Home-based high-intensity exercise training during neoadjuvant chemotherapy is associated with favorable chemotherapy outcomes and tumor microenvironment remodeling in breast cancer

Abstract

Background

A growing number of studies show that physical exercise is feasible and safe for cancer patients and enhances various physical and cognitive capabilities, thus improving the quality of life and disease outcomes. The molecular mechanisms behind these effects are, however, poorly understood. Here, we present the results from the CancerBeat study, where breast cancer (BC) patients receiving neoadjuvant chemotherapy (NAC) were enrolled in a high-intensity interval training (HIIT) program to investigate the effects of exercise on tumor biology and response to NAC.

Methods

Fifty BC patients were randomized into HIIT and control (CON) groups. Patients were enrolled in the study shortly after diagnosis, and the HIIT group took personalized home-based training sessions for the whole duration of NAC (mean 24 weeks). NAC response was evaluated using the standard Miller-Payne grading system. A retrospective control (Ctrl) group consisted of age, stage and subtype-matched BC patients receiving standard of care during NAC. Acute exercise-induced changes in the plasma proteome were analyzed with the Olink Immuno-Oncology panel. Bulk RNA sequencing analysis was performed on BC and normal breast tissues from 20 patients in the HIIT and Ctrl groups. Tumor-infiltrating immune cell composition was inferred from the bulk RNA-seq data using CIBERSORTx, and findings were validated with a 7-plex immunofluorescence assay.

Results

In the HIIT group, the proportion of patients with Miller-Payne grades 4 and 5 was significantly higher compared to CON (80% vs. 47%, p = 0.036) and Ctrl groups (80% vs. 50%, p = 0.047), representing the primary finding of this study. Exploratory plasma proteomic analysis identified 2 and 20 plasma proteins whose levels change after acute exercise in BC patients at diagnosis and after completing the training program/NAC, respectively; five of these had not previously been linked to exercise. Bulk RNA sequencing of post-NAC tumor tissues revealed HIIT/NAC-associated gene expression patterns consistent with less aggressive phenotypes, favorable outcomes, and altered wound-healing and immune-related processes at the tumor site. The findings suggest the possible involvement of fibroblasts, endothelial cells, and other stromal cells in exercise-associated tumor microenvironment remodeling. CIBERSORTx analysis suggested greater infiltration of activated NK cells in HIIT tumors compared to resting NK cells in Ctrl tumors. As an exploratory observation, both plasma and tumor tissue levels of decorin were upregulated in the HIIT group, and this increase was associated with better NAC response and favorable prognosis. These findings warrant further investigation in larger, dedicated studies.

Conclusions

Our study provides novel insights into the molecular mechanisms underlying the protective effects of exercise on BC progression. These preliminary results support the further investigation of physical exercise as a component of BC treatment and highlight the need for clinical trials exploring the impact of HIIT on immunotherapy efficacy.

Trial registration

retrospectively registered at ClinicalTrials.gov, NCT06522971.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12964-026-03053-1.

Background

Physical exercise is increasingly recognized not only for its beneficial effects on reducing cancer risk but also for improving survivor outcomes. In breast cancer (BC) patients, higher levels of post-diagnosis physical activity have been shown to reduce fatigue and treatment side effects and improve physical and cardiorespiratory fitness [1–3], thus leading to improved quality of life, as well as improved survival rates and reduced recurrence [4, 5].

These effects are mediated in part by exerkines - bioactive molecules released by tissues such as muscle, liver, brain, bone, and adipose tissue during exercise [6]. Plasma proteome analyses have identified hundreds of proteins that fluctuate immediately following activity [7] or during long-term training adaptation [8]. Several myokines - muscle-derived exerkines - including IL-6, SPARC, Oncostatin M, and Irisin, have been found to have anti-cancer properties. They work by directly inhibiting cancer cell growth, recruiting effector immune cells to the tumor microenvironment (TME), or normalizing tumor vasculature [9–11]. However, while exercise-induced proteomic changes are well-documented in healthy individuals, it remains unknown how the presence of cancer or chemotherapy influences these systemic responses.

Preclinical studies using rodent models support the protective effects of exercise in prophylactic and therapeutic settings, and suggest that exercise can directly modulate tumor biology [12]. Several independent studies have demonstrated that various forms of exercise enhance anti-tumor immune responses and reprogram the immune TME in mouse models. For instance, endurance exercise performed prior to the injection of mammary tumor cells has been shown to slow tumor growth and extend survival, potentially by suppressing the recruitment of FoxP3⁺ regulatory T cells into the tumor [13]. Furthermore, regular aerobic exercise after a tumor challenge has been shown to normalize intratumoral vascularization [14], decrease the accumulation of myeloid-derived suppressor cells, and increase the proportion of activated NK cells and CD8⁺ T cells within the tumor [15]. Another study using five different mouse tumor models, showed that exercise promotes the mobilization and intratumoral infiltration of NK cells [16]. However, to what extent these findings can be generalized and attributed to humans is not known, and the antitumor mechanisms of exercise in BC patients remain largely unknown.

Neoadjuvant chemotherapy (NAC), a treatment given before surgery to shrink a tumor, plays a crucial role in the management of locally advanced BC by making inoperable tumors operable, reducing the extent of surgery, and potentially preventing the spread of metastatic cells [17, 18]. However, NAC often induces deleterious effects including myelosuppression, reduced cardiorespiratory fitness and muscle wasting, fatigue and poor sleep, elevated risk of osteoporosis and impaired cognition, thus significantly impacting patients’ physical functioning, well-being and quality of life [19, 20]. Several intervention studies have demonstrated that regular physical exercise during NAC is feasible and safe, helps to maintain physical capacity, cardiopulmonary function and quality of life, and may lead to improved survival [2, 21–23]. We reasoned that enrolling BC patients undergoing NAC into a structured exercise program would not only improve clinical outcomes but also provide a unique opportunity to identify the exercise-induced cellular and molecular processes within tumor tissues.

Although multiple exercise modalities, such as aerobic exercise, resistance training, Nordic walking, etc. have shown beneficial effects in cancer patients [24], we opted for high-intensity interval training (HIIT), which is well-tolerated during NAC and has proven effective in improving cardiorespiratory fitness during chemotherapy [2, 25, 26]. Moreover, HIIT has been shown to induce a greater increase in the CD8⁺ effector memory T cell proportion in peripheral blood [27], mobilize NK cells with cytotoxic and tumor-homing phenotypes [28], and induce more pronounced alterations in peripheral blood mononuclear cell activation pathways than moderate-intensity continuous training [29]. We hypothesized that HIIT would trigger measurable systemic protein responses and modulate the immune TME to enhance immune-mediated tumor cell killing, thereby augmenting NAC efficacy.

The primary objective of this randomized pilot study was to evaluate the association between a home-based HIIT program during NAC and pathological responses in BC patients. Secondary exploratory objectives were to characterize acute exercise-associated changes in circulating proteins and to explore HIIT/NAC-associated post-treatment tumor transcriptomic signatures to provide insight into the molecular mechanisms mediating the effects of exercise on cancer biology. Gaining a deeper understanding of these mechanisms would enable rational design of personalized exercise protocols, finding the most effective dose and modality of exercise, and optimizing integration of exercise within treatment plans.

We found that HIIT significantly enhances the response to NAC in BC patients. Exercise-induced changes in the plasma proteome were assessed using the Olink Immuno-Oncology panel and alterations in the tumors’ gene expression profile were studied using bulk RNA sequencing analysis. In addition, the composition of tumor-infiltrating immune cells was estimated by the CIBERSORTx bulk gene expression profile deconvolution tool and subsequently studied by a 7-plex immunofluorescence assay. Finally, we demonstrate that plasma levels of decorin (DCN)— a known exerkine —increase during acute exercise. Furthermore, tumor DCN mRNA levels were elevated in the HIIT group and associated with improved NAC response and a favorable prognosis.

Methods

Participants and study design

Patients with locally advanced, non-metastatic breast cancer scheduled for doxorubicin/cyclophosphamide-based NAC were recruited at the Latvian Oncology Center, Riga East University Hospital. To ensure a homogeneous study population, eligibility was restricted to patients aged 30–65 with a core needle biopsy-confirmed primary diagnosis of stage IIA–IIIC (T1-4, N0-3, M0) disease who had not received prior oncological treatment. Candidates were excluded if they presented with cardiac pathologies, had a history of other malignancies, or had received blood transfusions within the previous six months. Additionally, those already engaging in more than 180 min of moderate-to-high-intensity aerobic training per week or participating in concurrent clinical trials were ineligible.

Patients were randomized to a HIIT group and a standard of care control (CON) group at an intended ratio of 2:1. The final achieved allocation was 32 and 18 participants, respectively, reflecting the inherent variation produced by an incomplete final block in the permuted block randomization sequence. Patients were enrolled in the study between diagnosis and the initiation of first NAC course. The HIIT group took personalized unsupervised home-based training sessions for the whole duration of NAC (8 courses; average of 24 weeks), whereas patients assigned to the CON group were advised to maintain their usual physical activity levels. All HIIT and CON group participants underwent maximal exercise tests at the Health Care and the Sports Science Research Laboratory, Latvian Academy of Sport Education, before the onset of NAC and 4 weeks after the last course of NAC (before the BC surgery). The UNCCRI Treadmill protocol [30], consisting of twenty-one 1-minute stages, with increments in speed and/or grade at each stage, was used to obtain HR_max and VO_2peak data that were further used for designing personalized exercise programs. VO_2peak and ventilatory parameters were measured by direct gas exchange analysis using Vyntus CPX system (Germany) and HR was measured continuously throughout the test with Polar H10 HR sensor. Blood samples from the HIIT and CON group participants were collected before and immediately after the test (designated Pre-Run and Post-Run samples). Sample sets collected at the time of diagnosis were designated HIIT-DG, while sample sets collected after completion of the HIIT/NAC were designated HIIT-NAC.

In addition, a retrospective control (Ctrl) group consisting of 20 age, stage and BC subtype-matched patients was selected from the BCPerMed study, receiving 8 courses of doxorubicin/cyclophosphamide-based NAC. The clinical samples from these patients were retrieved from the Latvian Genome Database. The study design is shown in Fig. 1a and the clinical and histopathological characteristics of the study population are provided in Table 1.

Fig. 1.

Effect of high-intensity interval training on response to neoadjuvant chemotherapy. a, Design of the CancerBeat study. b, Stacked bar plot showing the fraction of patients with each Miller-Payne grade in the HIIT, CON and Ctrl groups. The difference in Miller-Payne grades between the two groups was assessed by Mann-Whitney U test. c, Patients were dichotomized based on Miller-Payne grades as responders (R) (grades 4–5) or non-responders (NR) (grades 1–3), and statistical significance was evaluated by the Chi-square test. HIIT – high-intensity interval training, CON - prospective control group, NAC – neoadjuvant chemotherapy

Table 1.

Clinical characteristics of the study population

Characteristic	All samples				RNA-seq samples
	HIIT	CON	Ctrl	Chi-square, p-value	HIIT	Ctrl	Chi-square, p-value
Total enrolled (n)	32	18
Sample size (n)	20	17	20		10	10
Age mean, years	46.1	48.9	47.4		48.1	47.4
Age range, years	34–58	34–61	33–64	0.57*	35–58	35–64	0.79*
Tumor grade
Grade 1	2	1	0	0.32	0	0	0.12
Grade 2	15	12	12		9	6
Grade 3	3	4	8		1	4
TNM stage
T1 N1-3 M0	1	0	0	0.02	0	0	0.33
T2 N1-3 M0	15	14	8		7	2
T3 N1-3 M0	3	3	12		2	8
T4 N1-3 M0	1	0	0		1	0
Oestrogen receptor > 50%
Positive	10	9	11	0.95	5	4	0.65
Negative	10	8	9		5	6
Progesterone receptor > 50%
Positive	7	5	7	0.92	2	2	1
Negative	13	12	13		8	8
HER2 overexpression
0	9	9	8	0.17	4	3	0.5
1	6	2	6		3	3
2	0	0	3		0	2
3	5	6	3		3	2
Triple negative breast cancer
Yes	6	5	7	0.92	2	4	0.33
No	14	12	13		8	6
E-cadherin
Positive	13	12	14	0.94	9	5	0.32
Negative	2	2	3		1	2
Proliferation Index (Ki-67)
≤ 14%	6	0	5	0.05	2	2	1
> 14%	14	17	15		8	8
Neoadjuvant chemotherapy regimens**
AC	4	4	8	0.53	4	8	0.28
ACD	8	7	9		3	1
ACD + Phesgo	2	2	0		1	0
ACT	2	1	2		0	1
ACT + Carboplatin	2	0	0		1	0
ACT + Phesgo	1	0	0		1	0
D + Phesgo	1	0	0		0	0
Number of chemotherapy courses
4–6	0	0	2	0.17	0	1	0.3
8	20	14	18		10	9
Response to chemotherapy (Miller-Payne)
1–3	4	10	10	0.05	3	5	0.36
4–5	16	8	10		7	5
Pre-diagnosis physical activity level***
Inactive	3	2	3	0.34	3	2	0.72
Moderate	11	10	7		3	3
Active	4	4	10		3	5

*Kruskal-Wallis, p-value

**AC Doxorubicin + Cyclophosphamide, D Docetaxel, ACD AC + D, ACT AC + Paclitaxel

***Estimated using a self-reported health and lifestyle questionnaire

The study protocol was approved by the Latvian Central Medical Ethics Committee (approvals No. 4066 and 01-29.1/487). The CancerBeat study is registered at ClinicalTrials.gov (identifier: NCT06522971).

Physical exercise intervention

The HIIT intervention consisted of 2 to 3 running/walking sessions per week for 6 months. The HIIT session started with a 6-minute warm-up period at about 65–70% HR_max followed by 4 ⋅ 4-min high-intensity intervals (85%-95% of HR_max) combined with 3 min periods of active recovery (55–70% of HR_max). In total, the session lasted 34 min. This protocol is based on a well-established aerobic interval training model [31] that is safe and effective in improving cardiorespiratory fitness and quality of life in cancer patients [25, 26]. At the beginning of the intervention, each participant underwent one supervised training session to get familiar with the HIIT protocol, received an HR sensor (Polar H10) and uploaded the Polar Flow application. Subsequently, HIIT participants individually completed home-based training sessions and shared the sessions’ data with the investigators. The adherence rates were calculated as the percentage of sessions completed by each participant and were assessed at the midterm (week 12) and at the end of the study (week 24). Adherence was considered acceptable if participants completed > 60% of the sessions (from the total of 39 sessions). This threshold was selected to ensure sufficient exposure to the intervention to plausibly induce physiological and molecular adaptations, as adherence during chemotherapy in BC populations is typically moderate (60–70%), making this cut-off consistent with previously reported data [32].

Sample collection

Blood samples from the HIIT and CON group participants were collected into EDTA tubes before and immediately after the maximal exercise test and processed at room temperature within 2 h. Plasma samples were obtained by centrifugation of blood twice at 3000g for 10 min, aliquoted, and stored at -80 °C.

Once the NAC was completed, patients underwent BC surgery, during which tumor and adjacent normal tissue specimens were collected. Tissue specimens were macroscopically dissected by a histopathologist during surgery and stored in RNALater (Applied Biosystems, USA) at -20 °C until processing. The “tumor bed” was marked before the NAC with 3-millimeter titanium metal clips to allow precise identification of the tumor site after the NAC. Tumor tissue specimens were harvested directly from this site.

Response to NAC was assessed by standard histological examination of tissues at the tumor site using the Miller-Payne grading [33]. Patients with grades 1 to 3 were classified as non-responders (NR) and patients with grades 4–5 as responders (R).

The tissue specimens were collected after the patient’s informed written consent was obtained. The samples were stored in the Latvian Genome Database. The biobanking procedures have been approved by the Latvian Central Medical Ethics Committee (first approval No. 2007 A-7, renewed approvals No.1/19-04-05 and No. 01-29.1.2/6407).

Plasma proteome profiling

The Immuno-Oncology panel (panel file v.3114; Olink Bioscience, Uppsala, Sweden) was used to analyze 92 proteins simultaneously in each plasma sample using the Proximity Extension Assay (PEA) [34, 35]. Briefly, 1 µl of plasma samples pretreated as described above was incubated with antibody pairs modified with DNA oligonucleotides. The oligonucleotides of antibody pairs that were in close proximity hybridized and, after amplification with DNA polymerase, were quantified by real-time qPCR. Ct-values were transformed to the arbitrary units Normalized Protein Expression (NPX) using the NPX program manager (Olink Bioscience, Uppsala, Sweden). The data were further analyzed using the Olink analysis R package using t-test analysis for 2-group comparison and ANOVA test for multiple group comparison. These tests were used to identify differential abundance of proteins with p-value lower than 0.05 after adjusting the false discovery rate using the Benjamini-Hochberg procedure. Box plots were used for the visualization of the significant proteins.

Tissue RNA extraction

Ten patients were randomly selected from the HIIT group and age, stage and subtype-matched patients were selected from the Ctrl group and total RNA was extracted from tumor and adjacent normal specimens. Approximately 45–50 mg of tissue were overlaid with 1 ml of QIAzol Lysis Reagent (Qiagen, USA) in Lysing Matrix A tubes (MP Biomedicals, USA), and homogenized twice for 40 s at 6 m/s using FastPrep-24™ device (MP Biomedicals). The tubes were centrifuged at 1000g for 10 min at + 4 °C and total RNA was extracted from the supernatant using RNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. RNA was eluted in 30 µl of RNase-free water, quantified using NanoDrop 1000 (Thermo Fisher Scientific, USA) and the integrity of RNA was assessed with Agilent Bioanalyzer using RNA Nano chip (Agilent Technologies, USA).

Tissue RNA sequencing

Total RNA samples (200 ng) were preprocessed to remove ribosomal RNA (rRNA) utilizing the MGIEasy rRNA Depletion Kit (MGI, China) and purified using RNA XP Cleanup Beads (Beckman Coulter, USA). RNA sequencing libraries were prepared using the MGIEasy RNA Directional Library Prep Set (MGI, China) according to the manufacturer’s instructions. The process involved fragmenting the RNA into approximately 250 bp fragments, reverse transcribing these fragments to create cDNA, and synthesizing the second DNA strand. Adapters were ligated to the synthesized DNA, which was subsequently amplified using PCR. The length and concentration of the PCR products were measured using an Agilent High Sensitivity DNA Chip on an Agilent 2100 Bioanalyzer (Agilent Technologies, USA) and a Qubit fluorometer (Thermo Fisher Scientific, USA), respectively. The constructed transcriptome libraries underwent paired-end sequencing on a DNBSEQ-G400 high-throughput sequencing instrument using the DNBSEQ-G400RS High-throughput Sequencing Set (MGI, China).

RNA sequencing data analysis

The RNA sequencing data were acquired in FASTQ format and analyzed using an ad hoc pipeline in R version 4.3.2. The pipeline included the trimming of adapters using Cutadapt 4.4 [36], and mapping of the reads against Ensembl human genome reference sequence (GRCh38) using STAR 2.7.11a [37]. Uniquely mapped reads were counted using Rsubread 2.16 package and Ensembl human genome annotation. Alignments, where TPM coverage was greater than 100 in at least one sample, were kept for analysis. Differentially expressed genes (DEGs) were identified using DESeq2 1.42 [38] package and Benjamini and Hochberg adjustment for multiple testing. Genes with adj. p-value < 0.05 and |Log2 fold change| >1 were considered as differentially expressed.

Pathway analysis of the identified DEGs was performed using the multiGSEA [39] and msigdbr packages. For GO term analysis, the MSigDB [40] C5 subcollection Gene Ontology Biological Process was utilized. Thirty pathways with the lowest adj. p-value (< 0.015) were selected for visualization.

Quantitative RT-PCR

To validate RNA sequencing data, 15 genes were selected for qRT-PCR analysis. 480 ng of total RNA was converted to cDNA using random hexamer primers and Maxima H Minus Reverse Transcriptase (Thermo Fisher Scientific) according to the manufacturer’s protocol. cDNA reaction mixtures were diluted 1:20 and 2 µl were used for qRT-PCR reactions. Primer sets were designed using Primer-BLAST [41] and ordered from Metabion (Germany). qRT-PCR was performed using SYBR™ Green PCR Master Mix (Thermo Fisher Scientific) according to the manufacturer’s instructions using ViiA 7 Real-Time PCR system (Thermo Fisher Scientific). All qRT-PCR experiments were performed in duplicates. To normalize the expression data, the normalization factor was calculated as the geometric mean of two internal control genes – B2M and TBP. All primer sequences are available upon request.

CIBERSORTx analysis

Immune cell infiltration in tumors and normal breast tissues was analyzed using CIBERSORTx bulk gene expression profile deconvolution tool [42]. Raw TPM file of RNASeq data was uploaded into CIBERSORTx working environment (https://cibersortx.stanford.edu, accessed on Feb 1, 2024). LM22 signature matrix, containing 22 immune cell type annotations, was used for deconvolution. The analysis was performed in absolute mode, with 1000 permutations and batch correction enabled.

Multiplex immunofluorescence and image analysis

Immunophenotyping of tumor-infiltrating NK cells was performed by multiplex immunofluorescence (mIF) as described by Silina et al. [43]. Briefly, 3 μm-thick FFPE tissue sections were mounted on HistoBond adhesive microscope slides (Marienfeld, Germany), deparaffinized, and subjected to heat-induced antigen retrieval in Tris-EDTA pH 9.0 buffer. Endogenous peroxidases were inactivated by 0.02 N HCl and unspecific binding was blocked using 4% BSA /0.01% Triton X-100/TBS. Next, the sections were subjected to sequential detection of antigens as shown in Supplementary Table S1: primary antibodies were incubated overnight at 4 °C, followed by HRP-conjugated secondary antibodies for 1 h at room temperature and tyramide-conjugated fluorophores of the 7-plex Opal system (Akoya Biosciences) for 10 min according to the manufacturer’s protocol. Before each subsequent primary antibody incubation, the slides were stripped for 10-min using sodium citrate pH 6.0 buffer in + 95 °C water bath, and quenched for 15-min in 0.02 N HCl. Finally, the tissue sections were incubated with Spectral DAPI (Akoya Biosciences) for 5 min and mounted using Prolong Antifade mounting medium (Vectashield PLUS).

Whole slide scans and 5–10 representative 200X high power fields (HPF) were acquired for each tumor using PhenoImager HT (Akoya Biosciences), and images were processed using Inform v. 2.6 (Akoya Biosciences). The processing included spectral unmixing and automated cell segmentation steps. Per cell fluorescence data were exported and converted into flow cytometry data format (.fcs) by R using Biobase and Flowcore packages. Identification and quantification of various cell phenotypes were performed by supervised signal intensity gating using FlowJo v.10.10.0 software.

Statistical analysis

The nonparametric Mann-Whitney U test was used to compare continuous variables between two groups and ANOVA test - for multiple-group comparison. If multiple variables were tested simultaneously, multiple testing correction was done by FDR estimation and adj. p-value of ≤ 0.05 was considered to be significant. For categorical variables, Chi-Square test was used.

The prognostic significance of all identified DEGs was initially assessed in the Breast Invasive Carcinoma (BRCA) datasets available at The Cancer Genome Atlas (TCGA). Relevant data sets aligned and normalized to TPM values were downloaded from the Gene Expression Omnibus (accession number GSE62944) [44] and analyzed using the Log rank test and Cox proportional hazards regression model using overall survival (OS) as the survival endpoint. Genes with a Log rank test p-value < 0.05 and Hazard ratio > 1.5 or < 0.67 were considered prognostically significant.

The prognostic value of the selected candidate genes was further analyzed using Kaplan-Meier Plotter for BC (gene chip mode), which integrates gene expression and clinical data on 7830 BC cases [45]. Relapse-free survival (RFS) was selected as the survival endpoint and patient samples were split into two groups by “auto-select best cutoff” for the continuous variables. To assess the prognostic value of candidate genes in BC molecular subtypes, PAM50 classification was used. The impact of immunotherapy outcomes was assessed using KM Plotter for immunotherapy, which comprises clinical data on 1434 patients with various cancers treated with anti-PD-1, anti-PD-L1 or anti-CTLA-4 immunotherapy [46].

Results

Exercise enhances the response to neoadjuvant chemotherapy

A total of 50 patients with locally advanced BC were enrolled in the CancerBeat study (Fig. 1a) between January 2022 and June 2023 at Riga East University Hospital. Initially, 32 patients were randomized to the HIIT and 18 patients to the CON group. At the midterm assessment, 6 patients had completed less than 60% of the prescribed training program. These patients remained in the study but were excluded from per-protocol analyses. Twenty participants of the HIIT group completed the prescribed training program with an adherence level of over 60%. On average, only 10–12.5% of sessions involved minor reductions in intensity, where participants exercised slightly below the prescribed HR zone. Six participants from the HIIT group and 1 from the CON group dropped out of the study due to psychoemotional reasons, chemotherapy side effects or unavailability to travel to the testing facility. None of the participants experienced any exercise-related adverse events in this study. Unexpectedly, we observed that a fraction of CON group participants tended to increase their physical activity level, however precise monitoring of activity levels was not feasible. Therefore, we decided to include a retrospective control group (Ctrl) in this study: age-, BC subtype- and stage-matched BC patients from the BCPerMed study, who had received a matched NAC regimen and were treated at Riga East University Hospital between June 2019 and August 2021.

We found that the HIIT group responded to NAC significantly better than the CON and Ctrl groups. The fraction of patients who achieved pCR (Miller-Payne grade 5: no malignant cells identifiable; only vascular fibro-elastic stroma remains [47]) was higher in the HIIT group (50%) as compared to the CON (29%) and Ctrl group (20%). There were no patients with no response in the HIIT and CON groups, whereas 10% of the patients from the Ctrl group did not respond to the NAC (Miller-Payne grade 1: no change or some alterations to individual malignant cells but no reduction in the overall cellularity [47]). The Miller-Payne grades were significantly different between the HIIT and the Ctrl group (p = 0.012), whereas the difference between the HIIT and the CON group did not reach statistical significance (p = 0.068) (Fig. 1b). When the patients were dichotomized, the number of responders (80%) in the HIIT group was substantially higher than in the Ctrl group (50%) (p = 0.047) and the CON group (47%) (p = 0.036) (Fig. 1c).

Exercise increases plasma levels of secreted proteins after NAC

To identify acute exercise-induced changes in plasma proteins, we measured the levels of 92 proteins in plasma samples collected before and immediately after the maximal exercise test using the Olink Immuno-Oncology panel. We analyzed paired Pre- and Post-Run samples from the HIIT group at two distinct time points: diagnosis (HIIT-DG) and following the completion of HIIT/NAC (HIIT-NAC). By comparing Post vs. Pre-Run samples, we identified 2 and 20 exercise-responsive proteins in the HIIT-DG and HIIT-NAC time points, respectively (Fig. 2a, b). A greater number of exercise-induced proteins in the HIIT-NAC samples may suggest that the adaptation to exercise and/or NAC triggers a more prominent response to acute exercise. To determine whether long-term intervention altered baseline proteomic profiles, Pre-Run samples were compared between the HIIT-DG and HIIT-NAC time points; however, no significant differences were detected (Fig. 2c). Similarly, no significant differences were observed between Post-Run samples across these two time points (Fig. 2d).

Fig. 2.

Exercise-induced plasma proteins in HIIT group BC patients. a, Volcano plot showing differentially abundant proteins in Post-Run (Post) vs. Pre-Run (Pre) plasma samples from HIIT group patients at the time of diagnosis. b, Volcano plot showing differentially abundant proteins in Post-Run (Post) vs. Pre-Run (Pre) plasma samples from HIIT group patients after the NAC and training program. c, Volcano plot showing differentially abundant proteins between Pre-Run samples at the time of diagnosis and after the NAC and training program. d, Volcano plot showing differentially abundant proteins between Post-Run samples at the time of diagnosis and after the NAC and training program. Names of top 10 proteins are shown, except for panel b, where all significantly different proteins are shown. Red – adj. p-value < 0.05; blue – non-significant after multiple testing correction. e, Paired dot plots showing the relative abundance of selected proteins in Post-Run vs. Pre-Run plasma samples from HIIT group patients at the time of diagnosis (DG) and after the NAC and training program (NAC). Benjamini-Hochberg adj. p-values are shown

Among the exercise-induced plasma proteins are well-known myokines, including interleukins (IL6, IL7, and IL15) and chemokines (CCL13, CCL17, CX3CL1, CCL2, and CCL8), as well as angiopoietin 2 (ANGPT2) and decorin (DCN), which are released by muscles and play significant roles in exercise adaptation (Fig. 2e; Table 2). Other proteins, such as endothelial nitric oxide synthase 3 (NOS3) [48], granzymes B (GZMB) and H (GZMH) [49], and HGF [50] have also been identified as exercise-induced proteins released in the circulation by various cell types such as endothelial cells, cytotoxic T cells, NK cells, and adipocytes.

Table 2.

Exercise-induced plasma proteins

Gene name	Olink Assay	DG-Post vs. DG-Pre		NAC-Post vs. NAC-Pre		NAC-Pre vs. DG-Pre		NAC-Post vs. DG-Post		Myokine status	Functions*
		ΔNPX	Adj.p	ΔNPX	Adj.p	ΔNPX	Adj.p	ΔNPX	Adj.p
ANGPT2	ANGPT2	0.33	0.07	0.24	0.03	0.09	0.73	-0.02	0.92	Yes [51]	Angiopoietin 2, antagonist of angiopoietin 1, regulation of angiogenesis
CCL13	MCP-4	0.43	0.17	0.37	0.03	-0.13	0.76	-0.22	0.69	Yes [52]	Chemokine ligand 13, a chemoattractant for macrophages, eosinophils, monocytes, allergic inflammation, non-allergic inflammation, CCR3 receptor
CCL17	CCL17	0.48	0.15	0.40	0.03	0.04	0.93	-0.03	0.94	Yes [52]	Chemokine ligand 17, T cells trafficking and activation, CCR4, CCR8 receptors, pain, wound healing, fibroblast migration, pro-inflammatory
CCL2	MCP-1	0.31	0.17	0.33	0.02	0.13	0.73	-0.03	0.94	Yes [52–54]	Chemokine ligand 2, a chemoattractant for monocytes and basophils, CCR2 receptor, monocytic infiltrates
CCL8	MCP-2	0.27	0.17	0.27	0.02	-0.17	0.68	-0.25	0.64	Yes [52]	Chemokine ligand 8, a chemoattractant for monocytes, lymphocytes, basophils and eosinophils
CX3CL1	CX3CL1	0.23	0.14	0.18	0.04	0.13	0.67	-0.01	0.95	Yes [54]	C-X3-C Motif Chemokine Ligand 1, fractalkine, ligand for CX3CR1 and integrins ITGAV: ITGB3 and ITGA4:ITGB1, implicated in immune response, inflammation, cell adhesion and chemotaxis
DCN	DCN	0.23	0.17	0.25	0.01	-0.05	0.83	-0.07	0.73	Yes [55]	Decorin, proteoglycan, collagen fibril assembly, tumor suppression, autophagy, angiogenesis inhibition
GZMB	GZMB	0.67	0.07	0.81	0.01	0.17	0.54	0.22	0.63	Not known	Granzyme B, serine protease in the cytosolic granules of cytotoxic T-cells and NK-cells
GZMH	GZMH	0.47	0.15	0.68	0.03	0.06	0.88	0.18	0.69	Not known	Granzyme H, cytotoxic chymotrypsin-like serine protease, expressed in the NK cells
HGF	HGF	0.33	0.11	0.24	0.03	-0.05	0.84	-0.17	0.65	Not known	Hepatocyte growth factor, regulate cell growth, cell motility and morphogenesis, involved in angiogenesis, tumorigenesis, and tissue regeneration
IL15	IL15	0.17	0.25	0.30	0.02	-0.13	0.54	-0.13	0.73	Yes [56, 57]	Interleukin 15, cytokine, stimulates the proliferation of NK cells, T-cells and B-cells and promotes the secretion of cytokines
IL6	IL6	0.75	0.01	0.61	0.01	0.40	0.54	0.14	0.76	Yes [58, 59]	Interleukin 6, wide variety of biological functions in immunity, tissue regeneration, and metabolism
IL7	IL7	0.24	0.38	0.34	0.02	-0.28	0.68	-0.21	0.69	Yes [60]	Interleukin 7, hematopoietic cytokine, development, expansion, and survival of naive and memory T-cells and B-cells
KIR3DL1	KIR3DL1	-0.03	0.82	0.30	0.03	0.06	0.86	0.29	0.62	Not known	Killer cell immunoglobulin like receptor, inhibitory receptor on NK cells for HLA Bw4 allele
LGALS1	Gal-1	0.24	0.07	0.20	0.02	0.04	0.83	-0.06	0.78	Not known	Galectin 1, lectin that binds beta-galactoside and complex carbohydrates, regulates apoptosis, cell proliferation and differentiation
MMP7	MMP7	0.22	0.07	0.30	0.03	-0.13	0.75	-0.03	0.88	Not known	Matrix metalloproteinase 7, breakdown of extracellular matrix, implicated in wound healing, tumor invasion, tissue repair
NOS3	NOS3	0.54	0.04	0.57	0.03	0.29	0.41	0.17	0.74	Not known	Nitric oxide synthase 3, expressed by adipocytes and endothelial cells, NO is implicated in vascular smooth muscle relaxation, neurotransmission and antimicrobial and antitumoral activities.
PTN	PTN	0.50	0.12	0.56	0.01	-0.09	0.76	-0.10	0.82	Not known	Pleiotrophin, secreted growth factor, regulates cell proliferation, survival, growth, differentiation and migration
TNFRSF12A	TNFRSF12A	0.32	0.12	0.25	0.03	0.10	0.73	-0.10	0.77	Not known	TNF receptor superfamily member 12 A, TNF signalling, inducer of apoptosis, regulation of wound healing
TNFSF14	TNFSF14	0.30	0.16	0.31	0.01	-0.17	0.68	-0.15	0.69	Not known	TNF superfamily member 14, cytokine, ligand for TNFRSF14, a costimulatory factor for the activation of T cells

* keywords from the Gene (https://www.ncbi.nlm.nih.gov/gene) and The Human Protein Atlas (https://www.proteinatlas.org). ΔNPX, difference in the relative abundance of the protein between the groups. Adjusted p < 0.05 are marked in bold

In contrast, pleiotrophin (PTN), NK cell inhibitory receptor KIR3DL1, Galectin 1 (LGALS1), TNF receptor (TNFRSF12A) and MMP7 have not previously been implicated in the response to acute exercise. This suggests that the induction of these proteins may be linked to the presence and/or treatment of BC. Intriguingly, the levels of tumor necrosis factor superfamily member 14 (TNFSF14) have been found to decrease in survivors of childhood acute lymphoblastic leukemia after a 16-week exercise intervention [61].

Taken together, we identified 20 immuno-oncology-related plasma proteins that increased in the blood during acute exercise in BC patients undergoing NAC. While most have been previously described as exercise-induced, five of these proteins have not been linked to exercise before, implying that their increase might be associated with the presence and/or treatment of BC. However, it has to be taken into account that the changes in plasma proteins were measured after the maximal exercise test, which differs in the modality, intensity, and duration from the HIIT sessions performed during the intervention.

Exercise increases transcription of genes associated with improved outcomes

To assess the long-term impact of the HIIT program on the BC transcriptome, paired tumor (T) and normal (N) tissue specimens from 10 patients from the HIIT group and 10 patients from the Ctrl group were subjected to bulk RNA sequencing (RNA-seq) analysis. On average, 22.4 million raw reads were obtained per RNA-seq library, and approximately 69–95% of them were uniquely mapped against the human reference genome (GRCh38). Uniquely mapped reads were retained for further analysis and multi-mapped reads were excluded. To identify genes that are differentially regulated by exercise, two different comparisons were performed. Firstly, tumor tissues from the HIIT group (HIIT-T) were compared against tumor tissues from the Ctrl group (Ctrl-T). Secondly, tumor tissues were compared against normal tissues in each group, and then the identified DEGs were compared between the HIIT and Ctrl groups.

Comparison between HIIT-T and Ctrl-T groups revealed 309 DEGs (|Log2FC| >1, adj. p < 0.05) − 56 of them had higher expression levels in the HIIT-T, 253 – in the Ctrl-T group (Fig. 3a; Supplementary Table S2). On the contrary, only 4 DEGs were found between HIIT-N and Ctrl-N groups (data not shown), suggesting that the intervention primarily affected the tumor tissues. Survival analysis of the TCGA dataset showed that higher expression of 7 out of the 56 (12.5%) HIIT-T upregulated genes was associated with longer overall survival (HzR < 0.67; log-rank p < 0.05), whereas 22 of the 253 (8.7%) Ctrl-T upregulated genes were associated with poor prognosis (HzR > 1.5; log-rank p < 0.05) and only three (0.4%) with favorable prognosis.

Fig. 3.

Exercise-induced alterations in gene expression profiles of BC tissues. a, Volcano plot showing DEGs between tumor tissues from HIIT group (HIIT-T) and Ctrl group (Ctrl-T). The prognostic value of DEGs was assessed by log-rank test in TCGA BC datasets. DEGs with favorable prognostic value are highlighted in green (HzR<0.67; log-rank p<0.05), poor prognostic value in red (HzR<1.5; log-rank p<0.05), and no prognostic significance in blue. b, Box plots showing the validation of differential expression of selected DEGs by qRT-PCR. Statistical significance was evaluated by the Mann-Whitney U test. Box plots show mean and 0.25–0.75 quantiles, whiskers show the range of the data within 1.5 times the interquartile range (IQR) from the quartiles, displaying the spread of the bulk of the data, while points outside this range are considered outliers. Dots represent the mean of duplicates for individual samples. c, Comparison of relapse-free survival between high and low-gene expression groups stratified by the auto-select best cutoff by Kaplan Meier plotter and log-rank test. d, Gene set enrichment analysis of differentially expressed genes between tumor tissues from the HIIT vs Ctrl groups. Heatmaps show DEGs representing a response to wounding and aerobic respiration and gene set enrichment scores

Differential expression of 8 genes (OPRPN, PIGR, CCDC8, ILF2, ANLN, ESRP1, SNORD114-6 and SNORD81) was validated by qRT-PCR in the same sample set used for the RNA-seq analysis (Fig. 3b). OPRPN, PIGR, CCDC8, SNORD114-6 and SNORD81 were upregulated in the HIIT-T group, whereas ILF2, ANLN and ESRP1 had higher expression levels in Ctrl-T. Survival analysis using Kaplan-Meir plotter and log-rank test showed that high expression of OPRPN, PIGR and CCDC8 is associated with favorable prognosis (longer RFS) in all BC subtypes together (Fig. 3c). When BC subtypes were analyzed separately, both genes showed favorable prognostic value in basal, luminal B, HER2 + and normal-like BC but not in the luminal A subtype according to the PAM50 classification (Supplementary Fig. S1). On the contrary, high expression of ILF2, ANLN and ESRP1 was associated with shorter RFS in all BC subtypes together (Fig. 3c). ILF2 was significantly associated with poor prognosis in all subtypes except for basal BC, while ANLN and ESRP1 in all subtypes except normal-like BC (Supplementary Fig. S1).

Gene set enrichment analysis (GSEA) using MSigDB C5 collection containing gene sets from the Gene Ontology resource (GO) revealed that HIIT-T were enriched in genes implicated in vasculature development and wounding, while Ctrl-T were enriched for aerobic respiration, nucleosome assembly, DNA packaging and chromatin organization, most likely representing higher proliferative activity of Ctrl tumors (Fig. 3d; Supplementary Fig. S2a, b).

While the response to wounding was associated with both HIIT-T and Ctrl-T, the corresponding genes differed. In the HIIT-T, 8 genes were upregulated (MMRN1, PTN, EMILIN2, LOX, DST, ANXA1, CD34, DPYSL3) and 8 genes were downregulated (CERS2, PTPRF, IGF1R, HSPB1, SPP1, CLDN3, HBB, S100A8) as compared to Ctrl-T. Intriguingly, scRNAseq data set at the Human Protein Atlas showed that the HIIT-T upregulated genes are predominantly expressed in the breast tissue fibroblasts (EMILIN2, LOX, DST, ANXA1, CD34, DPYSL3, PTN), endothelial cells (MMRN1, LOX, DST, DPYSL3) and macrophages (EMILIN2, ANXA1) [62]. On the contrary, Ctrl-T upregulated genes are predominantly expressed by breast glandular cells (CERS2, PTPRF, IGF1R, CLDN3, SPP1), a subpopulation of macrophages (S100A8), erythrocytes (HBB) or ubiquitously expressed by multiple cell types (HSPB1) [62].

Taken together, these data suggest that combining NAC with physical exercise induces changes in the tumor gene expression profile, promoting less aggressive phenotypes and favorable outcomes, and reshaping the wound healing process at the tumor site. Additionally, the findings suggest that breast fibroblasts, endothelial cells, and macrophages may play a role in sensing and responding to physical activity.

Exercise boosts immune gene signatures in tumors

Comparison between tumor and normal tissues revealed 158 and 320 DEGs in the HIIT and Ctrl groups, respectively (Fig. 4a; Supplementary Table S3, Table S4). Eighty-two and 203 genes were upregulated in HIIT-T and Ctrl-T, respectively; However, only 4 of them overlapped (Fig. 4b). A substantial part of the HIIT-T vs. HIIT-N upregulated genes were immune-related genes encoding T cell receptors, immunoglobulins, chemokines and chemokine receptors (e.g., CCL18, CCR1, CXCR4), immune cell surface antigens (e.g., CD4, CD52, CD53, CD84), receptors or effector molecules (e.g., MSR1, CHIT1, TYROBP, FCER1G, LTBP1, TREM2, etc.). In contrast, immune-related genes (e.g., CXCL9-11, CXCL13, CCL19, IL4I1, etc.) constituted only a minor fraction of all DEGs in the Ctrl-T vs. Ctrl-N comparison. Here, the top upregulated genes included histones, collagens, matrix metallopeptidases and several oncogenes. Upregulation of LTBP1 and TREM2 in HIIT-T vs. HIIT-N, CXCL9 in Ctrl-T vs. Ctrl-N and CXCL10 in both comparisons was validated by qRT-PCR (Fig. 4c). Although the difference did not reach statistical significance for some comparisons, the expected trend was observed in all cases.

Fig. 4.

Comparison of DEGs between tumor and normal tissues from the HIIT and Ctrl groups. a, Volcano plots showing DEGs between HIIT-T vs HIIT-N and Ctrl-T vs Ctrl-N. b, Venn diagram showing the number and overlap of upregulated and downregulated DEGs between HIIT-T vs HIIT-N and Ctrl-T vs Ctrl-N. c, Validation of differential expression of selected DEGs by qRT-PCR. Statistical significance was evaluated by the Mann-Whitney U test. Box plots show mean and 0.25–0.75 quantiles, whiskers show the range of the data within 1.5 times the interquartile range (IQR) from the quartiles, displaying the spread of the bulk of the data, while points outside this range are considered outliers. Dots represent the mean of duplicates for individual samples. Tumor tissue (T). Normal tissue (N). d, Heatmap of DEGs representing “immune response” pathway and gene set enrichment score in HIIT-T vs N. Color code of sample groups: dark green – HIIT-N, light green – HIIT-T. e, Heatmap of DEGs representing “adaptive immune response” pathway and gene set enrichment score in Ctrl-T vs N. Color code of sample groups: dark red – Ctrl-N, light red – Ctrl-T

GSEA indicated that the HIIT-T group was enriched for genes involved in adaptive and innate immune response activation, while pathways related to lipid, organic acid, and small molecule metabolism appeared downregulated relative to the HIIT-N group (Fig. 4d, e; Supplementary Fig. S3a). Notably, distinct gene sets associated with lipid metabolism were active in both HIIT-T and HIIT-N, particularly among patients with Miller-Payne grades 4 and 5 (Supplementary Fig. S2c). In the Ctrl-T group, the most prominently activated biological processes involved nucleosome assembly and chromatin organization, potentially reflecting higher proliferative activity; conversely, various biosynthetic and metabolic processes were downregulated compared to Ctrl-N. While a cluster of immune-related pathways was also identified in the Ctrl-T group, these gene sets differed substantially from those observed in the HIIT-T group (Fig. 4e, Supplementary Fig. S3b, Fig.S2d).

Taken together, these results highlight significant differences in the immunological processes occurring in the tumor as compared to normal tissues between patients from the HIIT and Ctrl groups.

Immune cell deconvolution analysis suggests changes in the NK cell activation status

To estimate the composition of tumor-infiltrating immune cells, we employed CIBERSORTx to computationally infer cell fractions from bulk tissue gene expression profiles. In both the HIIT-T and Ctrl-T groups, tumors exhibited substantial predicted infiltration of CD4 + and CD8 + T cells, Tregs, B cells, and macrophages compared to adjacent normal tissue (Supplementary Fig. S4), consistent with typical cancer-related immune remodeling. Further in silico analysis revealed distinct transcriptomic variations between the HIIT-T and Ctrl-T groups, which likely reflect the activity states of infiltrated NK cells, CD4 + T cells, and macrophages (Fig. 5). While total NK cell infiltration appeared comparable across groups, the HIIT-T group was characterized by a significantly higher inferred proportion of activated NK cells, whereas the Ctrl-T group showed a higher frequency of resting NK cells. Additionally, the Ctrl-T group displayed an increased expression profile computationally associated with activated memory CD4 + T cells and M1 macrophages.

Fig. 5.

Alterations in the CD4+ T cell, macrophage and NK cell activation status in HIIT-T and Ctrl-T identified by CIBERSORTx deconvolution analysis of bulk RNA-seq data. Box plots show mean and 0.25–0.75 quantiles, whiskers show the range of the data within 1.5 times the interquartile range (IQR) from the quartiles, displaying the spread of the bulk of the data, while points outside this range are considered outliers, dots represent the individual samples

Next, we attempted to validate the density and activation status of tumor-infiltrating NK cells by mIF using a 7-plex marker panel in the tumor tissue sections from the same cohort of patients used for RNA-seq analysis. The panel included CD45 as pan-immune cell marker, CD3 as pan-T cell marker), pan-CK (pan-cytokeratin) to mark tumor and normal epithelial cells, CD56, a cell adhesion molecule predominantly expressed in NK cells and neurons [63, 64], and three activation markers or effector molecules: granzyme B (GZMB) [64], lymphotoxin alpha (LTA) [65] and CD57 [66]. To detect all these molecules in the same tissue section, we used the cyclic tyramide amplification-based staining approach, which, while being highly sensitive, proved damaging to multiple samples. We detected the presence of both resting NK cells (pan-CK⁻CD45⁺CD3⁻CD56⁺CD57⁻GZMB⁻LTA⁻) and activated NK cells (pan-CK⁻CD45⁺CD3⁻CD56⁺CD57⁺GZMB⁺LTA⁺) in remaining samples with well-preserved tissue morphology. (Supplementary Fig. S5). However, due to staining-related tissue damage and the limited number of evaluable samples, mIF could not provide a statistically powered between-group comparison of NK-cell phenotypes. Therefore, the observed group differences in NK-cell activation status should be interpreted primarily as CIBERSORTx-derived transcriptomic inferences, with mIF providing qualitative confirmation that both resting and activated NK-cell phenotypes were present in the analyzed tissues.

Decorin and pleiotrophin are increased both in plasma and tumor tissue and are associated with better outcomes

Finally, we sought to identify the overlap between exercise-induced changes in plasma protein levels and tumor gene expression. We found that decorin (DCN) and pleiotrophin (PTN) protein levels increased in plasma during acute exercise (Fig. 2b, e), and their mRNA expression was upregulated in tumor tissues following the long-term HIIT intervention (Figs. 3a and 6a). We then stratified patients (combining HIIT and Ctrl groups) as responders or non-responders to NAC and compared tumor gene expression between these groups. This analysis identified 154 genes upregulated in tumors from responders. Of these, 16 overlapped with genes upregulated in HIIT-T compared to Ctrl-T, with only one gene — DCN — also overlapping with exercise-induced plasma proteins (Fig. 6a).

Fig. 6.

Overlap between exercise-induced plasma proteins and mRNAs in breast cancer tissues. a, Venn diagram showing the overlap between exercise-induced plasma proteins in Pre- vs Post-run HIIT-NAC samples, mRNAs upregulated in the HIIT-T vs Ctrl-T, and mRNAs upregulated in responders (R) vs non-responders (NR). b, Box plots showing the relative expression levels of DCN and PTN mRNAs in tumor and adjacent normal breast tissues from the HIIT and Ctrl groups analyzed by qRT-PCR. Statistical significance was evaluated by the Mann-Whitney U test. c-h, Survival analyses performed by Kaplan-Meier plotter. c, the correlation between DCN level and relapse-free survival in BC patients (all subtypes). d, The correlation between DCN level and overall survival in patients with various cancers treated with anti-PD1 immunotherapy. e, The correlation between DCN level and overall survival in patients with various cancers treated with anti-CTLA-4 immunotherapy. f, The correlation between PTN level and relapse-free survival in BC patients (all subtypes). g, The correlation between PTN level and overall survival in patients with various cancers treated with anti-PD1-L1 immunotherapy. h, The correlation between PTN level and overall survival in patients with various cancers treated with anti-CTLA-4 immunotherapy. In all analyses, patients were stratified using auto-select best cutoff

qRT-PCR analysis confirmed the overexpression of DCN in HIIT-T compared to HIIT-N and Ctrl-T suggesting that exercise stimulated its expression specifically in the tumor tissues (Fig. 6b). scRNAseq data at the Human Protein Atlas show that DCN is mainly expressed in breast fibroblasts [62] (Supplementary Fig. S6). Survival analysis using Kaplan-Meier plotter showed that higher levels of DCN are associated with better RFS in all BC cases together (Fig. 6c). Moreover, it was associated with better overall survival (OS) in patients with various cancers treated with anti-PD-1 and anti-CTLA-4 immunotherapy (Fig. 6d, e).

PTN mRNA was significantly downregulated in the tumor tissue compared to normal tissues in the Ctrl group and increased in the HIIT tumors (Fig. 6b). Survival analysis showed that high expression of PTN was associated with favorable RFS in all BC cases together (Fig. 6f), better OS in patients with various cancers treated with anti-PD-L1 (Fig. 6g) and anti-CTLA-4 (Fig. 6h), but not anti-PD-1 immunotherapy.

Taken together, our findings reveal that DCN and PTN protein levels increase in the blood in response to acute exercise. Meanwhile, regular exercise promotes their mRNA expression in tumor tissues, with higher expression levels correlating with improved RFS in BC patients and improved response to immunotherapy in patients with various cancers.

Discussion

We conducted a randomized pilot trial to assess the effect of a regular HIIT program on response to NAC in patients with locally advanced BC, and to gain insight into the molecular mechanisms behind the effects of exercise. Multiple studies, including ours, have demonstrated the feasibility and beneficial effects of exercise on the physical fitness, cognitive function and quality of life of BC survivors [67–69]. However, only a few studies have reported the effects of exercise on the NAC efficacy, and the results are controversial. In line with our study, Sanft T. et al. showed that a healthy diet and exercise intervention during NAC was associated with significantly higher pCR rates in HR⁺/HER2⁻ and TNBC patients as compared to NAC only (53% vs. 28%; p = 0.037) [70]. Another study showed that moderate physical activity during NAC reduced the levels of several pro-inflammatory cytokines that may contribute to the higher rate of pCR [71]. On the contrary, two other studies found no difference in the pCR rate between exercise and control groups [72, 73].

These differences might be related to the type, intensity and duration of exercise, as well as the stage and subtype of BC and NAC regimen. The fact that the reported pCR rates in the control groups varied from 28% [70, 73] to 55.9% [72] suggests that the selection of patients and treatment regimen plays a greater role. Other factors that may affect the results of the clinical studies are patients’ adherence to the training programs and drop-out rates. We observed that a fraction of patients experienced difficulties in adhering to the training program during NAC and either withdrew from the study or remained on the trial but were excluded from per-protocol analyses. This may have led to the selection of physically fittest patients remaining in the HIIT group. At the same time, we observed that a fraction of patients from the prospective control group were inclined to increase their physical activity levels during the trial, which likely influenced their clinical response to NAC. To maintain a clear distinction between exercise and non-exercise cohorts, we opted to utilize a retrospective control group as a more stable and biologically ‘clean’ reference for the molecular analyses. These biases and deviations from the initial plan represent the main limitations of our study. Therefore, a definitive appropriately powered randomized trial is warranted to minimize potential biases and accurately measure the effect size of physical exercise on the efficacy of NAC.

To date, most insights into the cellular and molecular mechanisms driving the protective effects of exercise on cancer incidence and progression have come from animal studies, whereas only a few studies have been performed in humans. In the PreHAB study, BC patients were randomized to an exercise or mind-body control group and the intervention on average lasted for 29.3 days between enrollment and surgery [74]. While RNA sequencing showed no significant differences between the exercise and control group tumors, pathway analysis revealed an upregulation of immunity and inflammation-related pathways in surgical specimens compared to baseline biopsies in the exercise but not the control group. Consistent with these findings, our study demonstrates that the majority of genes overexpressed in tumors compared to normal tissues in the HIIT group were associated with immune responses. In contrast, only a few immune-related genes were overexpressed in the Ctrl group tumors versus normal tissues, suggesting that the immune cell composition and activation status were significantly altered in patients who participated in the exercise program. However, the same DEGs were not found when HIIT tumors were compared to Ctrl tumors. This could be attributed to the larger number of dysregulated genes observed in Ctrl tumors or the higher overall abundance of tumor-infiltrating immune cells in the Ctrl tumors, which may obscure differences in immune gene expression. Supporting this, CIBERSORTx immune cell deconvolution analysis suggested a notable distinction in NK cell activation status: while HIIT tumors appeared to be infiltrated with activated NK cells, Ctrl tumors predominantly contained resting NK cells.

Consistent with previous studies, we observed increased levels of IL6 and other exerkines in plasma following acute exercise in BC patients [58, 59]. Hence, our findings align with studies in mouse tumor models, which demonstrate that exercise suppresses tumor growth by enhancing NK cell infiltration in an IL6-dependent manner [16] and promoting the activation of NK and CD8⁺ T cells [15]. However, it is important to note that five patients from the HIIT group and three from the control group had achieved pCR by the time of surgery. In these cases, the anti-tumor immune response may have already resolved, leaving only remnants of the immunological activity visible at the time of tissue collection. Consequently, the specific contribution of NK cells to tumor suppression and the potential involvement of other immune cell populations in initiating the anti-tumor response remain open questions for further investigation.

In contrast to the PreHAB study [74], we observed substantial differences in gene expression between the HIIT and control tumor tissues. These divergent results may be primarily attributed to the longer duration of our intervention and the synergistic effect of combining exercise with NAC. The most strongly upregulated gene in HIIT vs. Ctrl tumors was OPRPN, which encodes opiorphin prepropeptide. We found that its expression is lost in the Ctrl tumors compared to the adjacent normal breast tissues, while in the HIIT group it has similar levels in the tumor and normal tissues. This suggests that exercise might have restored OPRPN expression in tumors and scar tissue. Importantly, high expression of OPRPN is associated with better outcomes in BC patients. Opiorphin is a pentapeptide that inhibits the enkephalin-inactivating peptidases, thus acting as a potent analgesic with a greater pain-killing effect than morphine [75, 76]. The Human Protein Atlas shows that its expression pattern in females is restricted to salivary glands and breast tissue, where it is produced by glandular epithelial cells [62]. Upregulation of OPRPN in the BC tissue is likely to result in the activation of enkephalinergic opioid receptors that are expressed not only by neural cells but also by cancer cells [77]. This may lead not only to the modulation of pain perception and prevention of hyperalgesia, but also trigger diverse effects on cancer progression [78, 79] and immune system function [78, 80].

Cancers have long been described as “wounds that do not heal” [81, 82]. While there are numerous molecular similarities between cancer and the wound-healing process, the key distinction lies in the dysregulation of the sequential phases — inflammation, new tissue formation, and tissue remodeling — which are tightly coordinated during normal wound healing but become chronic and uncontrolled in cancer [83]. Many of wounding-related DEGs, such as ANXA1 [84], LOX [85] and EMILIN2 [86], are typically involved in the later stages of wound healing, contributing to the resolution of inflammation, tissue regeneration, and ECM remodeling. In contrast, several genes that are downregulated in HIIT-T, such as SPP1, also known as osteopontin [87] and S100A8 [88] are implicated in promoting inflammation. Overall, these results indicate towards distinct wound‑healing dynamics in HIIT versus control tumors. Exercise appears to prompt fibroblasts, endothelial cells, and other stromal populations to reprogram their transcriptional profiles, thereby accelerating tissue repair and resolving inflammation.

Furthermore, our findings show that DCN protein levels in the blood increased following acute exercise. Concurrently, DCN mRNA expression was elevated in the tumor tissue of exercising BC patients compared to controls, and in patients who responded to NAC compared to non-responders. In line with previous studies, we also observed that higher tumor DCN mRNA levels were associated with a favorable prognosis in ER + BC [89, 90], as well as with better outcomes in patients with various types of cancer, including melanoma and glioblastoma, treated with anti-PD1 and anti-CTLA-4 immunotherapy.

DCN encodes decorin – a small, secreted proteoglycan involved in collagen fibril assembly [91]. Decorin also interacts with various cell surface receptors, including TGF-β receptors and various tyrosine kinase receptors, resulting in the suppression of downstream signaling cascades and growth inhibition [92]. Intratumoral delivery of decorin has been shown to inhibit tumor growth and metastasis [93, 94]. Additionally, decorin is released into circulation during exercise from skeletal muscle [55, 89]. It seemed plausible that decorin released by contracting muscle could enter BC tissues, thus contributing to the beneficial effects of exercise on delaying the progression of BC. However, a recent study by Hjorth et al. (2024) found that muscle-derived plasma decorin could not penetrate into tumor or normal breast tissue and consequently did not affect the growth of the primary tumor or the formation of metastases [89]. scRNAseq data indicate that DCN is predominantly expressed in mesenchymal cells, particularly fibroblasts and testicular Leydig cells (Supplementary Fig. S6) [62, 89]. Consequently, tissue-resident fibroblasts appear to be the primary source of decorin across various tissues, including the breast. Our findings suggest that BC-associated fibroblasts may upregulate decorin in response to physical activity. However, it remains possible that the elevated DCN mRNA levels observed in the HIIT-T group reflect alterations in fibroblast proportions or activation states within post-NAC tissue specimens from the HIIT group. Furthermore, while DCN expression correlates with improved clinical outcomes, whether decorin functionally promotes a less aggressive phenotype or serves as a molecular surrogate for a favorable prognosis remains to be determined.

Building on the exploratory findings of this study and prior literature, we propose the following hypothetical conceptual model by which exercise affects tumor biology (Fig. 7). Exercise-induced leukocytosis, a transient increase in total blood leukocyte counts, is a well-documented phenomenon that occurs within the first 20–30 min of dynamic exercise and diminishes 1–2 h post-exercise [95, 96]. This process is driven by increased blood pressure, shear forces and stimulation of β2-adrenergic receptors on the surface of leucocytes that leads to the detachment of leucocytes from the endothelium of reservoirs such as bone marrow, spleen and lymph nodes and mobilization of immune cells in the bloodstream [95]. The most responsive cell types are NK cells, followed by CD8⁺ T cells, γδ T cells, monocytes and neutrophils. The NK cell counts can increase up to 10-fold during exercise [97]. Moreover, higher differentiation status (i.e. KIR⁺/NKG2A⁻) was associated with greater mobilization and CD57⁺ NK-cells were mobilized more efficiently than CD57⁻ NK-cells. Importantly, the exercise-mobilized NK cells showed 1.6-fold higher cytotoxic activity against myeloma, lymphoma and leukemia cells than pre-exercise cells [96]. In support of this, a recent clinical trial in breast and colon cancer patients showed that regular exercise helps to maintain NK cell degranulation and cytotoxic activity during NAC, while it decreased in the control group [98]. It seems reasonable to assume that each bout of exercise could lead to transient mobilization of activated tumor-infiltrating NK cells, CD8⁺ T cells and other cytotoxic immune populations, potentially contributing to immune-permissive TME, although cumulative immune cell recruitment and tumor rejection were not directly demonstrated in this study.

Fig. 7.

Hypothetical conceptual model of exercise-associated tumor modulation. The model summarizes exploratory systemic responses and tumor microenvironment alterations observed in BC patients performing HIIT exercise during NAC, together with mechanisms proposed from prior literature. Image created in BioRender

Meanwhile, CAFs, other stromal cells and cancer cells may respond to exercise by modulating their gene expression, leading to enhanced wound healing response [99], remodeling TME, interactions with sensory neurons, and tumor growth suppression. In our study, the enrichment of stromal and wound-healing-related genes in HIIT-T tumors is consistent with the possible involvement of CAFs, endothelial cells, and other stromal compartments. Nevertheless, because our observations are based on bulk tumor transcriptomes, the relative contribution and activation states of individual stromal cell populations remain to be resolved. How stromal cells sense exercise remains an open question. Fibroblasts and endothelial cells are known as mechanosensing cells that detect deformations, shear force, tension, traction forces, or ECM stiffness, which trigger chromatin remodeling and gene expression changes [100, 101]. These cells may directly sense increased blood flow or body movement and respond accordingly. Alternatively, they could react to circulating exerkines or exercise-induced extracellular vesicles released during exercise.

Last but not least, several exerkines such as IL6, IL15, irisin, decorin, SPARC and BDNF can have direct effects on inhibiting the growth of primary tumors and metastases [59, 102].

Conclusions

Our study shows that regular physical exercise during NAC is associated with favorable treatment outcomes in patients with locally advanced BC and, for the first time, offers exploratory insights into potential molecular mechanisms underlying this effect. Specifically, our findings highlight CAFs and other stromal cells as putative effector cells that may sense and respond to exercise by secreting various molecules that could promote wound healing and inhibit cancer progression. While these results are preliminary, they provide a conceptual framework to support the integration of physical activity into BC treatment plans, and emphasize the need for clinical trials investigating the impact of HIIT on the efficacy of immunotherapy.

Abbreviations

BC

Breast cancer

CAF

Cancer-associated fibroblast

CON

Prospective control group

Ctrl

Retrospective control group

DEG

Differentially expressed gene

GO

Gene ontology

GSEA

Gene set enrichment analysis

HR

Heart rate

HIIT

High-intensity interval training

N

Normal tissue

NAC

Neoadjuvant chemotherapy

NK

Natural killer

pCR

Pathological complete response

snoRNA

Small nucleolar RNA

T

Tumor tissue

TME

Tumor microenvironment

Authors' contributions

AL, KS, JE, AK and ALl designed research, DS, KE, AM, BMG, EBC, LS, AB, FR, KS, SR, KG, TZ, MČ, RC and ILK performed research and participated in the analysis and interpretation of the results, PZ and AM performed the RNAseq data analysis and statistical analyses. AL, AM and DS wrote the manuscript, KS, AK and ALl edited the manuscript. KE, TZ and JE organized the recruitment of patients, and collected the clinical samples and information. All authors have read and approved the manuscript.

Funding

This work was funded by EEA and Norway Grants under project No EEA-RESEARCH-164 (CancerBeat) to AL, ALl, AM, AK, and Swiss National Science Foundation (SNSF) Prima grant (nr. PR00P3 201656) to KS.

Data availability

The RNA-seq datasets are submitted in ArrayExpress, accession E-MTAB-14051. Link for reviewers (without login): [https://www.ebi.ac.uk/biostudies/arrayexpress/studies/E-MTAB-14051?key=73eee47e-2396-4751-81b8-d27ec552dfc4].

Declarations

Ethics approval and consent to participate

The study was conducted according to the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.