Exercise Training Improves Tumor Control by Increasing CD8⁺ T-cell Infiltration via CXCR3 Signaling and Sensitizes Breast Cancer to Immune Checkpoint Blockade

It is now recognized that exercise helps prevent cancer, reduces morbidity associated with cancer therapy, and enhances overall survival in patients with breast cancer (1–4). Because exercise is a homeostatic physiologic intervention, it also affects the tumor microenvironment (TME; refs. 5–10). However, how exercise modulates host immune responses against cancer or in combination with immune modulators is not fully understood.

Preclinical studies indicate that tumor perfusion and oxygenation increase from the very first exercise session (5, 9, 11). Because the tumor vasculature is immature (12) and lacks myogenic and autonomic regulation (5), any transient increase in mean arterial pressure, as seen during exercise, increases tumor perfusion (13). Beyond the acute increase in perfusion, exercise training (ExTr; accumulated bouts of exercise sessions) induces vessel normalization, thus enhancing tumor blood flow and oxygenation in subsequent resting conditions as well (6, 7). As a consequence, the TME becomes less hypoxic, rendering tumors less aggressive (14). This may partly explain the association of ExTr with reduced tumor burden (6, 8) and suggests ExTr as a form of vascular normalization therapy. More importantly, a normalized tumor vasculature and microenvironment reprograms the immunosuppressive TME and enhances antitumor immunity (15). An improved vascular network favors adaptive antitumor immunity by promoting trafficking and relocalization of tumor-infiltrating lymphocytes (TIL), leading to improved outcomes of various immunotherapies, including immune checkpoint blockade (ICB; refs. 15–20).

Limited T-cell recruitment and function in the TME correlate with poor prognosis in breast cancer (21, 22). Among TILs, CD8⁺ T cells predict overall treatment response and overall survival in patients with breast cancer (23–25). A varying degree of infiltrating CD8⁺ T cells express multiple immune checkpoints [e.g., coexpression of programmed cell death protein 1 (PD-1) and cytotoxic T-lymphocyte–associated protein 4 (CTLA-4)], which cause these cells to become exhausted (26). ICB mitigates CD8⁺ T-cell exhaustion by targeting these receptors and is approved for a subset of patients with triple-negative breast cancer (27). An abnormal TME, caused in part by hypoperfusion, results in the exclusion of CD8⁺ T cells and reduces the impact of immunotherapies such as ICB. We previously demonstrated that vascular normalization could increase immune cells' infiltration into tumors, and multiple studies have confirmed these findings in several tumor models (15, 17–19).

Here, we investigated the effect of ExTr on the tumor vasculature, TME, and TILs in established orthotopic breast cancer models. We found that ExTr normalized the tumor vasculature, increased CD8⁺ T cells in primary breast cancer tumors, and delayed tumor growth. Mechanistically, we demonstrated that the antitumor effect of ExTr was dependent on CD8⁺ T cells and that the CXCL9/11–CXCR3 pathway mediated the ExTr-induced tumoral infiltration of CD8⁺ T cells and the observed antitumor effect. We also demonstrated that ExTr reprogramed the TME, enhanced infiltration and function of CD8⁺ T cells, and sensitized refractory breast cancers to combined PD-1 and CTLA-4 blockade. These findings highlight the potential role of ExTr as a physiologic strategy to revive antitumor immunity and improve the clinical outcomes of ICB therapy in breast cancer.

E0771 (CH3 BioSystems) and EMT6 (ATCC) cells were cultured in RPMI-1640 medium with 10% FBS (Gibco). MMTV-PyMT cells from transgenic breast cancer mouse models, as well as the MCa-M3C cell line (derived clusters from MMTV-PyMT tumors that spontaneously colonized the lungs; cells were isolated by Dr. Peigen Huang and established as a metastatic breast cancer cell line) were cultured in DMEM medium with 10% FBS supplemented with 4.5 g/L glucose (Corning). All cells were cultured at 37°C in a humidified incubator with 5% CO₂. Before starting the study, the cells were expanded and frozen in single vials (containing 10⁶ cells in 50% FBS/10% DMSO, Sigma-Aldrich), so vials from the same batch would be thawed anytime we started a new mice cohort. These cells were tested and authenticated (morphology check by microscopy, and antibodies for common viral contaminations, VRL Diagnostics). The cell lines were also tested negative for mycoplasma. Aliquots were thawed and implanted in mice after three (no more than five) passages within 5 to 7 days.

Syngeneic, orthotopic breast tumors were established by injecting 50 μL of media containing breast cancer cells in the third mammary fat pad of 8- to 10-week-old female C57BL/6 (E0771, 2 × 10⁵), FVB (MCa-M3C and MMTV-PyMT, 1 × 10⁵), and Balb/c (EMT6, 2 × 10⁵) mice. All these strains were bred and maintained in Cox-7 gnotobiotic animal facility. C57BL/6 mice (8 weeks old) were subjected to an ad libitum high-fat diet (HFD; cod. D12492, 60% fat) or low-fat diet (LFD; cod. D12450B, 10% fat; both from Research Diets) starting 12 weeks before tumor implantation, and throughout the study. After 3 months of HFD/LFD, 10⁶ E0771 cells were implanted in the third mammary fat pad, as previously described in ref. 14. Cxcr3^−/− mice were obtained from Jackson Laboratory and were in the C57BL/6 background. For tumor implantation and subsequent collection, mice were anesthetized for 30 minutes using intramuscular injections of ketamine/xylazine (K/X; 90 mg/9 mg per kg body weight; Patterson Vet Supply). Tumor dimensions were initially measured every 3 days by both ultrasound (Vevo 2100, FUJIFILM VisualSonics) and calipers after treatment initiation. Tumor volume (in mm³) was estimated using the formula: tumor volume = (long axis) × (short axis)² × π/6. When tumors reached approximately 100 mm³ (around 10 days for E0771 and 20 days for MCa-M3C), we started neoadjuvant intensity-controlled, moderate-to-vigorous aerobic ExTr treatment alone or in combination with ICB as described in the following sections. All animal procedures were carried out following the Public Health Service Policy on Humane Care of Laboratory Animals and approved by the Institutional Animal Care and Use Committee of Massachusetts General Hospital.

For patients with cancer, the current exercise recommendation follows the guidelines from the American College of Sports Medicine (ACSM; ref. 28), which is endorsed by the American Cancer Society (29) and similar to the World Health Organization (30) for general health. It states that cancer survivors (after a cancer diagnosis) should engage in moderate-to-vigorous intensity (rate at which the exercise is performed) exercise 3 to 5 or more days per week for 30 to 60 minutes per session. The intensity should be expressed as relative to the maximum amount an individual can perform. The goal is to identify the body's maximal capacity for uptake, transport, and use of oxygen, defined as the maximal oxygen uptake (VO₂ max). ACSM considers moderate-intensity exercise ranging from 46% to 63% and vigorous 64% to 90% of VO₂ max (28). In mice, similar physiologic responses to graded exercise have been demonstrated (31, 32). Thus, we measured the maximal oxygen capacity in mice to determine a proper prescription of the ExTr.

One week before tumor implantation, mice were subjected to acclimatization on a running treadmill (5 lanes with shock grid, Panlab/Harvard Apparatus). The acclimatization consisted of running on the treadmill at a low velocity (10 m/min) for 10 minutes on 3 alternate days. Once mice were familiarized with the treadmill, we performed a maximal exercise test, which is correlated with and predicts VO₂ max. The test started at 6 m/min with 3 m/min increments every 3 minutes until volitional exhaustion with 0° inclination. At the exhaustion point, mice were rapidly removed from the running treadmill, and the time (s), distance (m), and velocity (m/min) were recorded. Intensity-controlled, moderate-to-vigorous aerobic ExTr began when tumors reached approximately 100 mm³ and consisted of daily sessions at approximately 60% of maximal velocity in clusters of 5 mice per time (the number of lanes in the treadmill). In a cohort of MMTV-PyMT tumor–bearing mice, ExTr started when tumors were at a smaller size (∼20 mm³). ExTr began with 30 minutes in the first session, with consecutive increments of 5 minutes until reaching 45 minutes per day, which was sustained throughout the ExTr protocol. Compliance was accessed by the percentage of completion of the prescribed ExTr program. We excluded mice that did not reach at least 75% of prescribed ExTr after randomization, representing approximately 20% of C57BL/6 mice. All FVB mice completed the study. A 7-day ExTr protocol was used for analyzing TME parameters, and a 14-day protocol was used for tumor growth endpoints.

CD8⁺ T cells were depleted by administering i.p. injections of 250 μg of anti-CD8β (53–5.8, BioXcell) or IgG. Concurrently with the ExTr protocol, the depletion protocol started when tumors reached 100 mm³ by injecting every 4 days (days 1, 5, 9, and 13) until tumors in the control group reached 1 cm³.

ICB with anti–PD-1 alone (200 μg, RMP 1–14, BioXcell), anti–PD-1 combined with anti–CTLA-4 (100 μg, 9D9 BioXcell), or IgG was administered i.p. and readministered every 4 days (days 1, 5, 9, and 13, a total of 4 doses). ICB treatment started concomitantly with ExTr protocol when tumors had reached 100 mm³. The endpoint was determined by the time when the control group reached 1 cm³.

Twenty-four hours after the last exercise session, mice were anesthetized with K/X (90 mg/9 mg per kg body weight), and breast tumors were harvested, minced, and incubated for 1 hour at 37°C with DMEM media containing collagenase type 1A (1.5 mg/mL), hyaluronidase (1.5 mg/mL), and DNase (20 μg/mL). The digested sample was filtered through 70-μm strainers and centrifuged at 1,200 rpm for 5 minutes. The supernatant was discarded, and cells were resuspended in 4 mL of ACK lysing buffer (Lonza) for about 4 minutes. Six milliliter of flow cytometry buffer (PBS with 3% BSA and 0.05% sodium azide) was first added, cells were then centrifuged, and the pellet (∼2 million cells per panel) was resuspended (150 μL). An aliquot from each sample was collected for counting cells in a hemocytometer. To detect cytokine production (IFNγ, granzyme B, and CXCL9), we additionally stimulated cells for 4 hours at 37°C with cell activation cocktail (phorbol 12-myristate 13-acetate/ionomycin and Brefeldin A, BioLegend, 1:100). Single-cell suspensions were incubated with Fc block (anti-CD16/CD32 antibody for 10 minutes at 4°C, BioLegend, 1:500) and then stained with cell surface–conjugated antibodies. The stained cells were washed and resuspended in 100 μL of cold flow cytometry buffer (1% BSA, 5% FBS, 0.1% sodium azide in PBS; salts from Sigma-Aldrich). Flow cytometry data were obtained using an LSRII flow cytometer (BD Biosciences) and analyzed with FlowJo software. The double/aggregated cells were gated out using forward scatter area versus forward scatter width and side scatter area versus side scatter width. No viability dyes were used. Different fluorophores conjugated with the following mAb were used: CD45 (30-F11, BioLegend), TCRβ chain (H57–597, BioLegend), CD4 (RM4–5, BioLegend), FOXP3 (150D, BioLegend, 1:100), PD-L1 (10F.9G2, BioLegend), CD8a (53–6.7, BioLegend), PD-1 (29F.1A12, BioLegend), Tim3 (B8.2C12, BioLegend), CTLA4 (CD152, BioLegend), LAG3 (C9B7W, BioLegend), IFNγ (XMG1.2, BioLegend), Granzyme B (GB11, BioLegend), Ki67 (16A8, BioLegend), CXCR3 (173, BioLegend, 1:100), CD11b (M1/70, BioLegend), Ly6G (1A8, BioLegend), F4/80 (BM8, BioLegend), MHCII (M5/114.15.2, BD biosciences), CD206 (MMR, BD Biosciences), NK1.1 (PK136, BD biosciences), and CXCL9 (MIG-2F5.5, BioLegend). Unless specified otherwise, all the antibodies were used at 1:200. Representative dot plots and gating strategies can be found in the Supplementary Information. Markers used to define cell populations are in their respective figures.

Twenty-four hours after the last exercise session, mice were injected with pimonidazole hydrochloride (60 mg/kg; Hypoxyprobe) i.p. 1 hour before sacrifice. Mice were then anesthetized, and biotinylated lectin (Vector Labs) was administered in the retro-orbital sinus (in 100 μL of 1 mg/mL) 5 minutes before breast tumor removal. Harvested tumors were then fixed in 4% formaldehyde in PBS for 6 hours and incubated with 30% sucrose in PBS overnight. Tumors were then mounted in optimal cutting temperature gel (Tissue-Tek). Transverse 20 μm tumor sections were stained with antibodies for CD31 (Millipore, MAB1398Z, 1:100), Cy3-αSMA (Sigma, C6198, 1:100), lectin (Strepavidin-Alexa546, Invitrogen, S11225, 1:200), Pimonidazole (Hypoxyprobe HP6–100Kit antibody, 1:50), and CD8 (Cell Signaling Technology, 98941, 1:50). Last, they were counterstained with DAPI (Vector Labs).

Stained sections were imaged with a slide scanner (AxioScan Z1, Zeiss) for fluorescence quantification of mosaic images. Vessels (CD31) costained with lectin allowed the quantification of perfused vessel fraction (CD31⁺lectin⁺), whereas normalized vessels corresponded to CD31⁺αSMA⁺ costained vessels. Quantification of the hypoxia fraction was calculated by the percent area of tissue positive for pimonidazole. Unbiased image quantification was conducted using custom software designed in MATLAB (Mathworks). Briefly, whole slide images were imported into MATLAB and first manually cleaned to ensure no imaging artifacts remained. The images were then individually thresholded. For each stain on each image, the mean (μ) and variance (σ2) for each background intensity were first calculated, and a threshold value of at least μ+4σ was used. Colocalization was then measured on channels of interest, namely between CD31 and lectin or CD31 and αSMA. For quantification of CD8⁺ T-cell density, 5 to 10 random fields were imaged for distribution of CD8 (ImageJ, NIH) and for acquiring representative images using a confocal microscope (Olympus) at 20× oil magnification. To avoid bias during the image acquisition, we took images in a blinded manner. The histologic slides were named following a random numeric sequence with no group info given to the researcher, using DAPI channel to define a field of view (FOV) with no overlaps, no large necrotic areas, and avoiding tumor borders. We took at least 5 random qualified FOVs per slide and then, as many qualified FOVs as possible up to 10 FOVs per slide.

mRNA was extracted from snap-frozen fragments of E0771 tumors (n = 5 mice per group, control vs. ExTr) using the RNeasy Mini Kit (QIAGEN). For this analysis, tumors were time-matched and size-matched to avoid potential confounding factors. The Massachusetts Institute of Technology BioMicro Center performed RNA sequencing (RNA-seq) on extracted RNA samples. First, a Fragment Analyzer (Advanced Analytical Technologies Inc.) confirmed the quality of RNA samples (RIN 7.9–10). Illumina libraries were then prepared and indexed from approximately 250 ng of total RNA using the Kapa Hyperprep Kit following the manufacturer's recommendations (Roche). The libraries were confirmed using the Fragment Analyzer, quantified by qPCR, and then pooled and sequenced on an Illumina NextSeq500 using 40 + 40 paired-end reads. The samples were demultiplexed by custom scripts, which allowed a single mismatch to the index sequencing. The raw FASTQ files were aligned using STAR (33) to the mouse Ensembl GRCm38 reference genome. The Bioconductor R package, GenomicAlignments (34), was used to generate gene read counts. DESeq2 (35) was used to normalize and determine the differentially expressed genes. Gene set enrichment analysis (GSEA; ref. 36) was performed using GSEA software from the Broad Institute (37). Gene set enrichment network analysis was visualized with EnrichmentMap (36) in Cytoscape 3.0 v3.7.2 (https://cytoscape.org/cy3.html), using a cutoff of FDR < 0.10. Each node in the network corresponded to a gene set. The sizes of the nodes were proportional to the number of genes in the gene set. Gene sets with overlapping genes were organized as a network. Raw RNA-seq data of bulk E0771 mouse tumors performed in this study have been deposited in the NCBI Sequence Read Archive under BioProject ID PRJNA689055.

RNA-seq data from patients with human breast cancer from Ligibel and colleagues (38) were downloaded from the Gene Expression Omnibus database (accession number GSE129508, from the study registered on the clinicaltrials.gov as NCT01516190). The authors stated that the participants obtained a medical clearance before participation, signed an informed consent, and that the study was approved by the Institutional Review Board (Dana-Farber/Harvard Cancer Center). We used STAR to align the raw FASTQ files to the human Ensembl GRCh38 reference genome, gene read counts were generated using GenomicAlignments, DESeq2 was used to normalize the raw counts (expressed by Volcano Plot and heat map), and GSEA was performed on the normalized counts.

Homogenized fragments from snap-frozen bulk tumors from both E0771 and MCa-M3C mouse models were used to profile cytokines using the Proteome Profiler Mouse Cytokine Array Kit (Panel A; R&D Systems), following the manufacturer's protocol. Three hundred microgram of protein pooled from n = 6 mice per group (50 μg/mice). The array membranes (Thermo Fisher Scientific) were imaged (Kodak X-OMAT 2000A), and the densitometry analysis was carried out using ImageJ (NIH, USA).

Each E0771 and MCa-M3C tumors were homogenized in reagent diluent provided by the DuoSet ELISA (R&D Systems) and mechanically disrupted for protein extraction. Fifty microgram of protein were loaded in duplicate, and ELISAs for CXCL9/MIG (DY492), CXCL10/IP-10/CRG-2 (DY466), and CXCL11/I-TAC (DY572) and plates were read at 450 nm with wavelength correction at 540 nm (iMark, Bio-Rad), following the manufacturer's protocol. Absorbance values were interpolated from the standard values, and the final concentration was expressed in ng/mg of protein.

Whenever the presumptions of normality and homoscedasticity were achieved, we conducted the Student t test to compare two groups. One-way ANOVA was used for comparison among three or more groups, and two-way ANOVA was used for tumor growth analysis, followed by Tukey post hoc test whenever necessary to identify specific group differences. In the other cases, we conducted a nonparametric analysis, comparing two groups by the Mann–Whitney test and more than two groups by the Kruskal–Wallis test. Correlation between variables was accessed by regression analysis. All statistical analyses were performed using GraphPad Prism 8 software. A P value ≤0.05 was considered significant. Data presented as mean ± SEM or median ± interquartile range (box plot and violin plot) and distribution (violin plot).

We first conducted an incremental running exercise test until exhaustion to define ExTr intensity for C57BL/6, FVB, and Balb/c mice. Then, we implanted murine breast cancers orthotopically in syngeneic, immunocompetent mice [E0771 (C57BL/6), MCa-M3C (FVB), and EMT6 (Balb/c)]. When tumors reached approximately 100 mm³, mice were randomized to control (sedentary) or ExTr groups (Fig. 1A). The ExTr consisted of daily sessions of continuous moderate-to-vigorous intensity aerobic exercise on a running treadmill, progressing from 30 to 45 minutes per session (Supplementary Fig. S1A–S1C). We trained mice with the same relative exercise intensity (Supplementary Fig. S1D; ∼60% of maximal running velocity). FVB mice displayed a higher maximal running velocity and better compliance with the ExTr program than C57BL/6 mice (Supplementary Fig. S1E and S1F). Therefore, total ExTr volume in FVB mice was higher than that in C57BL/6 mice. Using different models, we showed that ExTr delayed tumor growth in breast cancer. Seven days of consecutive ExTr reduced tumor burden in the three models of breast cancer (Fig. 1B–G; Supplementary Fig. S2A–S2F). Using a more traditional exercise intervention in preclinical models (i.e., 5 days of consecutive exercise for 2 weeks), we saw a similar tumor control induced by ExTr (Supplementary Fig. S2G–S2I). The consistent effect of ExTr on tumor growth was also observed on individual tumor growth and time to triple tumor volume (Supplementary Fig. S2C, S2F, and S2I). After 7 days of ExTr, tumor weight was approximately 30% lesser in ExTr than in control groups (Fig. 1E–G; Supplementary Fig. S2H). Similar antitumor effects of ExTr were found when started in smaller tumors (∼20 mm³ on MMTV-PyMT in FVB; Supplementary Fig. S2J). ExTr also blunted obesity-induced accelerated tumor growth (Supplementary Fig. S2K). Interestingly, ExTr prevented cancer-induced chronic fatigue, as demonstrated by increased exercise capacity (Supplementary Fig. S3A–S3D), which is a biomarker of ExTr response and a predictor of overall survival in patients with breast cancer (39). More importantly, ExTr was associated with reduced recurrence rate and metastatic burden in the metastatic MCa-M3C model (Supplementary Fig. S4A–S4C).

ExTr did not change vessel density (Fig. 2A and B) but promoted tumor vessels' maturation. As expected, we found ExTr induced vessel normalization in both E0771 and MCa-M3C breast cancer models. ExTr led to an increased percentage of vessels associated with αSMA⁺ perivascular cells (Fig. 2C and D), an indicator of vascular normalization. ExTr also increased the fraction of vessels that were perfused (Fig. 2E and F), indicating improved vascular function. More functional vessels resulted in reduced hypoxia (Fig. 2G and H), indicating normalization of TME.

To determine potential mechanisms associated with the effects of ExTr on breast cancers, we performed unbiased RNA-seq analysis of E0771 tumor samples from control versus ExTr mice (Fig. 3A and B; Supplementary Fig. S5). GSEA of the RNA-seq data indicated several pathways modulated by ExTr compared with the control group (comprehensive list is in Supplementary Table S1). ExTr induced a metabolic reprogramming by reducing glucose uptake/insulin pathways and upregulated several gene sets associated with oxidative metabolism, such as oxidative phosphorylation (OXPHOS). It also enhanced gene expression linked to cytokine activation and downregulated several pathways involved with immunosuppression in breast cancer (20, 40, 41), whose inhibition is demonstrated in the clinic to synergize with anti–PD-1 against solid tumors (42). This molecular reprogramming suggests a less glycolytic and less immunosuppressive tumor, which is consistent with vessel normalization.

To test whether ExTr induced similar transcriptional signatures in the TME in patients with breast cancer, we analyzed the RNA-seq dataset previously published by Ligibel and colleagues (38). There were two groups of patients with breast cancer in the published study—exercise intervention (pre-ExTr and post-ExTr) and a mind–body control group (pre-mind and post-mind). We found via GSEA that in the ExTr group, there was upregulation of transcriptional pathways consistent with antitumor immune activity and vessel normalization in the patients with breast cancer (Fig. 3C and D), similar to what we found in mice. These results indicate that ExTr can improve antitumor immune activity and vessel normalization in not only mice, but also patients with human breast cancer.

We and others have shown that vessel normalization using antiangiogenic therapies enhances trafficking of CD8⁺ T cells in murine breast cancer models (17–20). Transcriptomic changes associated with ExTr support a shift from immunosuppression to activation of antitumor immunity (Fig. 3). Therefore, we studied the phenotypic changes in the immune TME induced by ExTr using flow cytometry. Indeed, we found that ExTr increased the infiltration of CD8⁺ T cells into the breast cancer TME (Fig. 4A–D; Supplementary Fig. S6A and S6B). We profiled the immune TME in breast cancers during ExTr and found that ExTr consistently boosted CD8⁺ T-cell infiltration, indicated by changes in relative (Fig. 4A) and absolute numbers (Fig. 4B) of CD8⁺ TILs. We also analyzed the intratumoral localization of CD8⁺ T cells with immunofluorescence. Consistent with flow cytometry analysis, we found that ExTr increased the intratumoral distribution of CD8⁺ T cells within the TME (Fig. 4C), indicated by the increased number of infiltrating CD8⁺ T cells per field (Fig. 4D). TILs are associated with favorable outcomes in breast cancers (22, 24, 43). We, therefore, determined the potential effects of ExTr in promoting the trafficking of TILs. Changes in tumor volume were inversely correlated with CD8⁺ TIL density (Supplementary Fig. S6C). The higher the increase in maximal aerobic exercise capacity (a biomarker of response to ExTr), the higher was the infiltration of CD8⁺ TILs (Supplementary Fig. S6D). On the other hand, no significant changes in conventional CD4⁺ (TCR⁺CD4⁺FOXP3⁻) or regulatory T cells (TCR⁺CD4⁺FOXP3⁺) were found (Fig. 4A). We did not see any changes in other cell populations potentially relevant to antitumor immunity (Supplementary Fig. S7), such as tumor-associated macrophages (CD45⁺Ly6G⁻Ly6C^lowCD11b⁺F4/80⁺), dendritic cells (CD45⁺Ly6G⁻CD11b^lowCD11c⁺), or natural killer (NK) cells (CD45⁺TCRβ⁻NK1.1⁺). Hence, we focused on the potential role of CD8⁺ T cells in mediating the antitumor effect of ExTr.

We measured the immune phenotype of CD8⁺ T cells in breast cancers in response to ExTr. The proliferation marker Ki67 was elevated in ExTr groups (Fig. 4E and F). In addition, ExTr increased the fraction of effector CD8⁺ T cells expressing granzyme B (Fig. 4G and H) and IFNγ (Fig. 4I and J). Because exhausted cells restrain the potential for antitumor effector function, we determined the expression of the checkpoint receptors PD-1, CTLA-4, TIM3, and LAG3 in CD8⁺ T cells (Supplementary Fig. S8A and S8B). We found ExTr reduced the amount of CD8⁺CTLA-4⁺ in MCa-M3C tumors and CD8⁺LAG3⁺ in E0771 tumors. ExTr also increased the number of nonimmune cells expressing PD-L1 (Supplementary Fig. S8C). We then checked for the coreceptor expression (PD-1⁺CTLA-4⁺) in CD8⁺ T cells (Supplementary Fig. S8D), which is considered a marker of exhausted T cells (26). We found that ExTr significantly reduced CD8⁺ T cells showing an exhausted phenotype in E0771 and did not change the percentage in MCa-M3C (Supplementary Fig. S8E). However, ExTr significantly increased the total number of CD8⁺ T cells, contributing to the increased number of exhausted PD-1⁺CTLA-4⁺ T cells within the tumors (Supplementary Fig. S8F).

After characterizing the selective infiltration of CD8⁺ T cells, we examined the association of higher CD8⁺ T-cell infiltration with the antitumor effect of ExTr. We subjected mice to 14 consecutive days of ExTr when tumors in the control group would have reached the maximum tumor size of approximately 1 cm³. Treatment with depleting antibodies targeting CD8⁺ T cells was started concurrently with ExTr when tumors were approximately 100 mm³ (Fig. 5A). We found that CD8⁺ T-cell depletion completely abrogated ExTr-induced tumor growth delay (Fig. 5B and C; individual values and time to triple tumor volume in Supplementary Fig. S9A–S9D; comparison between tumor growth curve from anti-CD8 and IgG in Supplementary Fig. S9E), indicating that CD8⁺ T-cell tumor infiltration is necessary for the antitumor effect of ExTr.

To dissect the mechanism of ExTr-induced CD8⁺ T-cell infiltration, we conducted a proteome profile array of bulk tumors and identified CXCL9 and CXCL11 as potential candidates (Fig. 6A; Supplementary Fig. S10). We then quantified all three CXCR3 ligands (CXCL9, CXCL10, and CXCL11) by ELISA in the breast cancer models. We found ExTr significantly increased CXCL9 in E0771 and CXCL11 in the MCa-M3C model (Fig. 6B–D). CXCL9 also showed a trend toward increase in MCa-M3C (Fig. 6B). There was no change in CXCL10 in both models (Fig. 6C). C57BL/6 mice do not express functional CXCL11. We also determined cellular CXCL9 expression and found several CXCL9-expressing cell populations (Supplementary Fig. S10A–S10J). CXCL9⁺ immune cells were predominantly macrophages (Supplementary Fig. S10H and S10I). It is possible that macrophages mediate beneficial effects of ExTr directly via CXCL9 expression and/or indirectly via vascular normalization (44–46). However, we could not identify a particular cell type that dominantly elevated CXCL9 expression with ExTr. CXCL9 and CXCL11 chemokines are specific ligands for the CXCR3 receptor. We also observed that ExTr increased the absolute number of CD8⁺ T cells expressing CXCR3 more than 3-fold in MCa-M3C (P < 0.05) and 2-fold in the E0771 (P = 0.07) tumor (Fig. 6E). Considering its role in the recruitment of CD8⁺ T cells into tumors (47–50), we asked whether CXCR3 deficiency would impair the antitumor effect of ExTr. To this end, we orthotopically implanted E0771 cells in Cxcr3^−/− mice (C57BL/6 background), and when tumors reached 100 mm³, we subjected them to ExTr for 11 consecutive days, when these tumors began to reach 1 cm³. Loss of Cxcr3 completely abolished antitumor effects of ExTr, as assessed by tumor growth kinetics and final tumor weight at day 11 (Fig. 6F and G). The number of CD8⁺ T cells in E0771 tumors grown in Cxcr3^−/− mice was 10-fold lower than that observed in wild-type mice after 7 days of ExTr (Fig. 6H). The frequency of CD8⁺ T cells was also lower in Cxcr3^−/− mice (Fig. 6I). Majority of the CD8⁺ T cells expressed IFNγ⁺, and there was no difference between groups (Fig. 6J). ExTr induced a small but significant increase in CD8⁺ T cells in Cxcr3-KO mice, suggesting CXCR3-independent mechanisms of CD8⁺ T-cell recruitment (Fig. 6H and I). However, these CD8⁺ T cells were not enough to induce tumor control.

ExTr increased perfusion, alleviated hypoxia, and enhanced the infiltration of CD8⁺ T cells in the TME. However, due to the increased total CD8⁺ T-cell recruitment in ExTr mice tumors, the number of exhausted CD8⁺ T cells also increased proportionally. Hence, we hypothesized that the combination of ExTr and ICB therapy would have a superior effect on breast cancers. When MCa-M3C tumors reached approximately 100 mm³ in size, we treated mice with either anti–PD-1 alone (Supplementary Fig. S11A and S11B) or anti–PD-1 in combination with anti–CTLA-4 (Fig. 7A; Supplementary Fig. S11C–S11G). During treatment, mice remained either sedentary or performed daily exercise (14-day ExTr protocol). We found no effect of ICB alone in the orthotopic breast cancers (Fig. 7B and C; Supplementary Fig. S11). On the other hand, the addition of ICB to ExTr exhibited superior breast cancer tumor control, delaying tumor growth, reducing the final tumor volume, and further enhancing the antitumor effect of ExTr (Fig. 7B and C; Supplementary Fig. S11). Further analysis indicated similar infiltration of CD8⁺ T cells in both groups subjected to ExTr (Fig. 7D–G). ExTr also increased the total number of intratumoral granzyme B⁺CD8⁺ T cells with ExTr, although the fraction of granzyme B⁺CD8⁺ T cells was increased only in the ExTr+ICB group (Fig. 7E–H). Finally, ExTr and ICB seemed to have a synergistic effect in inducing an IFNγ⁺ response in CD8⁺ T cells, as both the relative and the absolute number of CD8⁺IFNγ⁺ T cells increased only with the combined treatment (Fig. 7F–I). ExTr also prevented cancer-induced fatigue in ICB-treated mice (Supplementary Fig. S11G). Altogether, we demonstrated that ExTr intervention in established tumors normalized blood vessels, increased CD8⁺ T-cell infiltration, and improved control of refractory breast cancer tumors when combined with ICB.

ExTr is an emerging antitumor strategy (2, 10), yet its potential to enhance immunotherapy in established tumors has not been reported. To study exercise in a therapeutic setting, we needed to define the exercise dose. Because most previous preclinical studies use ad libitum physical activity models (6, 8, 10), which do not properly capture human exercise behavior, we adapted a controlled exercise. We used a treadmill to precisely control the intensity and duration of exercise dose. Unlike the voluntary running model, in which mice can run 20 km per day (51), our treadmill model's exercise volume is 10 to 20 times less than this. Thus, it better recapitulates ExTr in patients participating in clinically supervised ExTr programs. Using this clinically relevant ExTr protocol, we found that aerobic ExTr decreases tumor growth and reprograms the TME by increasing perfusion and decreasing hypoxia in established orthotopic breast cancer models. To investigate the causal role of CD8⁺ T cells on ExTr-induced tumor growth delay, we used antibody-mediated CD8⁺ T-cell depletion. The depletion of CD8⁺ T cells abolished the antitumor effect of ExTr, confirming that the mechanism of antitumor activity from ExTr is CD8⁺ T-cell dependent.

Emerging data suggest that exercise dose plays a significant role in modulating tumor physiology (2, 52). This is supported by clinical observations showing that even fulfillment of the recommended levels of physical activity (e.g., 150 minutes of accumulated moderate-intensity aerobic exercise most days of the week or 75 minutes of vigorous exercise every other day) reduces overall mortality in breast cancer (1, 4). Although meeting the minimum guidelines for physical activity may improve survivorship in the long term, it might not be sufficient to promote clinical benefits during cancer treatment. It seems only supervised ExTr, accounting for planned interventions considering both duration and intensity, has been proven beneficial (3, 53–55). We found a significant variability of ExTr dose within preclinical models, even though it is usually assumed that the same ExTr velocity would fit all mouse strains and experimental conditions. Our study demonstrated that the relative exercise intensity for C57BL/6 versus FVB female mice is distinct. This may explain, at least in part, why previous preclinical studies failed to show antitumor activity under similar ExTr protocols (7, 52, 56). More importantly, the lack of proper control of exercise dose might account for the large variability and absence of significant results from exercise interventions in clinical trials (2).

The timing and type of ExTr intervention might also induce different TIL phenotypes. In voluntary running models conducted in the prevention setting (i.e., starting ad libitum run before implantation or appearance of tumors), NK cells (8) and CD8⁺ T cells (10) were previously reported to explain the antitumor effect of ExTr. However, because the host was already primed by the adaptations induced by ExTr, including the intrinsic metabolic capacity of CD8⁺ T cells (10), exercise might have enhanced antitumor capabilities even before a tumor's existence. The problem is that, along with the much larger exercise volume that does not recapitulate human exercise behavior, the running wheel model does not provide a reliable preclinical model on the therapeutic ExTr either. This is particularly relevant considering that, although three out of four patients are willing to start an exercise program, nearly 60% of them do not follow an exercise routine (57).

Previous therapeutic ExTr interventions (i.e., starting treatment when tumors were established) on running treadmills could not demonstrate enhancement of cytotoxic NK or CD8⁺ T cells (7, 52, 56). This is probably due to differences in exercise intensity. It is well established that both NK and CD8⁺ T cells increase in the peripheral blood in response to acute exercise (58). This occurs as a function of exercise-induced adrenergic activation (8, 59), linked to exercise intensity. Hence, transient mobilization of immune cells might occur in response to every exercise session, thus favoring relocalization and immune surveillance of tumors. Naïve and memory subsets of CD8⁺ T cells are increased only beyond moderate-to-vigorous exercise intensity sessions (58). Therefore, if a certain threshold of exercise intensity is not achieved, one should not expect mobilization and redistribution of antitumor cells.

Consistent with previous preclinical reports (6, 7), we showed that ExTr induces vessel normalization, characterized by a larger number of vessels with higher pericyte coverage, resulting in less tumoral hypoxia. To gain insight into changes induced by ExTr, we conducted an unbiased RNA-seq analysis of bulk RNA isolated from tumors to characterize the effects of ExTr on the TME. We found that ExTr modulated antitumor immune responses in preclinical models, as well as a transcriptional signature, consistent with the metabolic switch toward OXPHOS metabolism. This shift can occur as a consequence of enhanced tumor perfusion due to vessel normalization (60). We compared our results with the transcriptomic changes induced by a short-term ExTr in the TME in treatment-naïve patients with breast cancer (38). In this study, the ExTr consisted of short-term (∼1 month) supervised sessions of strength training twice a week, and patients were asked to perform aerobic exercise sessions on their own. It was not possible to categorize transcriptome signatures based on aerobic exercise dosing in this study. Nevertheless, we found a transcriptional TME reprogramming in patients consistent with improved antitumor immunity and vessel normalization, similar to what we observed in preclinical models of breast cancer. Vessel normalization may occur at ExTr intensities lower than that used in our study. For example, Schadler and colleagues (7) showed increased perfusion in subcutaneous tumors in C57BL/6 mice by ExTr at 12 m/min. This velocity would have represented approximately 35% of maximal running capacity for C57BL/6 mice in our study, instead of 60%. These authors did not observe changes in tumor growth or immune cell infiltration with the low-intensity ExTr. Although vessel normalization can be observed with low-intensity ExTr (7), a higher exercise dose may be required for mobilization and redistribution of immune cells (8, 58, 59). Thus, the synergism between induced vessel normalization and CD8-mediated tumor control seems to occur when sufficient intensity (i.e., from moderate to vigorous) of ExTr is reached.

ExTr causes a ubiquitous homeostatic physiologic response, initiating several events that might contribute to the enhanced tumoral infiltration of CD8⁺ T cells. Tumors are predominantly glycolytic and with low glucose in the TME of aggressive tumors (61). Conversely, increased glycolytic rate and low perfusion/oxygenation contribute to intratumoral hypoxia and acidosis, which is alleviated with enhanced perfusion (12). Thus, a metabolic switch toward less hypoxic and more oxidative metabolism that favors CD8⁺ T-cell effector function (62) might have facilitated the enhanced CD8⁺ T-cell effector function with ExTr. Also, CD8⁺ T cells cannot compete with cancer cells for the scarce intratumoral glucose availability, which further impairs CD8⁺ T cells' density and activity (61). Therefore, by inducing vessel normalization and promoting a less hypoxic tumor, ExTr allows a more favorable metabolic TME to facilitate the trafficking of T cells and trigger antitumor effector function (63).

To understand mechanisms promoting enhanced CD8⁺ T-cell trafficking, we found ExTr increases the concentration of the chemokines CXCL9 and CXCL11, but not CXCL10, in breast tumors. CXCL11 is not functional in C57BL/6 mice, so it was only quantified in the MCa-M3C tumor model in FVB mice (48). CXCL9/10/11 are ligands of the CXCR3 receptor involved in increased trafficking of CD8⁺CXCR3⁺ T cells to tumors (47, 48, 64). These chemokines are induced by IFNγ and were potentially triggered in response to the ExTr-mediated increase of IFNγ-producing effector CD8⁺ T cells, thereby inducing positive feedback to attract more CD8⁺ T cells to tumors. There is an association of IFNγ-secreting T cells and vessel normalization (18), and IFNγ promotes the enhanced expression of adhesion molecules in endothelial cells (65). These cells display a strong affinity to CD8⁺CXCR3⁺ T cells (66), further increasing their infiltration. ExTr antitumor effect was entirely abolished in Cxcr3^−/− mice, demonstrating that the CXCR3 pathway participates in the antitumor effect of ExTr. Data suggest that CXCR3 is not essential for the early recruitment of CD8⁺ T cells into tumors (64). However, CXCR3 in CD8⁺ T cells is necessary to trigger initial effector responses in bystander CD8⁺ T cells (67), although not required for sustained IFNγ effector responses (68). Indeed, a large number of infiltrating CD8⁺ T cells expressed IFNγ in Cxcr3^−/− mice. This may explain why CD8⁺CXCR3⁺ T cells are required for improved response to immunotherapy. Their ligands are reported to function both as biomarkers and inducers of response to anti–PD-1 alone (64) or in combination with anti–CTLA-4 (47). It also explains why Cxcr3^−/− mice blunted the antitumor effect of ExTr. Future studies need to examine the inducers of activation and effector function of tumoral CD8⁺ T cells by ExTr.

The relevance of the CXCL9/10/11–CXCR3 pathway for recruiting and triggering cytotoxic T cells is well known (47–50, 64). Several therapeutic approaches have been tested to leverage this pathway to attract more CD8⁺ T cells, including radiotherapy, chemotherapy (50), and vaccination (49). Multimodal therapy increases ICB response, but it is associated with increased adverse events (69). Therefore, less toxic approaches are necessary. Our findings support ExTr as a safer strategy to stimulate the antitumor CXCL9/11–CXCR3 system in solid tumors.

ICB has revolutionized the treatment of multiple malignancies. However, its efficacy in breast cancer has been modest (27, 70). Lack of perfusion, tumoral hypoxia, and T-cell exclusion are considered significant resistance mechanisms (15, 17–19). Several strategies have been proposed to alleviate immune exclusion and sensitize tumors to ICB (70). Here, we show that controlled ExTr can facilitate immune cell Infiltration into tumors and sensitize them to ICB. ExTr increases CD8⁺ T cells by reprogramming the TME toward a less aggressive phenotype, but a large fraction of CD8⁺ T cells remains expressing immune checkpoint receptors, suggesting these tumors would benefit from ICB therapy. Our work shows that ExTr and ICB synergistically enhance the activity of effector CD8⁺ T cells (i.e., a fraction of IFNγ- or granzyme B–expressing CD8⁺ T cells). Both CD8⁺ T-cell infiltration and IFNγ signaling signatures are indicative of benefit from combined anti–PD-1 and anti–CTLA-4 in patients with solid tumors (71) or anti–PD-L1 alone in breast cancer (72). ExTr increased the number of cells expressing PD-L1, a biomarker of response for metastatic breast cancer (25). Our data indicate that supervised ExTr has the potential to unleash antitumor response from immunotherapy in the clinic.

R.K. Jain reports honorarium from Amgen; consultant fees from Chugai, Elpis, Merck, Ophthotech, Pfizer, SPARC, SynDevRx, XTuit; equity in Accurius, Enlight, Ophthotech, SynDevRx; Boards of Trustees of Tekla Healthcare Investors, Tekla Life Sciences Investors, Tekla Healthcare Opportunities Fund, Tekla World Healthcare Fund; and a grant from Boehringer Ingelheim. Neither any reagent nor any funding from these organizations was used in this study. No disclosures were reported by the other authors.

I.L. Gomes-Santos: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. Z. Amoozgar: Formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. A.S. Kumar: Data curation, formal analysis, validation, methodology, writing–review and editing. W.W. Ho: Formal analysis, investigation, writing–review and editing. K. Roh: Investigation, writing–review and editing. N.P. Talele: Validation, investigation, writing–review and editing. H. Curtis: Investigation, writing–review and editing. K. Kawaguchi: Conceptualization, investigation, writing–review and editing. R.K. Jain: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. D. Fukumura: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

The authors thank Dr. Joao Incio for preliminary discussions; Drs. Peigen Huang, Sylvie Roberge, and Carolyn Smith for technical support; Dr. Yao Sun for experimental assistance; and Drs. Patrik C. Anderson, Dan G. Duda, Gino B. Ferraro, and Timothy P. Padera for their helpful input on the article.

This work was supported by grants from the National Institutes of Health (to D. Fukmura and R.K. Jain); National Foundation for Cancer Research, Harvard Ludwig Cancer Center, Advanced Medical Research Foundation and Jane's Trust Foundation (to R.K. Jain); fellowships from Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP #2014/13690–8 and #2016/21320–1), in agreement with Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), as visiting fellow from Instituto do Coração do Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (InCor-HCFMUSP), Sao Paulo, Brazil (to I.L. Gomes-Santos); an Agency for Science, Research and Technology (A*STAR) Fellowship (to A.S. Kumar and W.W. Ho); a Cancer Research Institute/Merck Fellowship (to N.P. Talele); NCI grant #R25CA174650 through the CanCure Co-Op Program at Northeastern University in Boston, MA (to H. Curtis); an International Fellowship from the Alumni Association Sanjukai Kagawa University Faculty of Medicine; and an overseas research fellowship from the Japan Society for the Promotion of Science & KANAE foundation (to K. Kawaguchi).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.