A total of 59 healthy individuals (age 21–68) were included in the INHALE study between March and April 2023. Participants were eligible if they were between 18 and 75 years old, able to understand English or Danish, free of exercise-limiting or unstable medical conditions, free of autoimmune diseases requiring active treatment and excluded if using medications affecting cardiovascular, metabolic, or immune responses (e.g., >10 mg prednisone equivalents, immunosuppressive medications or beta blockers). Written informed consent was obtained. Health status and medication, and demographic, lifestyle, and habitual physical activity data were collected using the International Physical Activity Questionnaire (IPAQ). Participants were then selected to be HLA-A2 positive to maximize detection of virus-reactive CD8⁺ T cells restricted by the applied pMHC multimer panel (see below). HLA-A2 status was determined by flow cytometry analysis (see below) of capillary EDTA blood (Sarstedt); 27 (46.6%) were positive, of whom 23 (10 females, 13 males) aged 25–65 continued (Table 1). Study design is shown in Figure 1A. The study is registered at clinicaltrials.gov (NCT05826496) with ethics approval from Capital Region’s Ethics Board (H-23006672) and Danish Data Protection Agency (P-2023-95).

HLA-A*02 positive participants completed a CPET on a cycling ergometer (CPET Vyntus CPX, Ergometer ViaSprint 150p, Vyarie Medical) following 24 hours of avoiding alcohol, non-prescription meds, vigorous/unaccustomed exercise, and 2 hours caffeine fasting. Based on estimated aerobic capacity appropriate incremental workload protocols were chosen aiming for a test duration of 8–12 minutes and terminated at volitional exhaustion (25). Ventilation, heart rate (HR), and respiratory exchange rate (RER) were recorded every 15 s; VO2 peak, maximal power output, and HR were calculated as the average of a 30s period based on 2 subsequent measurements (25) and confirmed visually by three independent researchers. Anthropometric and body composition measurements by bioimpedance (SoZo, ImpediaMedia) were performed at CPET.

Within 24 days post-CPET, participants completed a physiotherapist-supervised, group-based exercise session after similar abstinence and fasting conditions, which took place during daylight between 8:00 and 14:00 to minimize bias by circadian rhythm. Baseline blood samples were collected by venipuncture at arrival. The warm-up included 5 minutes on a bicycle ergometer followed by 5 × 2-minute circuit training on: 1) air bike ergometer, 2) ski ergometer, 3) rowing ergometer, 4) cross trainer, and 5) step bench exercises. All warm-up exercises were self-paced but overseen by an experienced physiotherapist encouraging the participants to exercise with sufficient intensity to substantially increase HR and respiration. Bicycle training (Motion Cross 600, Emotion Fitness) started with 2 minutes low workload pedaling (approximately 50% of the maximal power output at CPET), then 3 × 3-minute high-intensity intervals alternating 20 s “all-out” (aiming at 100% of the maximal power output at CPET) with 20 s easy pedaling (aiming at 50%). These intervals were separated by two 3-minute steady-state sequences at 50-70% of the maximal power output during CPET. Resistance and cadence were pre-set to meet or exceed target workloads. Blood samples were taken at baseline (bsl), within 2 minutes post-exercise (ex02), and after 60 minutes (ex60) (Vacutainer Sodium Heparin, BD; Vacuette Serum, Greiner bio-one). Only water was consumed until ex60.

Samples were processed immediately at the sterile laboratory. Serum was obtained by centrifugation at 1300 g for 10 min, aliquoted within 1 h, and stored at –80°C. PBMCs were isolated using density gradient centrifugation in Leukosep tubes (Greiner bio-one) and counted. PBMCs were then cryopreserved in inactivated human AB serum (Sigma Aldrich) with 10% DMSO (Honeywell), and stored at –150°C. HLA genotyping was performed by DKMS Life Science Lab (Dresden, Germany) using amplicon sequencing.

For HLA-A2 staining, in short, 50µl whole blood were incubated with 2µl antibody (20 minutes, dark, room temperature), afterwards 2ml BD lysis buffer (BD Biosciences) were added. After 10 minutes, cells were resuspended in FACS buffer and acquired on the NovoCyte Quanteon (Agilent).Whole blood was stained with a 6-color TBNK panel in Trucount tubes (Supplementary Table S1) according to the manufacturer’s protocol and acquired on a BD FACSCanto II. Gating was standardized (FACSCanto Software, BD). Expression of CCR2, CCR5, CXCL2, CXCL4, CXCL6 and CX3CR1 was analyzed in cryopreserved PBMCs (antibodies listed in Supplementary Table S1). Cells were washed twice with PBS/2% FCS (FACS buffer), stained with chemokine receptor antibodies at 37°C for 15 min, then with antibodies targeting CD3, CD4, CD8, CD28 and CCR7 at 4°C for 30 min. Uniform Manifold Approximation and Projection (UMAP) analysis was performed on manually gated CD8⁺ T cells, using 5,000 events per sample, 15 neighbors, and a minimum distance of 0.5. The following markers were included for UMAP generation: CD28, CCR7, CX3CR1, CXCR4, CXCR2, CXCR6, CCR2, and CCR5. Unsupervised clustering was then conducted with the FlowSOM algorithm (grid size: 10 × 10) using the same markers, resulting in five distinct clusters.

Cells were acquired on a NovoCyte Quanteon (Agilent). Gating was guided by FMO controls and analyzed with Novoexpress v1.5.0 (Agilent) or FlowJo v10 (Tree Star).

Serum samples were thawed and analyzed immediately using the 2-CAT ELISA Fast Track kit for Norepinephrine and Epinephrine (LDN). Absorbance at 450nm wavelength was determined using Epoch plate reader and analyzed with Gen5 Take3 software (both Agilent). Concentrations were interpolated from standard curves according to the manufacturer’s protocol.

Thawed PBMCs were rested overnight in X-Vivo 15/5% human serum. 3–5 × 10⁵ PBMCs were plated in IFN-γ Ab-coated ELISPOT plates and stimulated in triplicates with 5 µM of GLCTLVAML (EBV, Schafer-N) peptide for 20–24 h. After washing, biotinylated IFN-γ Ab, streptavidin-AP, and BCIP/NBT substrate (all Mabtech) were applied sequentially. Spots were analyzed on a CTL ImmunoSpot S6 Ultimate-V using ImmunoSpot v5.1 (both ImmunoSpot). Responses were normalized to 5 × 10⁵ PBMCs and reported as the difference between average numbers of spots in wells containing peptides and negative control wells.

Peripheral blood mononuclear cells (PBMCs) were screened for virus-reactive CD8⁺ T cells (VARTs) using a panel of 250 unique HLA-A and B restricted peptides from 20 different viruses, which were incorporated into barcode-labeled pMHC-multimer complexes (26). The peptides are listed in Supplementary Table S2. Specifically, pMHCs and unique DNA barcodes were attached to Phycoerythrin (PE) fluorochrome-labeled dextran molecules (Fina Biosolutions), generating unique DNA barcode-labeled pMHC I multimers for each pMHC combination. The cells were then stained with a combination of these multimers, along a phenotype antibody panel including CD3, CD8, CD45RA, CCR7, CD57, CD95 and PD-1 (Supplementary Table S1). PE-labeled CD8⁺ T cells were subsequently sorted using a FACSAria flow cytometer (BD Biosciences). DNA barcodes from the sorted cells were amplified via PCR, along with a reference DNA barcode baseline sample from the pMHC multimer panel used for staining. The amplified barcodes, including those from the sorted cells and the baseline, were sequenced by PrimBio. The sequencing data were then uploaded to Barracoda for analysis (26), including additional information on primers, DNA barcodes, pMHC barcode annotations, and sample identification. Analysis was performed with FlowJo v10.

This study included all samples (n = 23 participants; timepoints bsl, ex02, ex60), unless otherwise stated. As this was an exploratory study, no formal power calculations were performed. Analyses were advised by two independent bioinformaticians/data scientists. Data were analyzed and visualized using GraphPad Prism v10 or R v4.3.0 with RStudio 2024.04.2 + 764, Figures were created in Inkscape 1.3.2. Normality was assessed by histogram inspection or Shapiro-Wilk test. Significance between three groups was tested using one-way ANOVA, unless otherwise stated. Continuous variables are presented as mean ± SD unless otherwise noted. Correlations were performed using Spearman’s method unless otherwise specified. Multiple-testing correction was applied for appropriate ANOVA post hoc analyses and for correlation analyses summarized in heatmaps (Holm adjustment). No correction was applied across figures or panels, as these represent independent, pre-specified biological endpoints. Further details are provided in figure captions.

A total of 24 HLA-A2+ participants were included in the INHALE study, of which 23 participants conducted the full exercise intervention, samples and tests (see study design in Figure 1A). Participants had a mean age of 37.3 years (25–65), with 43.5% females, further characteristics are presented in Table 1. No adverse events occurred. VO2 peak (25) reflected a healthy cohort with variable fitness levels (Supplementary Figures S1A, B), and maximal power output and HR relative to the maximal values at CPET confirmed high exercise intensity during HIIT (Supplementary Figures S1C-F). Analysis of peripheral blood taken as part of the acute HIIT confirmed profound mobilization of T cells, which increased 2.0-fold (1.5-2.9, p<0.0001) (Supplementary Figure S1G), and CD8+ T cells, which increased 2.6-fold (1.8-4.1, p<0.0001) (Figure 1B). At ex60, counts of CD8+ T cells declined below baseline (Figure 1B). We also observed that higher CD8+ than CD4+ mobilization lowered the CD4/CD8 ratio at ex02, with a subsequent rise at ex60 (Supplementary Figure S1H). Interestingly, both catecholamines norepinephrine (NE) (Supplementary Figure S1I) and epinephrine (EPI) peaked at ex02 (Supplementary Figure S1J), however CD8+ mobilization selectively correlated with NE (R = 0.531, p= 0.0091) (Figure 1C), but not with EPI (R = 0.097, p=0.66) (Figure 1D). Notably, robust CD8+ T cell mobilization was observed consistently across participants regardless of VO2 peak, sitting time, BMI, lean body mass, body fat mass, or alcohol consumption (Figure 1E) and did not differ between male and female participants. Also, mobilization occurred independent of exercise parameters (Supplementary Figure S2A). This held also true for CD8+ T cell egress, which was independent of individual factors (Supplementary Figure S2B). Overall, we only observed a correlation between CD8+ T cell mobilization and NE.

We characterized the exercise-mobilized CD8+ T cells by their differentiation status (27, 28). The composition of memory subsets based on staining with CD45RA and CCR7 (28) shifted at ex02 in favor of terminally differentiated effector memory (TEMRA) (p<0.0001) and effector memory (EM) (p<0.0001) with a concomitant decrease in naïve CD8+ T cells (p<0.0001) and central memory (CM) (p<0.0001) (Figure 2A). CD8+ T cell mobilization strongly correlated with the mobilization of TEMRA and EM subsets, suggesting that they constitute the majority of exercise-responsive CD8+ T cells (Supplementary Figures S2C, D). At ex60, naïve CD8+ T cells had increased beyond bsl (p=0.0330) as did CM (p=0.0096) accompanied by decrease in TEMRA (p=0.0030) (Figure 2A). Investigation of the association between mobilization of these subsets and individual participant factors revealed that VO2 peak was inversely correlated with mobilization of EM and PD-1+ CD8+ T cells (Figure 2B). To this end, lean body mass positively correlated with mobilization of naïve CD8+ T cells, while a negative correlation was observed with mobilization of TEMRA, EM, CD57+, CD95+ and PD-1+ CD8+ T cells (Figure 2B). In line with the memory subset dynamics, one acute HIIT induced a higher percentage of CD57+ (p<0.0001) (Figure 2C), CD28neg (p<0.0001) (Figure 2D), CD95+ (p<0.0001) (Figure 2E) and PD-1+ (p=0.0001)(Figure 2F) circulating CD8+ T cells, representing higher percentages of T cells expressing markers of late differentiation, indicative of both senescent (CD57+, CD28neg) and exhausted (CD95+, PD-1+) CD8+ T cells phenotypes in the peripheral blood (summarized in Supplementary Figure S2E). This increase was transient, resulting in lower percentages of CD57+ (p=0.0050), CD28neg (p=0.0055) and PD-1+ (p=0.0040) CD8+ T cells at ex60 compared to bsl (Figures 2C, D, F). In addition to the percentage, the median fluorescence intensity (MFI) of PD-1+ CD8+ T cells was decreased (Supplementary Figure S2F). This suggests, that acute HIIT induces transient mobilization of late-differentiated CD8+ T cells followed by a recovery of naïve and less-differentiated cells above bsl.

To investigate the impact of acute HIIT on virus-reactive T cells (VART) a large-scale analysis was performed using DNA barcode-labeled virus peptide-MHC multimers (26) (peptide and derived virus list in Supplementary Table S2). Representative stainings of virus multimer-binding T cells are shown in Supplementary Figure S3A. Overall, VARTs were detected for 58 peptides derived from 12 viruses with consistent exercise-induced mobilization patterns across all viruses, both latent herpesviruses or lytic viruses (Figure 3A). First, the number of VARTs recognizing one or more virus peptides was analyzed: Exercise-mobilized VARTs significantly increased at ex02 (p<0.0001) and declined below baseline at ex60 (ex60 vs. ex02 p<0.0001, ex60 vs. bsl p=0.0090) (Figure 3B). To discriminate the predominant mobilization of VARTs from a purely quantitative mobilization effect, we analyzed the estimated frequencies, which confirmed preferential mobilization of VARTs following acute HIIT (Figure 3C). We then quantified VARTs for the two viruses being reactive in the most participants (Supplementary Table S3) and at the same time the highest number of recognized peptides (Figure 3A): EBV and SARS-CoV-2. 21 participants exhibited quantifiable EBV peptide-specific CD8+ T cells at bsl, 22 at ex02, and 20 at ex60 (Supplementary Table S3) in response to a total of 19 EBV-derived peptides (Figure 3A). Acute HIIT significantly mobilized EBV-reactive CD8+ T cells (p=0.0003), followed by a marked decline (ex60 vs. ex02 p<0.0001) (Figure 3D). The findings were supported by IFN-γ ELISPOT analysis to the dominant EBV peptide and showed mobilization of a highly functional CD8+ subset capable of IFN-γ secretion: 2 participants showed enhanced, 1 showed an IFN-γ response only detectable after exercise (ex02 or ex60), 1 participant showed a strong response at both bsl and ex02 (ex60 sample not available), and 1 participant showed no response (Figure 3E). SARS-CoV-2 peptides elicited responses in 21 participants at bsl, 22 at ex02, and 20 at ex60 (Supplementary Table S3), frequencies (Figure 3C) and cell counts were minimally lower compared to EBV (Figure 3F). Memory subset analysis of multimer⁺ CD8⁺ T cells at bsl revealed a predominance of TEMRA and EM with minor CM and naïve fractions (Figure 3G). Overall, acute HIIT induced only minor shifts reflecting a homogeneous multimer+ CD8+ T cell population: At ex02, a shift toward TEMRA (p=0.0003) with reduced naïve (p=0.0002) and CM (p=0.0110) mirrored the pattern of total CD8⁺ T cells but was less pronounced. At ex60, CM had increased compared to bsl (p=0.0345) while other subsets remained largely unchanged (Figure 3G). Furthermore, the fractions of CD95+ and PD-1+ were unchanged by acute HIIT, while the CD57+ population was reduced at ex60 vs. ex02 (p<0.0001) (Supplementary Figures S3B-E). The mobilization of VARTs occurred independent of individual demographic, anthropometric and cardiorespiratory fitness parameters (Figure 3H).

Finally, after mobilization of CD8+ T cells, their cytotoxic functionality also depends on egression to relevant tissues. To this end, we investigated if acute HIIT causes a dynamic shift in chemokine receptor patterns of the circulating CD8+ T populations (Figure 4A). We identified CD8+ CD28neg CCR7neg T cells, corresponding to late-differentiated antigen-specific CD8+ T cells (29), which are characterized by uniquely high expression of CX3CR1 and medium expression of CXCR2 (pop1) to be preferentially mobilized by acute HIIT before decreasing below bsl (Figures 4B, C). Population 4 (pop4), a CD8+ CD28+ CCR7neg T cell population characterized by high CCR5 expression with medium CX3CR1, CXCR2 and unique CCR2 expression, is preferentially mobilized (Figures 4B, C) and corresponds to a population in the process of differentiation with enhanced tissue homing capacities (29). In contrast, the percentage of population 3 (pop3) as characterized by CD28+ CCR7+ CXCR4+ and otherwise low chemokine receptor expression below median (CCR2, CXCR2, CX3CR1, CXCR6 and CCR5) corresponds to a naive or early-differentiated population, decreased at ex02 and was above bsl at ex60 (Figures 4B, C). At ex60, a CD8+ CD28+ CCR7low population (pop5) had increased above bsl frequencies. Single UMAP plots illustrating the expression of distinct chemokine receptors are shown in Supplementary Figures S4A-H. Changes of the frequencies of individual circulating chemokine receptor-positive CD8+ T cells are shown in Supplementary Figures S4I-N. Elevated MFI of CXCR2 and CXCR6 suggests an increased chemotactic sensitivity (Supplementary Figures S4O-T). Age is the only variable influencing any of the chemokine receptor expressing subsets (data not shown). Chemokines specific to the analyzed receptors were strongly regulated by acute HIIT: Proteomic analyses revealed regulation of 9 of 11 analyzed ligands by acute HIIT, with 4 being significantly increased at ex02: CX3CL1 (Figure 4D), CCL2 (Figure 4E), CCL13 (Figure 4F) and CXCL12 (Figure 4G). These data show that exercise modifies chemokine receptor expression on CD8+ T cells and their ligands, suggesting a potential impact on the migratory capacity of immune cells.