Ethics regulations

Ethics approval for the yellow fever study (number 350_20 B; clinical trial ID: DRKS00034356) was granted by the local ethics committee (Medical Faculty, University Hospital Erlangen, Friedrich-Alexander University Erlangen-Nürnberg, Germany). Samples were collected after written consent had been provided by the donors, and donors received financial compensation. Additional cryopreserved peripheral blood mononuclear cells (PBMCs) from 22 donors vaccinated with the YFV-17D vaccine were provided by S. Rothenfusser (LMU Munich). Approval for this cohort was granted by the review board of the Medical Faculty of LMU (number 86-16), and cohort details are described in the ISRCTN registry (17974967). Animal experiments were approved by the district government of upper Bavaria (Department 5: Environment, Health and Consumer Protection).

Study cohorts

Yellow fever vaccination donors (n = 76) were 20–62 years old (median: 26; interquartile range: 23–31 years), 62% female, of European Caucasian ethnicity, overall healthy (no chronic medication), of normal weight, and received the YFV-17D vaccine (Supplementary Table 1). No previous YFV vaccination or yellow fever infection was reported for any donor. Thirty-one donors had received other vaccinations (rabies, typhoid, meningococcal disease, tetanus/diphtheria/pertussis/polio, cholera, hepatitis A, hepatitis B, rick-borne encephalitis, Japanese encephalitis, COVID-19) between day −14 before and day 13 after vaccination. Nine donors vaccinated with the SARS-CoV-2 mRNA vaccine Comirnaty were included from the previously described CoVa-Adapt cohort³², with blood collected from all donors on day 10 after the second vaccination. HLA typing was conducted at the University Hospital of Erlangen (Supplementary Table 6). PBMCs were isolated from citrated peripheral blood by density-gradient centrifugation using BioColl density medium (BioSell, BS.L 6115). Cells were analyzed directly in fresh whole blood or cryopreserved in fetal calf serum (FCS) + 10% dimethyl sulfoxide (DMSO) in liquid nitrogen.

Multimerization of pHLA monomers

Biotinylated HLA-A*02:01 molecules loaded with LLWNGPMAV peptide (yellow fever NS4B_214–222) or YLQPRTFLL peptide (SARS-CoV-2 spike protein) were generated⁵⁰ and multimerized on a streptavidin backbone conjugated with PE-fluorophores (Life Technologies, 12-4317-87) or BV421-fluorophores (BioLegend, 405225). Per 1 × 10⁶ cells, 0.2 µg pHLA was mixed with 0.125 µg streptavidin-PE or 0.05 µg streptavidin-BV421 in 25 µL FACS buffer (phosphate-buffered saline (PBS) + 0.5% bovine serum albumin) for 30 min (4 °C) directly before staining.

Flow cytometry

The following antibodies were used for human samples: anti-CD3-BUV496 (741206; 1:100), anti-CD4-PE/CF594 (562316; 1:200), anti-CD4-BV786 (740962; 1:400), anti-CD8-BUV395 (563795; 1:200), anti-CD8-BUV496 (612942; 1:200), anti-CD19-PE/CF594 (562294; 1:200), anti-CD56-PE/CF594 (564963; 1:200), anti-HLA-DR-BV421 (562805; 1:400), anti-HLA-DR-APC (560744; 1:200), anti-CD38-BV605 (562666;1: 400), anti-CD38-BUV395 (563812; 1:200), anti-CD95-BUV737 (612790; 1:20), anti-CD95-BV421 (566258; 1:25), anti-Ki-67-BV711 (563755; 1:20), anti-CD69-PE/Cy7 (561928; 1:100) anti-BCL-2-PE (556535; 1:100), anti-yH2AX-PE (562377; 1:20) and anti-HLA-A2-FITC (551285; 1:100) from BD Biosciences; anti-CD8-APC (301049; 1:200), anti-CD4-BV510 (300545; 1:50), anti-CD62L-FITC (304804; 1:200), anti-CD62L-APC/Cy7 (304813; 1:100), anti-CCR7-FITC (353215; 1:100), anti-mouse TCR β chain-APC/Fire 750 (109246; 1:100) and anti-CD45RA-PerCP/Cy5.5 (304121; 1:400) from BioLegend; anti-CD4-PE (12-0049-42; 1:400), anti-CD8-eF450 (48-0086-42; 1:200), anti-CD56-FITC (11-0566-42; 1:200), anti-CD45-PerCP/Cy5.5 (45-0459-42; 1:100) and anti-CD45-PE/Cy7 (25-9459-42; 1:400) from eBioScience; anti-puromycin-AF647 (MABE343-AF647; 1:200), anti-puromycin-AF488 (MABE343-AF488; 1:200) from Sigma-Aldrich; anti-CD8-FITC (A07756; 1:200) from Beckman Coulter; anti-CD45-PB (PB986, 1:50) from DAKO; anti-CD137-PE (130-119-885;1:100) from Miltenyi; and anti-clCasp-3-PE/Cy7 (64772S;1:50) from Cell Signaling Technology. A ZombieAqua or Zombie NIR Fixable Viability Kit (BioLegend; 4231017/423102; 1:500) was used for viability staining.

For murine samples, the antibodies were anti-puromycin-AF647 (MABE343-AF647; 1:200) from Sigma-Aldrich; anti-CD45.1-FITC (110706; 1:100), anti-KLRG1-PE/Cy7 (138416; 1:100), anti-CD27-mCherry (124228; 1:100), anti-CD4-APC/Cy7 (100414; 1:300), anti-CD8-BV785 (100750; 1:200), anti-CD19-APC/Cy7 (115530; 1:300) from BioLegend; and anti-CD62L-BUV737 (612833;1:200) from BD Biosciences. Viability staining was performed using Fixable Viability Dye eFluor-780 (Thermo Fisher, 65-0865-18; 1:1,000).

Staining was performed at 4 °C. PBMCs or murine cells were washed with FACS buffer. When required, cells (1–2 × 10⁶) were incubated with pHLA multimers (25 µl) for 25 min, followed by addition of 25 µl FACS buffer containing surface antibodies and viability stain with further incubation for 20 min. Samples without pHLA multimers were stained identically, except that the multimer incubation was omitted. For intracellular staining, cells were fixed and permeabilized using a BD Cytofix/Cytoperm Kit (BD Biosciences, 554714) for human samples or an eBioscience Foxp3/Transcription Factor Staining Buffer Set (Invitrogen by Thermo Fisher Scientific, 00-5523-00) for murine samples, then incubated for 60 min with intracellular antibodies in PermWash (50 µl per 1 × 10⁶ cells). Cells were analyzed using a LSRFortessa Cell Analyzer (BD Biosciences) with FlowJo v.10.7.2 (Tree Star Inc.). Unless indicated otherwise, we gated on single, living CD19⁻CD56⁻CD4⁻CD3⁺CD8⁺ A2/NS4B⁺ or A2/NS4B⁻ cells.

Metabolic profiling of cryopreserved PBMCs ex vivo

The SCENITH protocol was adapted from Argüello et al.³⁸. Cryopreserved PBMCs were rested overnight (37 °C; 5% CO₂) in cRPMI⁻ (RPMI 1640 Medium (Life Technologies; 21875091), 10% heat-inactivated FCS (anprotec; AC-SM-0027), 0.05 mM β-mercaptoethanol (Life Technologies; 31350010), 1.1915 g l⁻¹ HEPES (Carl Roth; HN77.3), 0.2 g l⁻¹ L-glutamine (Fisher Scientific; 31870025))_. For analysis at day 0–49 after vaccination, 1 × 10⁶ PBMCs (1 × 10⁶ cells ml⁻¹ in cRPMI⁻) were plated into a 24-well plate. For 1-year samples or SARS-CoV-2 samples, 2 × 10⁶ PBMCs per well were plated in a 12-well plate. Cells were rested for 1 h (37 °C; 5% CO₂). Metabolic inhibitors were prepared in PBS to reach the required final concentration: 2-DG (100 mM; Sigma-Aldrich; D6134), oligomycin (1 µM; Sigma-Aldrich; 75351), and harringtonine (2 µg ml⁻¹; Santa Cruz Biotechnology; sc-204771); controls received an equal volume of DMSO in PBS. Cells were treated for 15 min with inhibitors (37 °C; 5% CO₂), followed by addition of puromycin (10 µg µl⁻¹; Sigma-Aldrich; P7255) for 25 min (37 °C; 5% CO2). Cells were washed with FACS buffer, stained for extracellular markers, and stained intracellularly for puromycin and additional markers. BPS and metabolic dependencies were calculated from the geometric mean fluorescence intensity (MFI) of puromycin:

where CO represents the control, Har. is harringtonine and Oligo. is oligomycin. For metabolic dependencies 100, the value was set to 0 or 100, respectively.

Metabolic profiling of in vitro stimulated cryopreserved PBMCs

Cryopreserved PBMCs were rested overnight in cRPMI⁻ (37 °C, 5% CO₂). Then, 0.25 × 10⁶ cells (1 × 10⁶ cells ml⁻¹ in cRPMI⁻) per well were activated in a 96-well F-bottomed plate with plate-bound anti-CD3 (1 µg ml⁻¹; BioLegend, 317302) and anti-CD28 (1 µg ml⁻¹; BioLegend, 302902) for 24, 48 or 72 h (37 °C, 5% CO₂). Metabolic profiling was performed directly in the 96-well F-bottomed plate using SCENITH as described above. Afterward, staining was performed in a 96-well V-bottomed plate as described above.

Metabolic profiling in whole blood ex vivo

First, 5 µl metabolic inhibitor or control was added to 100 µl freshly drawn whole blood to final concentrations of 100 mM 2-DG, 1 µM oligomycin and 2 µg ml⁻¹ harringtonine for 15 min (37 °C, 100 rpm). Puromycin (15 µg ml⁻¹) was then added for 25 min, followed by washing with PBS and surface staining for 20 min (4 °C, 100 rpm). The required volume of surface antibodies and viability dye was prepared to achieve the indicated final concentration in blood samples (without FACS buffer). Erythrocytes were lysed with FACS lysing solution (BD BioScience; 349202), and cells were permeabilized, stained and analyzed as described for cryopreserved PBMCs.

Metabolic profiling of murine cells

Female C57BL/6 mice (CD45.1⁻, 6–8 weeks old) were purchased from Inotiv. OT-I or P14 donor mice (CD45.1⁺) were bred under specific-pathogen-free conditions at the mouse facility of Technische Universität München. Mice were fed a T.2018SMI.12 Global 18% Protein Rodent Diet.

Naive (CD44^lo) CD8⁺ T cells were sorted from the peripheral blood of OT-I or P14 donor mice on a FACS Aria II (Becton Dickinson), and 50,000 (OT-I) or 100,000 (P14) cells were injected intraperitoneally into C57BL/6 recipients. One day after transfer, recipient mice were infected by injecting 5 × 10³ colony-forming units of recombinant OVA-expressing L. monocytogenes (LM) intravenously or 2 × 10⁵ plaque-forming units of LCMV Armstrong intraperitoneally. Blood was sampled on days 6 and 835 (first LM-OVA experiment) or days 0, 6, 8, 10 and 30 (second LM-OVA and LCMV Armstrong experiments) after infection. Lysis was performed with Ammonium chloride-Tris (90% (v/v) 0.17 M NH₄Cl, 10% (v/v) 0.17 M Tris HCl, pH 7.2). Cells were plated in a 48-well plate at approximately 0.3–0.5 × 10⁶ cells (1 × 10⁶ cells ml⁻¹ in cRPMI⁻) per well. Cells were rested for 1 h (37 °C, 5% CO₂) and then SCENITH-treated as cryopreserved PBMCs ex vivo.

Metabolic tracker analysis

Freshly isolated PBMCs (1 × 10⁶ ml⁻¹) were incubated in cRPMI⁻ containing TMRM (0.05 µM; VWR, T5428-25mg) or MitoTracker Green (1:100; Life Technologies, M46750) for 30 min (37 °C, 5% CO₂), stained with pHLA multimers and surface antibodies (as described) and analyzed on an LSRFortessa Cell Analyzer.

Proliferation of CD8⁺ T cell subsets under metabolic perturbation

Cryopreserved PBMCs of healthy donors were rested overnight (37 °C, 5% CO₂) in cRPMI⁺ (cRPMI⁻ plus 0.05 mg ml⁻¹ gentamicin (Life Technologies; 15750060) and 100 U ml⁻¹ penicillin–streptomycin (Life Technologies; 15140122)). Cells were washed, resuspended in PBS (1 × 10⁶ cells ml⁻¹) and labeled with CellTrace Far Red (0.5 µM; Thermo Fisher; C34572) for 20 min (37 °C, dark). Cells were washed, resuspended in cRPMI⁺ and rested for 30 min (37 °C, 5% CO₂). Then, 0.25 × 10⁶ cells per well were stimulated in a 96-well F-bottomed plate with plate-bound anti-CD3 and anti-CD28 (37 °C, 5% CO₂), 50 U ml⁻¹ IL-2 for up to 72 h. Where indicated, metabolic inhibitors were added for the first 24 h (10 mM 2-DG, 0.05 µM oligomycin or 0.1 µg ml⁻¹ harringtonine). Counting beads (123count eBeads; Life Technologies; 01-1234-42) were added to each well 0 h, 24 h, 48 h and 72 h after stimulation. Cells were transferred to a V-bottomed 96-well plate, washed with FACS buffer, and stained for extracellular markers and analyzed as described. For quantification of cell counts, we normalized the acquired cell numbers to counting beads for each sample.

Analysis of sorted CD8⁺ T cell subsets under metabolic perturbation

Cryopreserved PBMCs were rested overnight in cRPMI⁺ (37 °C, 5% CO₂) and washed with FACS buffer, and CD8⁺ T cells were enriched by magnetic cell separation (MACS; Miltenyi Biotec; 130-096-495): 250 × 10⁶ cells were resuspended in 1 ml of MACS separation buffer (MACS buffer); 250 µl biotin–antibody cocktail was added for 5 min (4 °C), and 750 µl of MACS buffer and 500 µL CD8⁺ MicroBeads were added for 10 min (4 °C). Cells were separated according to the manufacturer’s instructions.

CD8⁺ T cells were stained with surface antibodies (CD62L, CD45RA, CD8, CD95) and sorted on a MoFlo Astrios EQ (Beckman Coulter) into CD8⁺ input populations: T_N/T_NM (CD45RA⁺CD62L⁺CD95⁻) input 1); T_SCM (CD45RA⁺CD62L⁺CD95⁺) plus T_CM (CD45RA⁻CD62L⁺) (input 2); and T_EM (CD45RA⁻CD62L⁻) plus T_E (CD45RA⁺CD62L⁻) (input 3). Cells were washed, resuspended in cRPMI⁺ and rested for 1 h (37 °C, 5% CO₂).

Next, 0.2 × 10⁶ cells per well were stimulated in a 96-well F-bottomed plate with plate-bound anti-CD3 and anti-CD28 and 50 U ml⁻¹ IL-2 (37 °C, 5% CO₂). Where indicated, cells were treated for the first 24 h of stimulation with metabolic inhibitors: 10 mM 2-DG, 0.05 µM oligomycin or 0.1 µg ml⁻¹ harringtonine. Before stimulation (0 h sample) or 72 h after stimulation, counting beads were added to each well. Cells were transferred to a V-bottomed 96-well plate, washed with FACS buffer, stained for flow cytometry (as described), fixed using a BD Cytofix Kit and acquired with a Cytek NorthernLights instrument. Cells were pregated on living lymphocytes. For quantification of cell counts, we normalized the acquired cell numbers to counting beads for each sample.

TCR reexpression in Jurkat cells

Six TCRs identified in scRNA-seq were reexpressed in Jurkat TCR-null cells⁵¹. The TCR constructs contained the identified variable regions of the α and β chains and murine constant regions (Supplementary Table 7). RD114 cells (in cDMEM⁻ (DMEM (Life Technologies, 11995073) with 10% heat-inactivated FCS, 0.05 mM β-mercaptoethanol, 1.1915 g l⁻¹ HEPES and 0.2 g l⁻¹ L-glutamine)) were transfected at 60–80% confluence with Lipofectamine 3000 transfection reagent (Thermo Fisher, L3000015) and 2 µg of plasmid DNA according to the manufacturer’s instructions. Cells were rested for 6 h (37 °C, 5% CO₂) and medium was replaced with cDMEM⁺ (cDMEM⁻ with 0.05 mg ml⁻¹ gentamicin and 100 U ml⁻¹ penicillin–streptomycin). Viral supernatant was collected after 48 h and stored at 4 °C. For transduction, 700 µl viral supernatant with 8 µg ml⁻¹ Polybrene was combined with 0.2 × 10⁶ Jurkat TCR-null cells in 200 µl cRPMI⁺, plated in a 24-well plate, centrifuged for 2 h (2,000g; 32 °C) and incubated for 48 h (37 °C, 5% CO₂) before virus was removed. Cells were cultured for another 48 h in cRPMI⁺ (with 50 U ml⁻¹ IL-2; 37 °C, 5% CO₂). Transduction efficiency and TCR specificity were evaluated by staining mTRBC and A2/NS4B-pHLA multimer (as described).

Single-cell RNA sequencing

scRNA-seq was performed on PBMCs from 18 donors across 9 time points after vaccination and PBMCs from 3 vaccination-naive and 5 long-term vaccinated donors preenriched for antigen-specific cells (Supplementary Table 6). Cryopreserved PBMCs were rested overnight (1 × 10⁶ cells ml⁻¹ in cRPMI⁻). Antigen-specific T cells were detected using PE- and DNA-barcoded MHC-I dCODE dextramers (Immudex) targeting eight YFV epitopes, alongside control dextramers for common viral antigens (Supplementary Table 3). Surface protein expression was assessed using CITE-seq antibodies.

Experiments 1 and 2 included time points spanning day 7 to 26 years. Dextramer cocktails were prepared directly before cell staining (all YFV and control virus dextramers regardless of HLA compatibility for experiment 1; only HLA-A2/NS4B_214–222 dextramer for experiment 2 (Supplementary Table 6)). Per 5 × 10⁶ cells, 1 μl of each dextramer and 0.2 μl of D-biotin (100 µM per dextramer) were combined in 50 µl FACS buffer. PBMCs from different donors were color- and DNA-barcoded using anti-CD45 fluorophore combinations (anti-CD45-PacificBlue, anti-CD45-PerCP/Cy5.5, anti-CD45-PE/Cy7) and TotalSeq-C hashtag antibodies (2.5 µl per 5 × 10⁶ PBMCs of TotalSeq-C anti-human hashtag antibodies 1–8; BioLegend: 394661, 394663, 394665, 394667, 394669, 394671, 394673, 394675). Cells were stained for 30 min (4 °C) and washed with FACS buffer, and up to 8 samples with different CD45 and hashtag antibodies were combined. Pooled samples (40–60 × 10⁶ cells) were stained with preprepared dextramer pools (50 µl per 5 × 10⁶ cells) for 30 min (4 °C). Surface antibodies and viability dye (anti-CD19-PE/CF594, anti-CD56-FITC, anti-CD8-APC, anti-CD4-BV510, Zombie NIR) and CITE-seq TotalSeq-C antibodies (per 5 × 10⁶ cells: 0.078 µg anti-human-CD45RA (304163), 0.078 µg anti-human-CD62L (304851), 0.3125 µg anti-human-CD95 (305651), 0.277 µg anti-human-CCR7 (353251), 0.25 µg anti-human-CXCR3 (353251); BioLegend) were added for 30 min (4 °C). Single, living CD19⁻CD56⁻CD4⁻CD8⁺dextramer⁺ lymphocytes were sorted on a BD FACS Aria II cell sorter into FCS-coated 1.5-ml tubes containing FACS buffer. In addition, single, living CD19⁻CD56⁻CD4⁻CD8⁺ lymphocytes irrespective of dextramer signal were added to the sample to provide a general map of CD8⁺ T cells. The donors were distinguished during sorting by CD45 color barcoding to avoid overrepresentation of individual donors.

In experiment 3, samples from YFV-naive or long-term-vaccinated donors were used. Only the HLA-A2/NS4B_214–222 dextramer was used, and this was prepared directly before cell staining (as described). PBMCs (10 × 10⁶ per donor) were collected, stained with dextramer (100 µl) for 30 min (4 °C) and washed with FACS buffer. Dextramer-specific cells were enriched by MACS (Miltenyi Biotec) with anti-PE microbeads according to the manufacturer’s instructions. Enriched cells were stained with surface, CD45 and CITE-seq antibodies and sorted (as described).

Sorted cells were loaded onto a Chromium Next GEM Chip K (10x Genomics) and Chromium Next GEM Single-Cell 5′ Kits (v.2) to generate gene expression (GEX), TCR (VDJ) and cell surface libraries (10x Genomics; 1000263, 1000256, 1000252, 1000286, 1000250, 1000215, 1000190). Libraries were sequenced at Novogene (Cambridge, UK) on an Illumina NovaSeq platform with the PE150 strategy.

scRNA-seq data analysis

The dataset comprised results from three experiments, run across nine sequencing lanes. Processing was performed per lane using Cell Ranger Multi (cellranger-7.1.0, 10x Genomics) with GRCh38 for gene expression (v. 2020A, 10x Genomics), vdj-GRCh38 for VDJ (v. 5.0.0, 10x Genomics) and custom feature barcode references for surface antibody detection.

Single-cell analyses were performed using Scanpy (v.1.10.1)⁵² and Scirpy (v.0.14.0)⁵³. For each sequencing run, the gene expression and antibody capture matrices were merged with TCR contig annotations, and filtering for doublets and dying cells was applied, based on UMI counts, detected genes and mitochondrial fractions (Supplementary Table 8).

Gene expression data were normalized to 10,000 counts per cell and log1p-transformed. Donor and time point assignments were performed with HashSolo⁵⁴. Samples were integrated, batch-corrected using combat, and cells without annotated TCRs were excluded. Analysis was based on the top 5,000 highly variable genes (excluding TCR genes). UMAP⁵⁵ embeddings were computed using 15 neighbors, and Leiden clustering⁵⁶ was performed at a resolution of 1. Differential gene expression was assessed using a t-test with Benjamini–Hochberg correction via the scanpy.tl.rank_genes_groups function.

Diffusion pseudotime was determined by scanpy.tl.dpt (root cell in cluster 4). Surface protein expression data were transformed using centered log-ratio normalization. Virtual gating was based on centered log-ratio-transformed values for CD45RA, CD62L and CD95, with thresholds of 1.3, 1.6 and 1.0, respectively.

Clonotypes were defined as having identical α- and β-CDR3 amino acid sequences on either the primary or secondary chain. Clonal expansion was assessed across the dataset. For experiment 1, a bivariate Gaussian distribution was used via the sklearn package (v.1.5.0)⁵⁷ to distinguish dextramer binding and nonbinding cells based on UMI counts and cell purity (proportion of epitope-specific UMIs among total UMIs). In experiments 2 and 3, cells were stained only with the HLA-A2/NS4B_214–222 dextramer, so purity metrics were unavailable. Instead, an in-house prediction package was established and applied (Supplementary Methods). A clone was considered to be epitope-specific if >60% of its cells were predicted to be epitope-specific in HLA-matched donors.

Visionpy (v.0.2.0)⁵⁸ (https://github.com/YosefLab/visionpy) was used for pathway analysis. Pathway annotations for KEGG legacy and GOBP were obtained from MSigDB (https://www.gsea-msigdb.org/gsea/msigdb) were supplemented with manually curated pathways (Supplementary Table 9). Weighted transcript levels were calculated using Visionpy⁵⁸. Scanpy (scanpy.tl.rank_genes_groups) was used to identify differentially active pathways between clusters. For visualization, mean scores per condition were z-scored. Transcription factor network activities were assessed using a univariate linear model on the ‘collectri’ library in Decoupler (v.1.8.0)⁵⁹.

Microarray gene expression data for different T cell subsets (T_N, T_SCM, T_CM and T_EM) from Gattinoni et al.³⁰ (GEO: GSE23321) were processed in R (v.4.5.1). The CEL files were imported and normalized with oligo (v.1.72.0) using array-specific package pd.hugene.1.0.st.v1 (v.3.14.1). Probes were annotated using hugene10sttranscriptcluster.db (v.8.8.0) and AnnotationDbi (v.1.70.0). Unmapped probes and duplicate genes were removed with dyplr (v.1.1.4), and the resulting expression matrix was used to determine differentially expressed genes in Python (v.3.13.5) with statsmodels (v.0.14.5). From this information, gmt files for the various subsets were generated for investigation using visionpy.

CITE-seq marker analysis

To determine which markers had the greatest potential to identify T_N/T_NM cells, we analyzed a previously published dataset³² containing CD8⁺ T cells stained with 130 CITE-seq antibodies. The T_N/T_NM Leiden cluster was identified based on marker genes, and the cells were annotated as naive. For each CITE-seq marker, we then determined the optimal signal cutoff for identification of naive cells by calculating positive and negative predictive values for each marker.

Statistics and reproducibility

No statistical methods were used to predetermine sample sizes, but our sample sizes were based on results from previous publications^1,3,28. The study was nonrandomized, and all consenting donors were included and analyzed. No blinding was performed, as the purpose of the study was not to perform a comparison between different donors but to provide methodological proof-of-concept in multiple donors. Samples were pseudonymized using study identification numbers. We excluded T cell subsets with fewer than ten cells from downstream analyses (exact donor numbers per panel are described in Supplementary Table 2).

Data analysis and visualization

Data graphs were generated with GraphPad Prism 10. Data were tested for normal distribution, then the appropriate statistical test was chosen. In all graphs, only statistically significant results are highlighted. For statistical testing of the timelines in Fig. 5d and Extended Data Fig. 8h, we used two-way analysis of variance with Tukey’s multiple comparison post hoc test, although the dataset contains samples from the same donor taken at different time points across the timeline and from donors taken at an individual time point. The test was performed for one subset against all A2/NS4B⁺CD8⁺ cells as indicated in the graph. Schemes and figures were generated with Affinity Designer (Serif (Europe) Ltd, v. 2.5.3).

Reporting summary

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