-8/CCK8) (ab228554) as an indicator of viable cells. The cell viability was expressed as relative values compared to the control sample, which was defined as 100%. MCC950 (5381200001) was purchased from Sigma-Aldrich; Fst1 (17729) was purchased from Cayman; zVAD-FMK (ALX-260-02) was purchased from Enzo Life Sciences; and Nec 1 s (2263) was purchased from BioVisiony.

In vitro tumour recognition assay

Expanded REP TILs were rested for 2 days in 60 IU ml⁻¹ IL-2. Autologous tumour cell lines were plated at 70–90% confluence in a flat-bottom 96-well plate and let to adhere overnight. One million TILs were added to the wells for overnight co-culture. TILs were then collected, and tumour recognition was assessed via flow cytometry analysis to quantify 41BB surface expression (CD137) or cytokine secretion by TNF and IFNγ intracellular staining. Cells were cultured with 1× Brefeldin A (eBiosciences, 00-4506-51) for cytokine secretion measurement.

RNA isolation and quantitative real-time PCR

Total RNA was isolated from 1 × 10⁶ cells with Trizol reagent (Life Technologies), followed by RNA purification using the RNA Easy Mini Kit (Qiagen). After treatment with RNase-free DNase I, 1 μg of total RNA was reverse-transcribed using PrimeScript First Strand cDNA Synthesis Kit (Takara Bio) as indicated by manufacturer. Quantitative real-time PCR was performed using TaqMan Fast Universal PCR reagents according to the manufacturer’s instructions. PCR amplification of the housekeeping gene GAPDH was performed for each sample as a control to allow normalization among samples. Each sample was run in triplicate, and each PCR experiment included three non-template control wells. The following primers were used: IL2RA (Hs00158122_m1), IL2RB (Hs01081697_m1), IL2RG (Hs00415671_m1), PTGER2 (Hs00168754_m1), PTGER3 (Hs00168755_m1), PTGER4 (Hs00168761_m1), PGC1A (Hs00173304_m1), PGC1B (Hs00993805_m1), HIF2A (Hs01026149_m1), CREB3L3 (Hs00962115_m1), CPT1A (Hs00912671_m1), GPX4 (Hs00989766_g1), Myb (Hs00920556_m1), TCF7 (Hs01556515_m1), GAPDH (Hs02786624_g1), ACSL4 (Hs00244871), LPCAT3 (Hs01553683), FSP1 (Hs00210845) and GLS2 (Hs00998733).

RNA sequencing and data analysis

RNA from unstimulated and RA CD8⁺ T cells was extracted using the RNA easy kit, and RNA quality was assessed using a Fragment Analyzer. RNA-sequencing libraries were prepared for Illumina TruSeq Stranded Total RNA reagents according to the manufacturer’s instruction. Cluster generation was performed with the libraries using the Illumina HiSeq PE Cluster Kit v4 cBot reagents and sequenced on the Illumina HiSeq 4000 SR using HiSeq SBS Kit V4 reagents. The Illumina Pipeline Software version 1.84 was used to process the sequencing data.

Illumina paired-end sequencing reads were aligned to the human reference GRCh37.75 genome using STAR aligner (version 2.6.0c) and the two-pass method as briefly follows: the reads were aligned in a first round using the –runMode alignReads parameter, then a sample-specific splice-junction index was created using the –runMode genomeGenerate parameter. Finally, the reads were aligned using this newly created index as a reference. The number of counts was summarized at the gene level using htseq-count (version 0.9.1). The Ensembl ID were converted into gene symbols using the biomaRt package (version 2.58.1) and only protein-coding, immunoglobulin and TCR genes were conserved for the analysis. Read counts were normalized into reads per kilobase per million (RPKM) and log₂-transformed after addition of a pseudo-count value of 1. Differential expression analyses were performed using the limma (v3.54.0). The PGE₂ signature was derived from the top 63 significant (FDR-corrected) upregulated genes in CD8⁺ PBLs upon PGE₂ treatment. Pathways scores were generated using ssGSEA from the R-package GSVA (version 1.44.5).

scRNA-seq and scTCR-seq in TIL-ACT patients

Thirteen patients were enrolled in a phase I trial designed to test the feasibility of ACT with TILs (ClinicalTrials.gov ID NCT03475134). Two datasets profiling the TME by scRNA-seq (13 patients sorted for viable cells) and matched scRNA-seq and scTCR-seq data (13 patients sorted for CD45⁺ cells) were used as described¹¹. We computed gene signature scores in CD8⁺ T cells for reactome pathways taken from MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/; extracted from the C2 collection) by using the AUCell R package. The IL-2 signalling signature was extracted from Reactome under the ‘REACTOME_INTERLEUKIN_2_SIGNALING’ name. The PGE₂ signature score was computed using the AUCell function and using as gene signature the 63 genes significantly upregulated by PGE₂ treatment (versus control in RA CD8⁺ T cells, adjusted P value < 0.05, FC > 1). The PGE₂ signature score was then computed for each cell. In patient-level analyses, signature scores were averaged pseudobulked per patient. CD8⁺ T cells were classified in several categorized according to their tumour-reactivity and ‘expanded in ACT product’ status. These two categories were defined as follows: annotated CD8⁺ clonotypes with their validated tumour-reactivity and their expansion profiles (using bulk TCRβ sequencing of the ACT product). Both categories, clinical responses and related data are fully described in Chiffelle et al.¹².

TCR cloning and tumour reactivity validation

Tumour reactivity interrogation from expanding TILs of the ACT products of the patients with melanoma and methodology was previously described¹². In brief, TCRαβ pairs were cloned into recipient activated T cells or Jurkat cell line (TCR/CD3 Jurkat-luc cells (NFAT), Promega, stably transduced with human CD8αβ and TCRαβ CRISPR-KO). Full-length codon-optimized DNA sequences including TCR mouse constant regions were synthesized at GeneArt (Thermo Fisher Scientific) or Telesis Bio. DNA served as template for in vitro transcription and polyadenylation of RNA molecules as per the manufacturer’s instructions (Thermo Fisher Scientific). Autologous T cells were activated with Dynabeads Human T Activator CD3/CD28 beads (Thermo Fisher Scientific) at a ratio of 0.75 beads: 1 total PBMC in the presence of 50 IU mL⁻¹ IL-2 (Proleukin). After 3 days of incubation at 37 °C and 5% CO₂, beads were removed and activated T cells rested for 2 days before use. To transfect TCRαβ pairs into T cells and Jurkat cells, the Neon electroporation system (Thermo Fisher Scientific) was used, following the manufacturer’s instructions. In brief, cells were mixed with 300 ng of TCRα chain RNA together with 300 ng of TCRβ chain RNA and electroporated with the following parameters: 1,600 V, 10 ms, 3 pulses and 1,325 V, 10 ms, 3 pulses, for T and Jurkat cells, respectively. To assess antitumour-reactivity, 10⁵ TCR RNA-electroporated cells and 2 × 10⁴ to 10⁵ autologous tumour cells pre-treated with IFNγ were co-cultured in 96-wells plate. After overnight incubation, T cells were recovered and the upregulation of CD137 in T cells was evaluated by staining with anti-CD137 (Miltenyi), anti-CD3 (Biolegend or BD Biosciences), anti-CD4 (BD Biosciences), anti-CD8 (BD Biosciences) and anti-mouse TCRβ-constant (Thermo Fisher Scientific) and with Aqua viability dye (Thermo Fisher Scientific). With Jurkat cells, the luciferase assay was performed using the Bio-Glo Luciferase Assay System (Promega). The LSRFortessa (BD Bioscience) and IntelliCyt iQue Screener PLUS (Bucher Biotec) flow cytometers were used for acquisition. Fluorescence-activated cell sorting (FACS) data analysis was performed with FlowJo v10 (TreeStar). Luminescense was measured with a Spark Multimode Microplate Reader (Tecan).

Bulk TCRβ sequencing and TCR repertoire analysis

Bulk TCRβ sequencing of the ACT products from control and COXi-expanded TILs was performed as previously described⁵². In brief, TILs mRNA was isolated and amplified using commercially available kits (from Life Technologies and Ambion, respectively) with the following modifications: in vitro transcription was performed at 37 °C for 16 h. First-strand cDNA was synthesized using the Superscript III (Thermo Fisher) and a collection of TRAV- or TRBV-specific primers. TCRs were then amplified by PCR with a single primer pair binding to the constant region and the adapter linked to the TRAV or TRBV primers added during the reverse transcription. A second round of PCR was performed to add the Illumina adapters containing the different indexes. The TCR products were purified, quantified and loaded on the MiniSeq instrument (Illumina) for deep sequencing of the TCRβ chain. The TCR sequences were further processed using ad hoc Perl scripts to: (1) pool all TCR sequences coding for the same protein sequence; (2) filter out all out-of-frame sequences; (3) determine the abundance of each distinct TCR sequence. TCRs with a single read were not considered for the analysis. Richness was assessed by the number of unique TCR sequences present in the repertoire. The clonality was described by the Shannon Entropy or 1-Pielou’s evenness⁷. 10x and 100x expanded (Exp) clones refers to the number of clonotypes with a frequency 10 or 100-fold higher than the median frequency of the repertoire.

Western blot

T cell pellet was lysed with RIPA Lysis and Extraction Buffer (Thermo Fisher) supplemented with protease and phosphatase inhibitors. Protein concentrations were quantified using a Quick Start Bradford assay kit (BioRad). Samples containing 20 μg of protein in NuPAGE LDS Sample Buffer (4×) were separated using 8–12% pre-cast SDS–PAGE gels (BioRad). Proteins were transferred to PVDF membranes and blocked for 1 h in 5% Milk in TBST buffer. PVDF membranes were incubated overnight at 4 °C in 1% BSA TBST buffer with the following primary antibodies at dilution 1:1,000 or otherwise specified: β-actin (K2713, Santa Cruz, sc-47778, 1:2000), JAK1 (B-3 Santa Cruz sc-376996, 1:500), pJAK1 (D7N4Z Cell Signaling 74129), JAK3 (B-12, Santa Cruz, sc-6932, 1:500), pJAK3 (D44E3, Cell Signaling, 5031), STAT1 (D4Y6Z, Cell Signaling, 14995), pSTAT1 (D4A7, Cell Signaling, 7649), STAT3 (D3Z2G, Cell Signaling, 12640), pSTAT3 (D3A7, Cell Signaling, 9145), STAT5 (D206Y Cell Signaling 94205), pSTAT5 (D47E7, Cell Signaling, 9351), AKT (C67E7, Cell Signaling, 4691), pAKT (D9E, Cell Signaling, 4060), mTOR (7C10, Cell Signaling, 2983), pmTOR (D9C2, Cell Signaling, 5536), S6 (5G10, Cell Signaling, 2217), pS6 (D57.2.2E, Cell Signaling, 4858), PGC1a (3G6 Cell Signaling, 2178 s), GPX4 (EPNC1R144, Abcam, 125066).

PVDF membranes were washed 3 times for 10 min at room temperature in TBST buffer and incubated with appropriate secondary horseradish peroxidase-linked antibodies for 1 h at room temperature with 1:10,000 dilution in 1% BSA in TBST buffer: anti-mouse HRP (Dako, p0447), anti-goat HRP (Dako, P0449), anti-rabbit HRP (Dako, P0448). Blots were developed on Fusion FX imaging system (Vilber) using ECL plus reagent (GE Healthcare). Densitometry analysis was performed in ImageJ. Images of all uncut blots can be found in Supplementary Figs. 2 and 3.

Flow cytometry

The following antibodies were used at dilution 1:50 for staining cells for flow cytometry: CD45 (BV570, HI30 Biolegend, 304034), CD4 (BV605, OKT4, Biolegend, 317438), CD8 (BV650, RPA-T8, Biolegend, 301042), Tim3 (APC fire 750, F38-2E2, Biolegend, 345044), CTLA4 (PE, BNI3, Biolegend, 369604), PD-1 (BV421, EH12.2H7, Biolegend, 329920), CD39 (BV711, TU66, BD Bioscience, 563680), Lag3 (AF488, 11C3C65, Biolegend, 369326), Ki67 (PE–Cy7, Ki67, Biolegend, 350526), Ki67 (AF700, B56, BD Bioscience, 561277), CD57 (BV605, QA17A04, Biolegend, 393304), TOX/TOX2 (PE, E6G50, Cell Signaling, 25202), TOX (PE, REA473, Miltenyi, 130-120-716), CD28 (AF700, CD28.2, Biolegend, 302920), CD27 (APC–Cy7, M-T271, Biolegend, 356424), IL-2Rα (FITC, BC96, Biolegend, 302604), CD122 (PE, TU27, Biolegend, 339006), CD132 (APC, TUGh4, Biolegend, 338608), TCF1/TCF7 (AF647, C63D9, Cell Signaling, 6932), CD56 (PE–Cy7, 5.1H11, Biolegend, 362510), CD137 (PE-Cy5, 4B4-1, Biolegend 309808), CD3 (BV510, UCHT1, Biolegend, 300448), CD3 (BV711, UCHT1, BD Bioscience, 563725), CD4 (PE-CF594, RPA-T4, BD Bioscience, 562281), pS6 (PE, cupk43k, eBioscience, 12-9007-42), IFNγ (APC, B27, Biolegend, 506510), TNF (PE–Cy7, MAb11, BD Bioscience, 557647).

For flow cytometry phenotyping analysis, T cells were washed with phosphate-buffered saline (PBS) and stained with Zombie UV fixable (Biologend, Cat:423107) or LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Thermo Fisher, L34957) for 15 min on ice in PBS. Cells were subsequently stained for surface markers in PBS + 2% FBS for 15 min on ice. For intracellular staining, eBioscience Foxp3/Transcription Factor kit was used (Thermo Fisher Scientific, 00-5523-00). Cells were fixed 1 h in fix/perm buffer (Thermo Fisher) and intracellular staining was performed for 45 min at room temperature in perm buffer. After staining, cells were acquired on a four-laser Fortessa (BD Biosciences) with FACS DIVA software v.9.0 (BD Biosciences) and analysed with FlowJo (TreeStar).

For mitochondrial mass and membrane potential staining, cells were incubated for 20 min at 37 °C in an incubator with 25 nM TMRM (Thermo Fisher Scientific, M20036) and 100 nM Mitotracker Green (Thermofisher, M46750) in medium before staining, respectively.

Lipid peroxidation was assessed via Bodipy 581/591 C11 (Thermofisher, D3861). Cells were incubated for 30 min at 37 °C with 2 μM Bodipy 581/591 C11.

To assess proliferation, cells were stained for 7 min in 1 μM CFSE and subsequently washed with PBS + 2% FBS to quench the reaction.

The MitoSOX Mitochondrial Superoxide Indicator (MitoSOX) was used to quantify mitochondria superoxide production following manufacturer’s instructions.

O-propargyl-puromycin (OPP) in TILs was measured with Click-iT Plus OPP Alexa Fluor 488 Protein Synthesis Assay Kit (Thermo Fisher, C10456).

For the analysis of the effect of PGE₂ on downstream signalling pathways (IL-2, CD3, CD3/CD28), pre-REP TILs were cultured for 48 h in the absence of IL-2. On the day of the experiment, TILs were washed and resuspended for 2 h in RPMI with or without PGE₂ (1μM). TILs were then washed and stimulated during 30 min with IL-2 (100 IU ml⁻¹) alone, coated anti-CD3 (0.5 μg ml⁻¹) alone, or combinations of anti-CD3/anti-CD28 in RPMI medium complemented with the following labelling antibodies: CD3 (BioLegend), CD8 (BioLegend) and Aqua viability dye (Thermo Fisher). TILs were then fixed with paraformaldehyde (PFA, 1.6%) for 10 min at room temperature, washed and permeabilized with methanol on ice for 30 min (99.9%). TILs were then washed in PBS 1% BSA and labelled with phosphorylated ribosomal protein S6 (pS6, eBioscience). After staining, cells were acquired on a three-laser iQue Screener PLUS (Sartorius) with iQue ForeCyt software v.6.2 (Sartorius) and analysed with FlowJo X (TreeStar).

For profiling of tumour antigen-specific COXi-treated versus control TILs, cells were labelled with an in-house MART-1 (A27L) HLA*A0201-multimer (ELAGIGILTV, developed by the Peptide and Tetramer Core Facility of the Department of Oncology, UNIL-CHUV, Lausanne, Switzerland).

Representative flow cytometry gating strategies for analysis of CD4⁺ or CD8⁺ PBLs or TILs can be found in Supplementary Fig. 1a,b.

FACS sorting of TILs for metabolic flux reconstruction

Tumours were dissociated as described¹¹, and single-cell suspension samples were stained for FACS sorting. Viable single cells TILs were gated using RedDot-1 /DAPI, CD4 and CD8 markers as depicted in Supplementary Fig. 1c and used for further analyses.

Mass cytometry acquisition and analysis

For the profiling of COXi-treated versus control products, REP TILs were stained with 34 metal-labelled antibodies (Standard BioTools and in house, described in Supplementary Table 5). Cells were first incubated with a 5 μM solution of cisplatin in PBS for viability assessment. Cells were then washed and resuspended in Maxpar Cell Staining Buffer (MCSB, Standard BioTools) with human Fc-receptor blocking solution (Miltenyi) (10 min at room temperature). Following surface staining (30 min at room temperature), cells were then fixed using Cytofix fixation buffer (12 min at room temperature) (BD Biosciences) and permeabilized using Phosflow Perm Buffer III solution (20 min at 4 °C) (BD Biosciences). Intracellular staining was then performed (30 min at room temperature). Cells were next incubated with cell intercalation solution (Standard BioTools) (overnight at 4 °C).

Cells were then washed, resuspended in Maxpar Cell Acquisition Solution MCAS (Standard BioTools) containing EQ Four Element Calibration Beads and filtered into cell strainer cap tubes, immediately prior to CyTOF data acquisition. Data were acquired on a Helios Mass Cytometer (Standard BioTools). Raw mass cytometry data were normalized with the bead passport EQ from the CyTOF Software version 7 (Standard BioTools). Data were then pre-processed using FlowJo v10 (TreeStar) and selected live cells were exported for further analysis.

The following antibodies were used for staining cells for CyTOF at specified dilutions: Granzyme B (106 Cd, GB11, Abcam ab103159, 1:100), Ki67 (111 Cd, B56, Abcam, ab279657, 1:100), granzyme (K145Nd, GM6C3, Santa cruz, sc-56125, 1:200), TCF1 (150Nd, 7F11A10, Biolegend, 655202, 1:100), Eomes (154Sm, WD1928, Invitrogen, 14-4877-82, 1:100), p-p38 (156Gd, D3F9, Standart BioTools, 3156002 A, 1:50), TOX (159Tb, REA, Miltenyi, 130-126-455, 1:100), Tbet (161Dy, 4B10, Standart BioTools, 3161014B, 1:200), FoxP3 (162Dy, PCH101, Standart BioTools, 3162011 A, 1:50), KLRG1 (166Er, SA231A2, Biolegend, 367702, 1:100), CTLA4 (170Er, 14D3, Standart BioTools, 3170005B, 1:50), CD45 (089Y, HI30, Standart BioTools, 3089003B, 1:400), CD57 (110 Cd, HCD57, Standart BioTools, MBS140192, 1:100), CD8a (112 Cd, RPA-T8, Biolegend, 301053, 1:100),CD4 (113 Cd, RPA-T4, Biolegend, 300502, 1:100), HLA-DR (114 Cd, L243, Biolegend, 307602, 1:100), CD3 (141Pr, UCHT1, Standart BioTools, 3141019B, 1:100), OX40 (142^Nd, ACT35, Standart BioTools, 3142018B, 1:50), CD45RA (143Nd, HI100, Standart BioTools, 3143006B, 1:200), CCR5 (144Nd, NP-6G4, Standart BioTools, 3144007 A, 1:200), CD28 (146Nd, CD28.2, Biolegend, 302937, 1:100), CD127 (149Sm, A019D5, Standart BioTools, 3149011B, 1:200), CD103 (151Eu, Ber-ACT8, Standart BioTools, 3151011B, 1:100), TIM-3 (153Eu, F38-2E2, Standart BioTools, 3153008B, 1:200), IL-2Rα (155Gd, 2A3, Biolegend, 356102, 1:100), CD27 (158Gd, L128, Standart BioTools, 3158010B, 1:400), CD39 (160Gd, A1, Standart BioTools, 3160004B, 1:100), CXCR3 (164Dy, G025H7, Biolegend, 353702, 1:100), CCR7 (167Er, G043H7, Standart BioTools, 3167009 A, 1:100), ICOS (169Tm, C398.4 A, Standart BioTools, 3169030B, 1:200), 41BB (173Yb, 4B4-1, Standart BioTools, 3173015B, 1:200), PD-1 (174Yb, EH12.2H7, Standart BioTools, 3174020B, 1:100), LAG3 (175Lu, 11C3C65, Standart BioTools, 3175033B, 1:100), CD56 (176Yb, NCAM16.2, Standart BioTools, 3176008B, 1:400), Viability (Cis-pt,Standard BioTools, 201064) and DNA (195-Ir, Standart BioTools 201192 A).

Subsequently, a hierarchical gating strategy was implemented with the openCyto library, resulting in the generation of three distinct populations: the ‘root’ population, the CD4^hiCD8^low population, and the CD4^lowCD8^hi population, labelled as CD8⁺ TILs in cumulative figures. For clustering within the CD4^lowCD8^hi population, FlowSOM was applied to ensure balanced clustering and each sample was sub-sampled within these populations to a maximum of 150,000 cells. The data underwent clustering using FlowSOM and ConsensusClusterPlus, resulting in metaclusters. The optimal number of metaclusters (11 clusters) was determined based on the average Silhouette width.

Based on the metaclustering results, separation between the COXi-treated and control TILs was visualized the using UMAP. Additionally, violin plots were then created and that belonging to specific clusters (1, 2, 3, 8 and 9) for both COXi-treated and control conditions were used. Cohen’s D effect sizes were calculated for each comparison to assess the significance of differences.

Polychromatic imaging cytometry

For Imagestream analysis, live TILs at 1 × 10⁷ cells per ml were run at 100 cells per second on the ImageStreamX MarkII (Merck Millipore). TILs were stained with relevant antibodies for CD4, CD8, IL-2Rα, IL-2Rβ, IL-2Rγ_c and DAPI. Single stained cells were used as compensation controls. Images were captured at 60× magnification. Data were analysed using the ImageStream Data Analysis and Exploration Software (IDEAS). Colocalization was calculated based on Bright Detail Similarity score, a log-transformed Pearson’s correlation coefficient computed by Amnis.

Confocal microscopy and dSTORM

Cells were plated in chamber slides at a 70%–80% confluence. Following PGE₂ treatment, cells were washed with PBS and fixed with PFA for 8 min at room temperature. After blocking with 5% BSA, fixed cells were incubated overnight at 4 °C with primary antibodies. Secondary antibodies were incubated for 1 h at room temperature. Nuclei were counterstained with DAPI (2 μg ml⁻¹ in PBS) for 5 min at room temperature. Slides were then mounted using Fluoromount-G.

Colocalization was quantified following this analysis procedure: Circular patches surrounding a cell or a group of cells or manual evaluation of IL-2Rβγ_c levels (region of interest) and non- IL-2Rβγ_c signal (random region of interest) were selected. Colocalization values were calculated using a pixel-wise Pearson’s test. Frequency quantification of Pearson’s test values (−1: opposing, 0: no and 1: maximum colocalization).

Immunofluorescence staining was performed on formalin-fixed paraffin-embedded (FFPE) tumoural tissue sections. Four µm-thick FFPE sections were subjected to routine deparaffinization and rehydration, blocked with 5% BSA and incubated overnight at 4 °C with primary antibodies. Secondary antibodies were incubated for 1 h at room temperature. Nuclei were counterstained with DAPI (2 μg ml⁻¹ in PBS) for 5 min at room temperature. Slides were then mounted using Fluoromount-G.

LipidSpot (Biotium, 70065-T) was used following manufacturing protocol on live cells for lipid droplet imaging. Cells were mounted in Vectashield HardSet Mounting Medium.

Proximity ligation assay was performed using Duolink PLA (Sigma) for IL-2Rβ and IL-2Rγ_c using manufacturer’s instruction. Manual evaluation of IL-2Rβ–γ_c signal or circular patches surrounding a cell or a group of cells were draw to assess the numbers of dots from the IL-2Rβ–γ_c signal. All samples were imaged on a Zeiss LSM 780 confocal microscope and analysed using ImageJ.β

For dSTORM, 25 mm round coverslips (Marienfeld, 1.5H, 0117650) were washed with ethanol plasma cleaned for 30 s and coated with poly-l-lysine (Sigma, P8920). Cells wμere seeded on coverslips, fixed with 4% PFA, and stained according to standard immunofluorescence staining with primary antibodies against IL2Rγ and IL2Rβ and secondary antibodies goat anti-mouse AF647 and goat anti-rabbit AF555 (Thermofisher A-21241 and A-21428). Coverslips were mounted using an 35 mm adapter (Okolabs, RA-35-18-2000-06) and covered with 450 μl of STORM buffer (Idylle, KMO-ETE-450-IDY, Everspark 1.0). 555 and 642 lasers were aligned before each round of imaging. Imaging was performed on a Zeiss Elyra 7 microscope with the 63× oil-immersion Plan APOCHROMAT objective (Zeiss, NA 1.46, 1.6× lens). Alexa647 and Alexa555 dyes were imaged in sequential time-series of approximately 20,000 frames each. Both molecules were ground-state depleted and in ultra-high power mode.For each dye, ground-state return was elicited by continuous illumination with a 405 laser. Images were recorded with an Andor iXon + 897 EMCCD. 2D dSTORM data analysis and visualization were carried out with the Zen Black 3.0 SR software (Zeiss) as previously described⁵³.

Fluorescence resonance energy transfer

FRET assays included IL-2Rγ_c–Bv510 as a donor and IL-2Rβ–PE as an acceptor. FACS-FRET measurements were performed using a FACSLSRII SORP (BD Bioscience) equipped with 355 nm, 405 nm, 488 nm, 561 nm and 633 nm lasers. To measure IL-2Rγ_c (donor) signal and FRET, cells were excited with the 405 nm laser and fluorescence was collected in the BV510 channel with a standard 530/30 filter, while the FRET-signal was measured with a 586/15 filter. To measure IL-2Rβ (acceptor) signal, cells were excited with the 561 nm laser while the emission was also taken with a 586/15 filter. For each sample, we evaluated a minimum of 250 positive cells that fell within the background-adjusted gate.

Ca²⁺ signalling

TILs were loaded with 1 μM Fluo-4 (Thermo Fisher) for 30 min at 37 °C, cells were then washed and resuspended in loading medium (RPMI + 10% FCS) in a poly-l-lysine-coated coverslips mounted in a RC-20 closed bath chamber (Warner Instrument). Fluorescence was excited at 490 nm and detected at >515 nm, with an acquisition rate of 10 Hz. The Fluo-4-loaded cells were treated or not with PGE₂ (1 μM), after which healthy cells were identified by their responsiveness to 1 μM ionomycin (Calbiochem). Single-cell video images were obtained on a Nikon Ti2 spinning-disk microscope.

Determination of mitochondrial DNA copy number

Mitochondrial DNA Copy number of CD8⁺ T cells and control or COXi-expanded TILs was determined as previously mentioned⁵⁴. Quantitative real-time PCR (rtPCR) was performed using KAPA SYBR FAST qPCR Kit Master Mix on a QuantStudio 6 Flex Real-Time PCR System (Thermo Fisher) after total DNA was extracted using Genomic-tip 20/G (QIAGEN) (KAPA Biosystems).

ELISA

PGE₂ level in supernatant from expanded TILs was determined using a human PGE₂ ELISA Kit (abcam, ab287802) according to the manufacturer’s instructions. For the solid tumour cohort, PGE₂ levels were measured by ELISA during the first medium change. For the breast and melanoma cohort PGE₂ levels were measured between day 7 and 10. Total and reduced glutathione levels were determined with the GSH + GSSG/GSH Assay Kit (Colorimetric) (abcam, ab239709). Intracellular ATP levels were quantified with ATP Detection Assay Kit (ab113849). Intracellular MDA levels were measured with the TBARS fluorometric microplate assay (FR45, Oxford Biomedical Research).

Electron microscopy

Unstimulated and RA T cells were fixed in glutaraldehyde solution (EMS) 2.5% in phosphate buffer (PB 0.1 M (pH 7.4)) for 1 h at room temperature and post-fixed in a fresh mixture of osmium tetroxide 1% (EMS) with 1.5% of potassium ferrocyanide (Sigma) in PB buffer for 1 h at room temperature. The samples were washed twice in distilled water and dehydrated in ethanol solution (Sigma, St Louis) at graded concentrations (30%, 40 min; 50%, 40 min; 70%, 40 min; 100%, 2 × 1 h). This was followed by infiltration in Spurr resin (EMS) at graded concentrations (Spurr 33% in ethanol, 4 h; Spurr 66% in ethanol, 4 h; Spurr 100%, 2 × 8 h) and finally polymerized for 48 h at 60 °C in an oven. Ultrathin sections of 50 nm thickness were cut transversally at 2, 5 and 6 mm from the root tip and at 2 mm below the hypocotyl-root junction, using a Leica Ultracut (Leica Mikrosysteme), picked up on a copper slot grid 2 × 1 mm (EMS) coated with a polystyrene film (Sigma). Sections were post-stained with uranyl acetate (Sigma) 4% in H₂O for 10 min, rinsed several times with H₂O, followed by Reynolds lead citrate in H₂O (Sigma) for 10 min and rinsed several times with H₂O. Micrographs were taken with a transmission electron microscope Philips CM100 (Thermo Fisher Scientific) at an acceleration voltage of 80 kV with a TVIPS TemCamF416 digital camera (TVIPS) using the software EM-MENU 4.0 (TVIPS). Panoramic alignments were performed with the software IMOD.

Seahorse XFe96 metabolic flux analysis

Oxygen consumption rate was measured at 37 °C using an XFe96 extracellular analyser (Seahorse Bioscience). Twenty-four hours before the experiments, PGE₂ (1 μM) was added to the cultures. On the day of the assay, T cells were plated in Seahorse XFe96 Microplates (2 × 10⁵ cells per well) previously coated with Cell-Tak (22.4 mg ml⁻¹), using Seahorse medium supplemented with glucose (10 mM), pyruvate (1 mM) and glutamine (2 mM). Mitochondrial function was interrogated by the sequential injection of oligomycin (1.5 μM, ATP synthetase inhibitor), FCCP (0.5 μM, uncoupling agent) and antimycin A (0.5 μM, complex III inhibitor) in combination with rotenone (0.5 μM, complex I inhibitor), following standard Seahorse XFe96 protocol. Every point represents an average of n = 6 per experiment.

Reconstruction of a metabolic model for T cells

We generated a reduced model around the metabolic subsystems of interest to study the effect of PGE₂ in CD8⁺ RA T cells and TILs. To this end, we applied the redHUMAN method25 to the human genome-scale metabolic network Recon 3D⁵⁵. We used the composition of the RPMI medium to define the extracellular medium in the model, and we allowed all the inorganic metabolites to be uptaken or secreted.

We selected 45 starting subsystems, namely aminosugar metabolism, arachidonic acid metabolism, arginine and proline metabolism, cholesterol metabolism, chondroitin sulfate degradation, chondroitin synthesis, citric acid cycle, coA catabolism, coA synthesis, eicosanoid metabolism*, fatty acid oxidation*, fatty acid synthesis*, fructose and mannose metabolism, galactose metabolism, glutamate metabolism, glutathione metabolism, glycerophospholipid metabolism, glycine, serine, alanine, and threonine metabolism, glycolysis/gluconeogenesis, glycosphingolipid metabolism, haem synthesis, heparan sulfate degradation, inositol phosphate metabolism, keratan sulfate degradation, keratan sulfate synthesis, leukotriene metabolism, linoleate metabolism, methionine and cysteine metabolism, miscellaneous*, NAD metabolism, nucleotide interconversion*, pentose phosphate pathway, phenylalanine metabolism, purine synthesis, pyrimidine synthesis, pyruvate metabolism, ROS detoxification, sphingolipid metabolism, starch and sucrose metabolism, steroid metabolism, tetrahydrobiopterin metabolism, triacylglycerol synthesis, urea cycle, oxidative phosphorylation, and all the mitochondrial reactions. For subsystems tagged with an asterisk, we included only the part that was deregulated by PGE₂. We used the redHUMAN parameters, D = 1 for redGEM, Smin for redGEMX, and Sminp3 for lumpGEM. As a result, we reconstructed redTcellPGE2, a metabolic model with 2,602 metabolites, 1,898 genes and 5,051 reactions associated with 81 metabolic subsystems.

Transcriptomics data integration to generate a metabolic context-specific CD8⁺ T cell model

Based on the experimental data, we assumed a maximum doubling time of 24 h for the RA T cells and TILs and 36 h when cultured with PGE₂.

We identified in the RNA-seq data 1,072 metabolic genes present in the redTcellPGE2 model. We computed the averaged fold change for each condition (RA PGE₂ versus RA control) for the three replicates. We evaluated the gene–protein–reaction rules in the metabolic model to assign the gene expression to the corresponding enzymes. Next, we classified the enzymes into up- or downregulated using a threshold of 1.3. Therefore, enzymes with a fold change above 1.3 are considered upregulated, and enzymes with a fold change below 0.77 are considered downregulated. The 1,072 genes code for 2,722 reactions, of which 491 are upregulated, and 55 are downregulated based on the transcriptomics data.

In order to integrate the data into the redTcellPGE2 model, we used the method REMI⁵⁶, which integrates transcriptomics data into metabolic models, assuming that deregulations in the gene expression translate to deregulations of the corresponding enzyme abundance and, therefore, in the reaction rate. In particular, REMI imposes constraints so that if the gene associated with a reaction is upregulated, its reaction rate must be higher. Conversely, if the gene is downregulated, the reaction rate will be lower. Then the method maximizes the number of reaction rates simultaneously constrained in the network according to the fold change of the corresponding gene expression between the two conditions¹².

From the REMI results, we observed that out of the 546 deregulated reactions, a maximum of 425 reaction rates could be simultaneously constrained in the network according to their expression profile. We identified 19 alternative sets of 425 reactions and 399 reactions that are common across alternatives.

Metabolic reaction fluxes representative of each CD8⁺ RA T cell treatment

We fixed the ratios of the 399 reactions consistent with the network and used the Artificial Centering Hit-and-Run sampler (ACHR) to sample 100,000 points from the solution space. We then computed for each reaction the mean rate and the mean fold change of the populations of samples between both conditions, and we used them as representative for the analysis of the metabolic state of RA T cells with PGE₂.

Metabolic reaction fluxes representative of each TIL treatment

Similar to the reconstruction of RA T cell specific models treated with PGE₂, we reconstructed TIL-specific models by integrating scRNA-seq data for TILs into the redTcellPGE2 model (Supplementary Table 3). We first identified 60 deregulated metabolic genes in TILs treated or not with PGE₂ (selecting genes with a threshold of at least 1.2-fold change and P values up to 0.5). Next, we used REMI to integrate the data into the metabolic redTcellPGE2 model. Subsequently, we mapped the 60 deregulated genes to 111 metabolic reactions and we identified with the computational analyses that 86 of these fluxes could be simultaneously constrained in the network according to the corresponding deregulation profile. We then sampled 100,000 points from the flux solution space and we computed the fold changes of each reaction rate for TILs treated or not with PGE₂.

Minimal network enrichment analysis to study metabolic functions

To study the effects of PGE₂ on T cells’ metabolic functions, we defined seven metabolic tasks associated with the proliferation, energy, and production of ROS and lipid droplets. Using the composition of the biomass reaction in the model, we extracted the corresponding reactions representing the synthesis of the macromolecules, ATP and superoxide anion production, and the Recon 3D lipid droplet production reaction. To generate minimal networks for the metabolic tasks²⁷, we formulated a mixed integer linear programme and identified the minimum number of reactions required to synthesize each metabolic task and possible alternatives.

Next, after integrating the transcriptomics data into the redTcellPGE2 model, we performed minimal network enrichment analysis (MiNEA) using either the gene expression data or the representative of the metabolic fluxes computed by sampling for RA T cells treated or not with PGE₂.

Metabolomics

Cell lysate from 5 different RA CD8⁺ T cells and sorted CD8⁺ TILs (n = 4) treated or not with 1 μM PGE₂ for 24 h were pre-extracted and homogenized by adding 400 µl of methanol:H₂O (4:1), in the Cryolys Precellys 24 sample Homogenizer (2 × 20 s at 10,000 rpm, Bertin Technologies) with ceramic beads. The bead beater was air-cooled down at a flow rate of 110 l min⁻¹ at 6 bar. Homogenized extracts were centrifuged for 15 min at 4,000g at 4 °C (Hermle). The resulting supernatant was collected and evaporated to dryness in a vacuum concentrator (LabConco). Dried sample extracts were resuspended in methanol:H₂O (4:1, v/v) according to the total protein content.

The protein pellets were evaporated and lysed in 20 mM Tris-HCl (pH 7.5), 4 M guanidine hydrochloride, 150 mM NaCl, 1 mM Na2EDTA, 1 mM EGTA, 1% Triton, 2.5 mM sodium pyrophosphate, 1 mM β-glycerophosphate, 1 mM Na₃VO₄, 1 µg ml⁻¹ leupeptin using the Cryolys Precellys 24 sample Homogenizer (2 × 20 s at 10,000 rpm, Bertin Technologies) with ceramic beads. BCA Protein Assay Kit (Thermo Scientific) was used to measure (A_562 nm) total protein concentration (Hidex).

FFA analysis

Extracted samples were analysed by reversed phase liquid chromatography coupled to high-resolution mass spectrometry (RPLC-HRMS) operating in negative mode using a 6550 Ion-Funnel Q-TOF instrument interfaced with 1290 UHPLC system (Agilent Technologies). Chromatographic separation was carried out on a Zorbax Eclipse Plus C18 (1.8 μm, 100 mm × 2.1 mm internal diameter column) (Agilent Technologies). The mobile phase comprised 60:40 (v/v) acetonitrile:water with 10 mM ammonium acetate and 0.1% acetic acid (A) and 88:10:2 isopropanol: acetonitrile:water with 10 mM ammonium acetate and 0.1% acetic acid (B). The linear gradient elution from 15% to 30% B was applied for 2 min, then from 30% to 48% B for 0.5 min, from 48% to 72% B, and the last gradient step from 72% to 99% B followed by 0.5 min isocratic conditions and a 3 min re-equilibration to the initial chromatographic conditions. The flow rate was 600 μl min⁻¹, with a column temperature of 60 °C and a sample injection volume of 2 µl. Electrospray ionization source conditions were set as follows: dry gas temperature 200 °C, nebulizer 35 psi and flow 14 l min⁻¹, sheath gas temperature 300 °C and flow 11 l min⁻¹, nozzle voltage 1,000 V, and capillary voltage −3,500 V. Full scan acquisition mode in the mass range of 100–1200 m/z was applied for data acquisition.

Statistical analyses and reproducibility

All statistical tests were performed using R (version 3.3.0) and GraphPad Prism software (v8 and v9.3.1; GraphPad). Data points represent biological replicates and are shown as mean ± s.d. Statistical tests to derive P values were performed as specified in the figure legends. Metabolomics statistical analyses were performed using MetaboAnalyst v5.0⁵⁷.

Data were collected using biological replicates to ensure reproducibility. The number of independent replicates for each experiment is noted in all figure legends.

Reporting summary

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