Exercise-associated epigenetic remodeling and TCR repertoire dynamics in Lynch syndrome carriers

Summary

Lynch syndrome (LS) carriers are at elevated cancer risk. Emerging evidence suggests that exercise may serve as a non-pharmacologic preventive strategy, yet the epigenetic and immunological mechanisms underlying its protective effects in this population remain unclear. Here, we perform integrative multi-omics profiling of DNA methylation, gene expression, and the T cell receptor (TCR) repertoire in LS carriers undergoing a 52-week aerobic cycling intervention. We identify compartment-specific DNA methylation changes, including innate immune activation in cfDNA and oncogenic pathway repression in tissue. Integrative transcriptomic analysis highlights ISL1 as a key exercise-repressed, epigenetically regulated gene, and identifies FLCN as a colorectal cancer (CRC)-associated methylation target. TCR analysis reveals an exercise-associated increase in systemic repertoire diversity and tissue-specific clonal convergence, thus suggesting antigen-driven recruitment. Collectively, these findings uncover epigenetic and immune remodeling as potential mechanisms of exercise-mediated protection in LS.

Graphical abstract

Highlights

• •Exercise remodels DNA methylation toward immune activation and oncogenic suppression
• •ISL1 emerges as an epigenetically regulated key exercise-responsive gene in LS carriers
• •Exercise reshapes TCR diversity and clonality in blood and colonic tissue in LS carriers
• •Exercise promotes tissue enrichment of antigen-driven TCR clonotypes in LS carriers

Immunology; Epigenetics; Cancer

Introduction

Lynch syndrome (LS), one of the most common hereditary cancer syndromes with autosomal dominant inheritance, predisposes carriers to colorectal cancer (CRC), endometrial cancer (EC), and other tumors due to germline alterations in the DNA mismatch repair (MMR) genes.^1,2 Individuals harboring a defective MMR system have a 40%–80% greater risk of developing CRC compared to the general population.^3,4 Therefore, identifying effective minimally invasive strategies to mitigate CRC risk in LS carriers is a critical focus in clinical and translational research.

Aerobic exercise has been extensively considered to be a potential prophylactic and therapeutic intervention for various disorders in the general population, and accumulating evidence also links it to a reduced risk of CRC in high-risk populations.^5,6,7 In line with this, a large cohort study revealed that higher levels of physical activity in individuals with LS were associated with a lower risk of CRC.⁸ Furthermore, we have previously demonstrated that 52 weeks of aerobic cycling in LS carriers led to enhanced anti-tumor immunity in the colonic mucosa and reduced inflammation, thus suggesting that exercise may serve as a valid non-pharmacologic preventive strategy in this high-risk population.⁹ To further elucidate the molecular mechanisms underlying these effects, we leveraged the same cohort of LS carriers to comprehensively profile exercise-induced changes in the epigenetic landscape and T cell receptor (TCR) repertoire dynamics.

DNA methylation, a key epigenetic mechanism regulating gene expression, is implicated in cancer and shaped by genetic, environmental, and lifestyle factors, including physical activity.^10,11,12,13 In cancer, methylation patterns serve as stable biomarkers of cell-of-origin and disease state.¹⁴ Despite this, the functional impact of the methylation changes is highly context-dependent, as changes in CpG methylation only exert meaningful regulatory effects within the appropriate cellular and tissue environment.¹¹ This context specificity is essential for driving downstream gene expression changes and long-term phenotypic adaptations. In high-risk populations such as LS carriers, early methylation changes may precede histological abnormalities, making DNA methylation profiles in colonic tissue and circulating cell-free DNA (cfDNA) crucial for deciphering the epigenetic mechanisms through which exercise may exert its health benefits. Our study, therefore, aims to explore how DNA methylation functions as an epigenetic transducer of exercise-induced adaptations in LS carriers by integrating DNA methylome and transcriptome profiles derived from colonic mucosa and cfDNA.

Immune surveillance is another critical mechanism potentially modulated by exercise. In LS, germline MMR mutations lead to microsatellite instability (MSI) upon acquisition of a second somatic hit, generating neoantigens that trigger robust T cell-mediated immune responses.^1,15,16,17 LS carriers are considered to be in a constant “pre-cancer state” throughout their lives, and as such their immune microenvironment may already exhibit features of dysregulation that precede tumor development. Aerobic exercise is known to modulate T cell function and enhance repertoire diversity in both healthy individuals and patients with cancer.^18,19,20 Given that, we hypothesize that long-term aerobic exercise may reshape the TCR repertoire in LS carriers, and potentially contribute to enhanced immunosurveillance and delayed tumorigenesis. To address this, we comprehensively analyzed the TCR repertoire of the colonic mucosa and peripheral blood, characterizing the clonal architecture at the repertoire, nucleotide, and amino acid sequence levels to uncover exercise-associated immune signatures.

By integrating the DNA methylome, transcriptome, and TCR repertoire data from both colonic mucosa and peripheral blood, this study addressed the molecular and immunological effects of a 52-week aerobic exercise intervention in LS carriers. Also, we aimed to identify exercise-induced methylation-regulated genes and characterize shifts in TCR repertoire diversity and clonal compositions in response to exercise. This multi-omics approach provide mechanistic insights into how exercise modulates epigenetic and immune responses in this genetically predisposed population, ultimately contributing to the development of non-pharmacologic strategies for cancer prevention.

Results

Study participants and clinical characteristics

Participants’ clinical characteristics have been previously described⁹ and summarized in Table S1. A total of 16 participants were enrolled in the epigenetics study, with 9 in the exercise group and 7 in the control group. Among them, 8 (4 per group) were also included in the immune cohort study (Table S2). The age of the participants differed between the exercise group (46 years, IQR 40–48) and the control group (35.5 years, IQR 34–38). No participants had history of diabetes or hypertension, and they were not taking non-steroidal anti-inflammatory drugs (NSAIDs) such as aspirin or naproxen. One participant in the usual care group was on treatment with statins and was excluded from this study. The overall study design is illustrated in Figure 1A.

Figure 1.

Schematic overview of the study design and identification of eDMRs

(A) Study design. 16 patients with Lynch syndrome (LS) were included in this study, with 9 in the exercise group and 7 in the control group. Peripheral blood and colonic biopsy tissues were collected at baseline and after the 52-week intervention for genomic-wide epigenomic, transcriptomic, and T cell receptor (TCR) repertoire analyses.

(B and C) Global methylation changes induced by exercise in cfDNA (B) and tissue DNA (C) between the pre- and post-intervention timepoints in the exercise and control groups, respectively. Mean RMS was used to quantify global methylation levels. Each line represents a paired sample from the same individual. Statistical significance was determined by a paired t test (∗p ≤ 0.05 and ∗∗p ≤ 0.01).

(D) Venn diagram illustrates the overlap of DMRs selected by LASSO regression, Random Forest, and Support Vector Machine.

(E) Receiver operating characteristic (ROC) curves illustrate the performance of the three machine learning models—LASSO regression (blue), Random Forest (green), and Support Vector Machine (red)—in distinguishing exercise from control samples. The x axis shows the false positive rate (1-Specificity), and the y axis shows the true positive rate (sensitivity). The area under the curve (AUC) for each model is indicated in the legend, with higher AUC values reflecting better discriminative performance.

(F) Heatmap displays the methylation changes of the 350 selected features across exercise and control groups. Rows represent individual DMRs, and columns represent samples, with hierarchical clustering applied to highlight distinct epigenetic patterns.

Exercise-induced changes in global DNA methylation patterns and eDMRs identification

Global methylation patterns in cfDNA and tissue-derived DNA were compared between the two timepoints across groups (Figures 1B and 1C). In cfDNA, the exercise group showed a non-significant hypomethylation trend, while the control group had significantly increased methylation (p = 0.047). Conversely, tissue-derived DNA showed increased methylation in the exercise group (p = 0.019), with a non-significant hypomethylation trend in the control group.

Differentially methylated regions (DMRs) identified from cfDNA and tissue-derived DNA are summarized in Figure S1A cfDNA yielded more DMRs than tissue DNA, with the control group showing the highest count (32,233). In the exercise group, 75% of cfDNA-derived DMRs were hypomethylated, whereas 77% were hypermethylated in the control group. Conversely, tissue-derived DMRs showed 90% hypermethylation in the exercise group and 80% hypomethylation in the control group. Venn diagrams (Figure S1B) revealed minimal overlap among groups, indicating that methylation changes were largely specific to both the biological material and intervention condition. Based on the predefined statistical and directional criteria (see STAR Methods), we further identified a subset of DMRs as exercise-induced differentially methylated regions (eDMRs), resulting in 9,727 cfDNA-related and 4,241 tissue-related eDMRs.

To further characterize these eDMRs, we examined their genomic distribution relative to gene features (Figure S1C). Most cfDNA-related hypomethylated eDMRs were enriched in promoter regions (42.54%), whereas cfDNA-related hypermethylated and tissue-related eDMRs were mainly located in distal intergenic regions. Chromosomal density plots further highlighted distinct distribution patterns across groups, with cfDNA-related hypomethylated eDMRs showing the highest density (Figure S1D). To evaluate the group-specific methylation changes of the identified cfDNA- and tissue-related eDMRs, we visualized the methylation patterns using heatmaps. Tissue-related eDMRs showed higher methylation levels in the exercise group (Figure S2A), whereas cfDNA-related eDMRs exhibited lower methylation levels, indicating opposing trends between tissue and cfDNA (Figure S2B). Principal component analysis (PCA) of eDMR methylation profiles further demonstrated clear separation between exercise and control groups for both tissue and cfDNA samples (Figures S2C and S2D), thus highlighting context-specific epigenetic responses to exercise in LS carriers.

Given that the initial eDMRs were identified using unadjusted p values, we applied three machine learning models including LASSO regression, Random Forest (RF), and Support Vector Machine (SVM) to further filter high-confidence tissue-derived eDMRs associated with exercise. Each model was implemented with 10 iterations of 5-fold cross-validation (50 iterations per model). Feature ranking was based on both model-specific importance scores and selection frequency. For LASSO, cross-validation and coefficient analyses were used to determine the optimal regularization parameter and CpG site weights (Figures S2E and S2F). Feature importance for RF and SVM was based on the Gini index and weight vectors, respectively, and is described in the STAR Methods. The top 100 DMRs for each criterion were combined, and their union was retained as the candidate feature set. This resulted in 175, 186, and 184 DMRs selected by LASSO, RF, and SVM, respectively. To evaluate model concordance, we constructed a Venn diagram (Figure 1D), which revealed the greatest overlap to be between LASSO and RF (60%), followed by LASSO and SVM (41%), and RF and SVM (33%). In total, 350 unique DMRs were identified across all three models. ROC analysis demonstrated robust classification performance for all three models, each achieving an AUC value > 0.90 (Figure 1E). The final 350 DMRs effectively distinguished groups, as visualized in the heatmap (Figure 1F). For downstream interpretation, the 350 DMRs were mapped to their nearest genes, yielding a final set of 269 unique genes, which were subsequently used for integrative analyses to explore potential exercise-associated transcriptional regulation.

Identification of exercise-induced epigenetically regulated genes in tissue

To explore potential epigenetic regulation of the 269 genes, we built gene-specific linear regression models to assess the relationship between promoter DNA methylation and gene expression, incorporating trial group (exercise vs. control) as an interaction term and adjusting for age and sex. Among the 269 candidate genes, four genes (HMGCS1, ISL1, ITGAV, and MFHAS1) showed significant exercise-associated negative correlations between promoter methylation and gene expression (p < 0.05). The volcano plot summarizes the interaction term analysis based on both statistical significance and effect size (Figure 2A). Regression analyses further confirmed stronger negative associations in the exercise group, consistent with epigenetic repression (Figure 2B). To further prioritize biologically relevant candidates, we integrated these findings with transcriptomic data acquired from the same cohort of patients (GEO: GSE183150).⁹ Differential expression analysis identified 13 statistically significant upregulated and 33 downregulated genes in the exercise group compared to controls. Notably, among the four genes, only ISL1 showed significant downregulation in the exercise group (BH-adjusted p < 0.05, Figure 2C), while the remaining genes did not exhibit statistically significant expression changes. This integrative analysis suggests that ISL1 represents the most robust candidate in response to exercise intervention supported by both epigenetic and transcriptional evidence.

Figure 2.

Identification of exercise-induced methylation-regulated genes

(A) Volcano plot illustrates the interaction effects between promoter methylation and gene expression for 269 candidate genes, analyzed using a linear regression model with group as an interaction term. The x axis represents the interaction coefficient (effect size), and the y axis shows the -log10 (p value). Genes with significant interaction terms (p ≤ 0.05) are highlighted in blue, with four genes (HMGCS1, ISL1, ITGAV, MFHAS1) showing significant negative regulation, indicating stronger methylation-driven repression of gene expression in the exercise group.

(B) Regression plots show the association between promoter methylation and gene expression for the four genes identified as significantly negatively regulated in (A). Each image displays methylation levels (x axis) versus gene expression (y axis) for exercise (blue) and control (red) groups.

(C) Heatmap displays 46 differentially expressed genes from transcriptomic data, including 13 significantly upregulated and 33 significantly downregulated genes in the exercise group compared to control. ISL1 was highlighted as a significantly downregulated gene in response to exercise.

(D) Volcano plot highlights three CRC-associated genes (FLCN, MDM2, RAF1) showing significant exercise-responsive methylation-expression relationships.

(E) Regression plots show the association between promoter methylation and gene expression for three CRC-associated genes. Group-specific patterns highlight exercise-responsive epigenetic regulation.

(F and G) Boxplot shows changes in promoter methylation levels, with significantly elevated methylation of FLCN in the exercise group compared to the control group (F). Boxplot displays the changes in gene expression levels, revealing a decreasing trend of FLCN in the exercise group (G). Statistical significance was determined by the Wilcoxon rank-sum test (∗p ≤ 0.05 and ∗∗p ≤ 0.01).

To explore whether known CRC-associated genes were subject to exercise-induced epigenetic modulation, we performed a focused analysis on the curated gene sets of 178 genes. Among these, three genes (FLCN, MDM2, RAF1) exhibited a significant exercise-associated negative relationship between promoter methylation and gene expression (p = 0.0241, Figures 2D and 2E). To further characterize their regulatory patterns, we compared the changes in promoter methylation and gene expression between groups. Specifically, FLCN showed a significant increase in promoter methylation in the exercise group (p = 0.018), along with a trend toward reduced expression that did not reach statistical significance (Figures 2F and 2G). These findings highlight the complexity of epigenetic regulation and suggest that FLCN may represent an exercise-responsive gene potentially regulated by promoter methylation.

Functional enrichment analysis and pathway-level validation using tumor reduce representative bisulfite sequencing (RRBS) data

To investigate the biological significance of eDMGs, defined as genes associated with eDMRs based on nearest gene annotation (see STAR Methods), we performed GO and KEGG enrichment analyses on three eDMGs sets: cfDNA-derived hypomethylated eDMGs, tissue-derived hypermethylated eDMGs, and tumor-derived hypomethylated DMGs (RRBS data).

In the cfDNA-derived hypomethylated gene set, enrichment analysis was predominantly related to immune response and infection (BH-adjusted p value < 0.05, Figure 3A; Table S6), including Fc gamma R-mediated phagocytosis, leukocyte transendothelial migration, endocytosis, and lysosomes, along with multiple pathogen-related pathways. Additional enrichment was observed in cytoskeletal regulation, autophagy, chemokine signaling, and metabolic processes such as AMPK signaling. These results suggest that exercise-associated hypomethylation events detected in cfDNA reflect immune, metabolic, and cellular stress response activation.

Figure 3.

GO and KEGG enrichment analysis of exercise-induced differentially methylated genes (eDMGs) across cfDNA, tissue, and tumor datasets

(A–C) Dot plot shows significantly enriched GO and KEGG terms for hypo-eDMGs identified in circulating cfDNA (A), for hyper-eDMGs detected in colonic tissue DNA (B), and for hypomethylated genes derived from RRBS data in LS-associated colorectal tumor tissues (C). The dots represent the count of core enrichment genes, and the dot colors represent the BH-adjusted p value.

(D) Pathway-level overlap between hypermethylated genes in exercise-induced colonic tissue (from (B)) and hypomethylated genes in colorectal tumor tissues (from (C)), highlighting shared biological processes or pathways relevant to exercise and CRC.

Tissue-derived hypermethylated eDMGs were significantly enriched in cell signaling and adhesion pathways (BH-adjusted p value < 0.05, Figure 3B; Table S7), including PI3K-Akt and calcium signaling pathways associated with cell proliferation and survival. Given that promoter hypermethylation is generally associated with gene silencing, these findings suggest that exercise may epigenetically suppress oncogenic signaling pathways. Other enriched pathways included axon guidance, neuroactive ligand-receptor interaction, and adherens junction, indicating regulation of neuronal development and tissue polarity.

To further assess the relevance of these findings, we analyzed tumor-derived hypomethylated genes using RRBS data. Enriched pathways were primarily associated with developmental regulation, hormonal signaling, and cellular organization (BH-adjusted p value < 0.05, Figure 3C; Table S8). These pathways are often involved in endocrine disruption, tissue remodeling, and loss of structural stability, which are hallmarks of malignant progression.²¹ Comparison of tissue-derived hypermethylated eDMGs and tumor-derived hypomethylated DMGs identified five overlapping pathways: axon guidance, axonogenesis, adherens junction, arrhythmogenic right ventricular cardiomyopathy, and hypertrophic cardiomyopathy (Figure 3D). Notably, these pathways showed opposite methylation patterns between exercise and tumor samples, suggesting a potential epigenetic counter-regulatory mechanism whereby exercise may suppress tumor-associated activation of adhesion- and migration-promoting pathways, helping to maintain tissue architecture and reduce invasion potential.

Exercise promotes systemic TCR repertoire diversity while decreasing local (tissue) diversity

To assess the temporal dynamics of TCRβ repertoire diversity in both circulating and tissue-derived samples across the two groups, we applied multiple complementary diversity metrics rather than relying on a single index.^22,23 Specifically, rarefaction analysis, the inverse Simpson index,²⁴ and the Gini coefficient were used to capture different facets of TCR repertoire diversity and evenness.

First, we quantified within-individual changes over time (Δ = post − pre) and compared these changes between the exercise and control groups (Figures 4A and 4B). In peripheral blood, the exercise group showed increased diversity over time compared to control, as evidenced by elevated rarefaction curves (Figures S3A and S3B), higher Inverse Simpson index, and lower Gini coefficient (p < 0.05; Figures 4A and 4B; Figures S3E and S3F), thus indicating lower clonal dominance and greater evenness. On the other hand, in the tissue samples, the exercise group exhibited the opposite pattern with lower rarefaction curves (Figures S3C and S3D), a reduced Inverse Simpson index, and an increased Gini coefficient (Figures 4A and 4B). These findings suggest that immune responses to exercise are compartment-specific, characterized by increased peripheral diversity and decreased tissue diversity.

Figure 4.

Exercise-induced changes in TCR repertoire diversity, evenness, similarity, and clonal overlap in both blood and tissue

(A) Bar plot shows the changes in the Inverse Simpson diversity index between the exercise and control group. Each dot represents an individual participant. A statistically significant difference was observed in whole blood-derived repertoires between groups. Values are shown on a log₁₀ scale for visualization purposes only; all statistical analyses were performed using the original (non-transformed) data. Statistical significance was determined by the Wilcoxon rank-sum test (∗p ≤ 0.05 and ∗∗p ≤ 0.01).

(B) Gini coefficient reflects TCR repertoire evenness. In blood, the change in Gini coefficient between the exercise and control groups was statistically significant. In tissue, a trend toward increased Gini coefficient in the exercise group and decreased Gini coefficient in the control group was observed, although the difference was not statistically significant.

(C–F) Longitudinal analysis of TCR repertoire similarity using Morisita’s overlap index. In the exercise group, all four individuals showed high pre-post repertoire similarity in whole blood (C). In contrast, the control group showed lower overlap, with only two individuals retaining moderate similarity (D). In tissue samples, both groups exhibited low repertoire similarity across time points (E and F). Plots show Morisita’s index for paired samples from each individual, with color intensity (redder indicating higher similarity) reflecting the degree of similarity.

(G and H) Tracking the overlap and abundance of top expanded TCR clonotypes between peripheral blood and matched tissue across two time points. Proportion of the top 100 most abundant peripheral blood clones shared in matched tissue, representing potential circulating clones infiltrating the tissue (G). Cumulative abundance of these shared peripheral-derived clones in the tissue, indicating the degree to which dominant blood-derived clones are established locally (log₁₀ scales) (H). Each line represents a paired sample from the same individual.

To further quantify the magnitude of these changes, particularly given the limited sample size, effect size analysis (Cohen’s d) was performed on paired samples, reflecting within-individual comparisons across two time points (Δ = post − pre) with interpretation of effect sizes as described in the STAR Methods. In peripheral blood, the exercise group showed a moderate-to-large effect size in the Gini coefficient (d = −0.81), thus indicating decreased evenness and reduced clonal dominance over time, consistent with the observed increase in diversity. In contrast, tissue samples exhibited substantially larger effect sizes. Notably, the control group demonstrated a large effect size in both diversity and clonality metrics (inverse Simpson index: d = 1.19; Gini coefficient: d = −1.31), indicating pronounced remodeling of the local TCR repertoire over time. Taken together, these findings are consistent with the between-group comparisons of longitudinal changes and highlight distinct patterns of TCR repertoire remodeling across compartments following exercise.

Exercise enhances peripheral TCR repertoire stability and facilitates the migration of tissue-infiltrating T cells

To quantitatively evaluate the overall repertoire similarity between baseline and follow-up timepoints in both tissue and peripheral blood across groups, we employed the Morisita overlap index (MOI), which accounts for both the presence and relative abundance of clonotypes within each repertoire.²³ In blood, the exercise group showed high repertoire similarity across all participants (Figures 4C and 4D), while the control group exhibited greater variability, with two participants showing lower MOI. In tissue samples, both groups exhibited variable repertoire similarity (Figures 4E and 4F), likely due to spatial heterogeneity resulting from the biopsy location.

To assess exercise-associated recruitment of blood-derived T cells into tissue, we analyzed the top 100 most expanded clones in blood that were detectable in the corresponding tissue samples, along with their cumulative frequency changes (Figures 4G and 4H), aiming to capture the changes in clonal overlap and potential exercise-induced infiltration. In the exercise group, three of four participants showed increased shared clones (Figure 4G) and higher cumulative abundance in tissue (p < 0.05; Figure 4H). The control group showed no consistent changes. These findings suggest exercise promotes peripheral TCR repertoire stability and enhances tissue infiltration of dominant peripheral T cell clones in LS. Additionally, V and J gene usage and V-J pairing frequencies were analyzed at baseline and follow-up. In both blood and tissue, V and J gene usage remained stable in both groups (Figures S4A and S4D). V-J pairing showed no significant changes in blood, but subtle differences in tissue were noted, likely due to biopsy location variability (Figure S4E). These findings suggest exercise-associated immune modulation primarily affects TCR repertoire diversity and tissue infiltration, not V or J gene selection or recombination patterns.

Exercise reshapes TCR clonal architecture at both the amino acid and nucleotide levels

To characterize amino acid-level clonal organization, we constructed CDR3 sequence similarity networks based on pairwise Levenshtein distances. Networks were generated for each sample using three sampling strategies, each selecting 2,000 clonotypes from the same repertoire: the most expanded, randomly selected, and least abundant clones (Figures S5A–S5C). Consistent across samples, networks derived from the most expanded clones exhibited the lowest density, while those from the least abundant clones showed the highest density. This pattern indicates that rare clones tend to share higher similarity at the CDR3 amino acid level, while highly expanded clones exhibit greater sequence specificity.

We then assessed how exercise influences CDR3 network density over time by analyzing representative samples from the exercise group (Figures 5A and 5B) and the control group (Figures 5C and 5D) at baseline and 52-week timepoints. In peripheral blood, all four individuals in the exercise group showed increased network density after the intervention, whereas the control group exhibited a declining trend (Figure 5E). Interestingly, an opposite pattern emerged in tissue samples: exercise was associated with reduced network density, whereas the control group exhibited a slight increase (Figure 5F). These findings suggest that exercise enhances CDR3 sequence-level similarity among circulating clonotypes, while supporting a more selective and distinct TCR repertoire within tissue.

Figure 5.

Exercise-induced changes in TCR repertoire network structure and degeneracy in blood and tissue

(A–D) Network analysis of T cell CDR3β amino acid sequences for the top 2000 most abundant clonotypes in the exercise group (A and B) and control group (C and D), with one representative sample per group; tissue networks are not shown. Each node corresponds to a unique T cell CDR3β amino acid sequence, while connections are made by edges when Levenshtein distance (LD = 1). The node size reflects clonal frequency, and color indicates private (unique to individual) or public (shared across individuals) clones. Mean degree was used to assess network density.

(E and F) Line plots show individual changes in network density across four paired samples per group between baseline and 52 weeks. Exercise group shows a significant increase in network density in blood (E). Control group shows a decreasing trend in blood.

(G and H) Bubble plots show the relationship between TCR degeneracy level and clonotype frequency in blood (G) and tissue (H). In both compartments, clonotypes with higher expansion frequencies tend to exhibit lower degeneracy levels.

(I) Stacked bar plot shows the proportion of public versus private clonotypes stratified by degeneracy level. Low-degeneracy clonotypes (≤1) are predominantly private, while high-degeneracy clonotypes (≥5) are mostly public.

(J and K) Line graph shows the changes in abundance of high-degeneracy clonotypes (≥2) in two groups across baseline and 52 weeks in blood (J) and tissue (K). In blood, the exercise group shows a significant reduction in high-degeneracy clonotypes over time, while the control group shows an increasing trend; in tissue, the exercise group displays a consistent increase. Statistical significance was determined by paired t test (∗p ≤ 0.05 and ∗∗p ≤ 0.01).

We further examined TCR degeneracy, which is defined as the number of unique nucleotide sequences encoding the same CDR3 amino acid sequence, to explore its relationship with clonotype abundance and clonal sharing across individuals. Across both compartments, clonotype frequency was negatively correlated with TCR degeneracy level, with highly expanded clones tending to exhibit low degeneracy levels and thus high CDR3 nucleotide sequence specificity, whereas low-abundance clones were often associated with high levels of TCR degeneracy (Figures 5G and 5H). This pattern suggests that highly expanded clonotypes, which are typically antigen-driven and highly specific, tend to exhibit low degeneracy. To further explore the relationship between TCR degeneracy and clonal sharing, we stratified clonotypes by their degeneracy levels. This analysis revealed that low-degeneracy clones (degeneracy ≤1) were predominantly private, whereas high-degeneracy clones (degeneracy ≥5) were commonly shared across individuals (Figure 5I), supporting the notion that TCR degeneracy contributes to the formation of public clonotypes.²⁵ Following exercise, we observed a significant reduction in the proportion of high-degeneracy clones in whole blood (p < 0.05, Figure 5J; Figure S5D) and an increase in colonic tissue (Figure 5K; Figure S5E). In contrast, the control group exhibited more variable and inconsistent patterns over time across both compartments. This compartment-specific response suggests that exercise facilitates the migration and local enrichment of selective, antigen-driven TCR clonotypes from peripheral circulation to tissue.

Discussion

In this study, we provide novel insights into how aerobic exercise could reshape the DNA methylation landscape, modulate the expression of functionally relevant genes, and alter adaptive immune dynamics in LS carriers by integrating DNA methylomes, transcriptomics, and immunogenomics. Epigenetic mechanisms play a role in the complex adaptation to exercise.^26,27 Previous studies in non-hereditary healthy populations have demonstrated that physical activity is associated with beneficial effects across multiple chronic diseases, including metabolic disorders, cardiovascular disease, and cancer, in part through epigenetic remodeling, including both global and locus-specific changes in DNA methylation in a tissue- and gene-specific manner.²⁸ Extending these observations to a genetically high-risk population, our study demonstrates that exercise induces similar context-specific epigenetic changes in LS carriers, with enrichment in immune-related pathways and cancer-associated loci. Notably, we observed a significant increase in global DNA methylation in histologically normal tissues following exercise in this population. This observation is particularly relevant given that global DNA hypomethylation is a well-established hallmark of cancer, contributing to genomic instability and malignant transformation.^14,24 Within this context of high-risk for cancer, our results suggest that exercise may partially counteract tumor-associated epigenetic alterations, thereby enhancing genomic stability and potentially reducing cancer risk in high-risk individuals.

In contrast to the hypermethylation pattern observed in tissue, global DNA methylation levels in circulating cfDNA exhibited a decreasing trend in response to exercise. Most hypo-eDMRs were located in promoter regions, suggesting potential transcriptional activation. Consistent with this, functional enrichment analysis indicated that these hypomethylated genes were significantly involved in immune and inflammatory pathways, particularly those associated with innate immune activation.

These findings extend our previous observation that a 52-week aerobic exercise intervention significantly reduced circulating prostaglandin E2 (PGE2),⁹ a key pro-inflammatory lipid mediator that inhibits the function of neutrophils, macrophages, natural killer cells, and cytotoxic T cells.²⁹ Mechanistically, PGE2 has been shown to induce the expression of DNA methyltransferases such as DNMT3B,³⁰ thereby contributing to immune gene silencing via promoter hypermethylation. Conversely, epigenetic regulation of cyclooxygenase-2 (COX-2), the rate-limiting enzyme in PGE2 synthesis, has been implicated in modulating PGE2 levels via promoter methylation dynamics.³¹ Together, these observations suggest that exercise-induced reductions in PGE2 may not only alleviate immunosuppressive signaling but also promote promoter hypomethylation of immune-regulatory genes, thereby facilitating innate immune activation. Collectively, our findings support a model in which exercise orchestrates systemic immune-metabolic reprogramming through intertwined inflammatory and epigenetic pathways in LS carriers, potentially contributing to cancer-preventive effects.

Pathway enrichment analysis revealed that tissue-derived hypermethylated eDMGs were significantly enriched in the PI3K-Akt and calcium signaling pathways, two canonical oncogenic cascades implicated in colorectal tumorigenesis. The PI3K-Akt pathway plays a central role in promoting cell survival, proliferation, and metabolism, and its aberrant activation has been reported in LS-associated CRC, contributing to uncontrolled growth and apoptotic resistance.^32,33 Likewise, dysregulation of calcium signaling facilitates increased calcium influx, promoting cell migration and invasiveness. In this context, the observed hypermethylation of genes within these pathways in exercise-exposed tissues may reflect a protective epigenetic response that attenuates pro-tumorigenic signaling. This interpretation is further supported by cross-validation with an independent RRBS dataset from LS-CRC tumors and adjacent normal tissues. Notably, hypomethylated genes in tumors overlapped with hypermethylated genes observed in non-tumor tissues from exercise participants, particularly within pathways related to cell adhesion, polarity, and motility. This cross-platform, cross-cohort validation approach provides further evidence that exercise can epigenetically enhance tissue homeostasis in high-risk individuals by suppressing oncogenic signaling involved in tumor invasion and metastasis.

To facilitate integrative analysis with tissue-level transcriptomic data, we applied a two-step feature selection strategy to identify the most informative eDMRs, combining statistical modeling with machine learning-based approaches. This multi-method framework is consistent with current recommendations for feature prioritization in high-dimensional omics data.^34,35 Given that not all DNA methylation alterations have functional consequences,^36,37,38 we further integrated methylation and matched transcriptomic data using a linear regression model to identify functionally associated events. Although four genes (HMGCS1, ISL1, ITGAV, and MFHAS1) showed significant exercise-associated inverse methylation-expression relationships, only ISL1 demonstrated significant downregulation at the transcriptomic level. This concordance across epigenetic and transcriptional analyses suggests that ISL1 may represent a functionally relevant target of exercise-associated regulation. ISL1 encodes a LIM-homeodomain transcription factor involved in transcriptional and metabolic regulation. Overexpressed ISL1 has been demonstrated to be implicated in the initiation, progression, and metastasis of multiple tumor types.^39,40,41 Moreover, highly expressed ISL1 is also associated with advanced tumor stage,⁴² increased tumor size, and poorer overall survival, and promotes angiogenesis.^43,44,45 The observed downregulation of ISL1 in response to exercise, accompanied by an inverse association between promoter methylation and gene expression, suggests a potential tumor-suppressive mechanism by which physical activity may mitigate early oncogenic signals in LS carriers. Although HMGCS1, ITGAV, and MFHAS1 also demonstrated canonical epigenetic regulatory patterns, they did not show consistent significant transcriptional changes. HMGCS1 is a key enzyme involved in cholesterol biosynthesis and metabolic regulation⁴⁶; ITGAV plays an important role in cell adhesion and integrin-mediated signaling, particularly in immune and extracellular matrix interactions; and MFHAS1 has been implicated in immune and inflammatory responses.⁴⁷ These genes may still contribute to exercise-induced biological responses; however, their roles appear to be more context-dependent and warrant further investigation.

The initiation and progression of CRC are driven by multiple abnormal signaling cascades, including Wnt/β-catenin, MAPK, phosphoinositide 3-kinase (PI3K)/AKT, Notch, Hedgehog, Hippo, and JAK/STAT, among others,^48,49 which regulate cellular proliferation, survival, differentiation, and apoptosis.⁵⁰ Beyond genetic mutations, epigenetic mechanisms play a critical role in modulating the activity of these pathways.⁵¹ In our study, we identified three CRC-associated genes (FLCN, MDM2, and RAF1) that exhibited an exercise-responsive methylation pattern. Among these, FLCN (Folliculin) showed one notable increase in promoter methylation in exercise-exposed tissues. FLCN has been implicated in tumor suppression, colorectal tumorigenesis, and energy/nutrient sensing.⁵² In this context, the observed hypermethylation of FLCN in exercise-exposed tissues may reflect a context-specific regulatory mechanism contributing to tissue adaptation and potential tumor-suppressive effects in LS carriers.

Circulating cfDNA captures systemic immune responses and is largely derived from apoptotic hematopoietic cells, especially neutrophils and monocytes, with minimal contributions from T cells.^53,54 While cfDNA-based methylation analysis offers insight into innate immune activation, it does not fully reflect adaptive immune dynamics. Therefore, we extended our analysis to characterize exercise-induced changes in adaptive immunity. Previous studies in healthy populations have demonstrated that exercise modulates T cell dynamics, including enhanced mobilization of effector and redistribution of memory T cells, thereby supporting immune surveillance.⁵⁵ In addition, physical activity has been associated with the maintenance of T cell function and attenuation of immune aging.⁵⁶ However, most of these studies are based on cross-sectional or short-term observations and primarily reflect systemic immune changes. In contrast, our longitudinal analysis integrates both systemic and tissue-specific adaptive immune responses. Our results reveal distinct and coordinated remodeling of the TCR repertoire. In circulation, exercise increased repertoire diversity and clonal evenness, suggesting enhanced flexibility in recognizing novel antigens. In contrast, tissue-derived TCR repertoires exhibited reduced diversity with increased dominance of specific clones, consistent with a more focused, antigen-driven response. These observations are further supported by a Spearman correlation analysis between exercise-associated changes in CD8⁺ T cell density in colonic mucosa measured by immunohistochemistry (IHC) in our previous study of the same cohort⁹ and changes in tissue TCR clonality measured by the Gini coefficient in the current study. We observed a positive association between CD8⁺ T cell infiltration and TCR repertoire remodeling in LS tissues. Although bulk TCR sequencing does not allow direct assignment of individual TCR clonotypes to specific phenotypically defined immune cell subsets, the concordant findings across the two studies collectively support an exercise-associated immune remodeling process. Moreover, increased clonal persistence, tissue sharing of expanded clones, and enhanced clonotype convergence, together with reduced network density in tissue repertoires, consistently indicate that exercise promotes the recruitment and expansion of antigen-specific T cell clones to the tissue. This dual effect of systemic diversification and tissue-specific clonal focusing may enhance immune surveillance against premalignant lesions in the heightened cancer risk population. Together, these findings suggest that exercise-induced activation of cytotoxic immune populations is associated with remodeling of the TCR repertoire, linking immune cell phenotypic changes to adaptive immune dynamics. However, although increased clonal convergence in tissue may reflect antigen-driven expansion and recruitment, prior studies in MSI-H CRC have shown that clonal expansion of T cell clonotypes does not necessarily equate to effective anti-tumor immunity.^57,58 Instead, their functional states, such as cytotoxic versus exhausted phenotypes, critically determine their anti-tumor efficacy. Therefore, the functional properties of these expanded clonotypes will need to be further investigated using single-cell RNA sequencing (scRNA-seq) coupled with paired scTCR sequencing to better resolve their roles in anti-tumor immunity.

Overall, the sample size in this study is relatively small. This reflects the practical challenges of conducting a 52-week longitudinal study design with repeated colonic tissue biopsies in a well-characterized LS cohort. Such procedures are inherently invasive and can limit participant recruitment and retention. In addition, the relatively young age of the cohort may also limit generalizability to the broader CRC population. Particularly, for TCR analysis, due to the limited sample size in the TCR cohort, comprehensive age-adjusted comparisons are statistically challenging. Nevertheless, it is well established that aging is generally associated with reduced TCR diversity and increased clonal dominance.^59,60 Notably, in our study, the exercise group, despite comprising older participants, exhibited increased systemic TCR diversity and greater clonal convergence in tissue, thus suggesting that the observed immune remodeling is more likely attributable to exercise rather than age. Collectively, despite this limitation, the integrative multi-omics design, together with complementary statistical and effect size analyses, provides consistent and biologically meaningful insights into exercise-associated epigenetic regulation and immune repertoire remodeling. Importantly, these mechanistic insights offer a potential link between molecular and clinical outcomes. Recent clinical evidence from the CHALLENGE trial⁶¹ has demonstrated that structured exercise can reduce recurrence and improve outcomes in patients with sporadic CRC. Although our study focuses on LS carriers, which represents a highly immunogenic context driven by MMR deficiency and increased neoantigen load, our mechanistic findings provide a potential multi-layered biological framework that may help explain these clinical benefits. These observations suggest that exercise-induced epigenetic and immune repertoire remodeling may reflect shared biological processes across CRC settings. Further studies integrating both hereditary and sporadic CRC cohorts, as well as broader age ranges, will be required to validate this hypothesis.

Taken together, our study highlights a potential mechanistic basis through which exercise may influence cancer-related outcomes. These findings provide a foundation for future studies integrating multi-omics profiling with clinical endpoints to further define the role of physical activity in cancer prevention and survivorship.

Limitations of the study

Our multi-omics approach underscores a novel interplay between epigenetic and immune remodeling, offering mechanistic insights into exercise-mediated CRC prevention in LS carriers. However, several limitations should be noted. First, the modest sample sizes (16 particiants for epigenetics, 8 participants for immune profiling) limit our statistical power. Second, although RRBS supported MeDIP-seq findings at specific loci, its limited genomic coverage and reduced sample size restricted the comprehensive detection and generalizability of exercise-induced DMRs. Third, while several genes demonstrated statistically significant interaction effects (p < 0.05), none of these findings remained significant after false discovery rate (FDR) correction. Future studies are needed to validate these findings using larger independent cohorts. In particular, single-cell multi-omics approaches including paired scTCR- seq and scRNA-seq will be essential to resolve the functional properties and cytotoxic potential of expanded T cell clonotypes. Experimental models will further help clarifying the roles of candidate genes such as ISL1 and FLCN in exercise-mediated tumor suppression. Finally, extending these investigations to other cancer-predisposition syndromes (e.g., BRCA1/2 and Li-Fraumeni syndrome), as well as to sporadic CRC across broader age ranges, will be important to determine the generalizability of these mechanisms and define the preventive potential of exercise.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Eduardo Vilar (evilar@mdanderson.org).

Materials availability

This study did not generate new, unique reagents.

Data and code availability

• •The previously published mRNA-seq data analyzed in this study are available in the NCBI Gene Expression Omnibus (GEO) under accession number GEO: GSE183150. The MeDIP-seq and RRBS data generated in this study have been deposited in GEO under accession numbers GEO: GSE310884 and GEO: GSE309783, respectively. The TCRβ-seq data generated in this study have been deposited in Zenodo (https://doi.org/10.5281/zenodo.19886210).
• •All custom scripts and code used in this study have been deposited in Zenodo (https://doi.org/10.5281/zenodo.19886210).
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank the patients and their families for their participation. We thank the MDACC Clinical Cancer Prevention Research Core for their assistance in the conduction of this study. We thank the staff of the Epigenomics Profiling Core at MD Anderson Cancer Center for performing DNA methylation assays. This work was supported by gifts from the Feinberg Family Foundation, the Ginn Family Foundation, the Petty Family and the Beirne Family; a partial support for this study by a grant from the University of Texas MD Anderson Cancer Center Duncan Family Institute for Cancer Prevention and Risk Assessment; the generous philanthropic contributions to MD Anderson Cancer Center SPORE in Gastrointestinal Cancer P50 CA221707 (US National Institutes of Health/National Cancer Institute); and P30 CA016672 (US National Institutes of Health/National Cancer Institute) to the University of Texas MD Anderson Cancer Center Core Support Grant; FD was supported by the CPRTP Postdoctoral Fellowship in Cancer Prevention Program RP170259 (Cancer Prevention and Research Institute of Texas).

Author contributions

J.W.: data curation, formal analysis, writing – original draft and revision, and writing - review and editing; N.D.: technical support, data curation, and writing - review and editing; A.M.: technical support and writing - review and editing; F.D.: writing - review and editing; K.M.S.: supervision, investigation, and writing - review and editing; S.G.: supervision and investigation; L.R.-U.: data curation; S.T.: investigation and writing -review and editing; E.V.: conceptualization, supervision, funding acquisition, investigation, methodology, and writing - review and editing.

Declaration of interests

EV had a consulting or advisory role with Janssen Research and Development, Recursion Pharma, Nouscom, Moderna, AbbVie, Permanence Bio and Parabilis. EV has received research support from Janssen Research and Development and Nouscom. EV has equity in Permanence Bio.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Human plasma	CYCLE-P clinical trial	ClinicalTrials.gov: NCT03495674
Human colonic biopsy tissues	CYCLE-P clinical trial	ClinicalTrials.gov: NCT03495674
Human whole blood	CYCLE-P clinical trial	ClinicalTrials.gov: NCT03495674
Critical commercial assays
Quick-DNATM Miniprep Plus Kit	ZYMO research	Cat# D4068
Lymphocyte Separation Media	Corning	Cat# 25-072-CV
Deposited data
mRNA-seq data	GEO9	GEO: GSE183150
MeDIP-seq data	This paper	GEO: GSE310884
RRBS data	This paper	GEO: GSE309783
TCRβ-seq data	This paper	https://doi.org/10.5281/zenodo.19886210
Custom scripts and code	This paper	https://doi.org/10.5281/zenodo.19886210
Software and algorithms
R v4.5.1	R Software for Statistical Computing	https://www.r-project.org/
GraphPad PRISM 10	GraphPad Software, LLC	https://www.graphpad.com/
RStudio v2025.05.1 + 513	Posit Software, PBC	https://posit.co/products/open-source/rstudio/
FastQC v0.11.9	Babraham Bioinformatics	https://www.bioinformatics.babraham.ac.uk/projects/fastqc/
bwa-mem2 v2.2.1	Li et al.62	https://github.com/bwa-mem2/bwa-mem2
BioRender	BioRender	https://www.biorender.com/
Cytoscape v3.10.3	Shannon et al.63	https://cytoscape.org/
edgeR v4.8.2	Chen et al.64	https://bioconductor.org/packages/devel/bioc/html/edgeR.html
Immunarch v1.0.0	Vadim et al.65	https://immunarch.com/
MEDIPS v4.4.0	Lienhard et al.66	https://bioconductor.org/packages/release/bioc/html/MEDIPS.html
clusterProfiler v4.12.6	Yu et al.67	https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html
glmnet v4.1-8	Tay et al.68	https://CRAN.R-project.org/package=glmnet
randomForest v4.7–1.1	Svetnik et al.69	https://cran.r-project.org/web/packages/randomForest/index.html
caret v7.0.1	Kuhn et al.70	https://topepo.github.io/caret/
pROC v 1.19.0.1	Robin et al.71	https://xrobin.github.io/pROC/
ChIPseeker v1.38.0	Wang et al.72	https://bioconductor.org/packages/release/bioc/html/ChIPseeker.html
HOMER v5	Heinz et al.73	http://homer.ucsd.edu/homer/
Bismark v0.24.2	Krueger et al.74	https://www.bioinformatics.babraham.ac.uk/projects/bismark/
stringdist	Van der Loo et al.75	https://github.com/markvanderloo/stringdist
Other
CYCLE-P clinical trial	ClinicalTrials.gov	NCT03495674

Experimental model and study participant details

We included the same cohort of participants previously enrolled in the CYCLE-P trial (ClinicalTrials.gov identifier: NCT03495674),⁹ a prospective, non-randomized, single-site controlled study conducted at the University of Texas MD Anderson Cancer Center (MDACC) from April 2018 to January 2019. Written informed consent was obtained from all participants, and the trial protocol and all amendments were approved by the MDACC Institutional Review Board (IRB #2017-1035). Detailed demographic and clinical characteristics of the study participants are provided in Table S1. Briefly, 21 participants diagnosed with Lynch syndrome were enrolled, including 11 participants in the exercise group who underwent a 52-week aerobic exercise intervention consisting of 45-min cycling classes three times per week and 10 participants in the usual-care control group who received standard-of-care counseling without enrollment in cycling classes. Colonic tissue and peripheral blood samples were collected at both baseline (Visit 1) and at 52 weeks (Visit 3) during routine cancer surveillance visits. Of the original 21 participants, 16 had paired blood samples collected at both baseline and 52-week follow-up timepoints, including 9 from the exercise group and 7 from the control group. Among these, 15 participants (9 exercise, 6 control) also had paired colonic tissue DNA from both time points, thus enabling comprehensive epigenetic analysis across cfDNA and tissue compartments. In addition, TCR repertoire sequencing was performed on a subset of participants (n = 8; 4 exercise and 4 control) with matched paired blood and tissue samples available at both time points. Detailed sample availability for each omics layer and group is summarized in Table S2. Study design, inclusion and exclusion criteria, and exercise intervention protocol have been previously described in Deng et al.⁹ and summarized in Table S3.

Method details

MeDIP-seq and analyses

Genome-wide DNA methylation profiling was performed using MeDIP-seq on both colonic biopsy tissue DNA and cfDNA as previously described.^76,77 Briefly, genomic DNA was sheared to 200–600 bp fragments, denatured, and immunoprecipitated using a monoclonal antibody against 5-methylcytosine, followed by high-throughput sequencing on the Illumina HiSeq platform.⁷⁸

Sequencing reads were aligned to the human reference genome (GRCh38/hg38) using bwa-mem2. Genome-wide methylation signals were quantified in 300 bp sliding windows using the MEDIPS R package⁶⁶ and normalized by the trimmed mean of M-values (TMM) methods. Promoter regions were defined as −2000 bp to +500 bp relative to transcription start sites (TSS), and promoter methylation levels were calculated as the average Relative Methylation Score (RMS) across overlapping windows.

DMRs in tissue and cfDNA were identified separately within the exercise and control groups by comparing pre- and post-intervention samples using the MEDIPS edgeR framework.⁷⁹ As no windows met the adjusted p < 0.05 threshold in tissue-derived DNA, DMRs were defined using relaxed criteria (minimum coverage ≥5 reads, |log₂FC| ≥2, and unadjusted p < 0.01). Exercise-induced DMRs (eDMRs) were defined as DMRs either unique to the exercise group or shared between groups but exhibiting opposite methylation changes between pre- and post-intervention, capturing both time-dependent and exercise-specific effects.

To further prioritize discriminative eDMRs, three machine learning approaches were applied for feature selection, including least absolute shrinkage and selection operator (LASSO) logistic regression, Random Forest (RF), and Support Vector Machines (SVM). Models were evaluated using 10 repeated 5-fold cross-validation. LASSO was performed using glmnet R package⁸⁰ (α = 0.5) with lambda optimized by cross-validation, and features with non-zero coefficients were retained. RF was implemented using the randomForest R package to evaluate the importance of each DMR based on mean decrease in Gini impurity. SVM classification was performed using the radial basis function (RBF) kernel (svmRadial) implemented via the caret R package.⁷⁰ Model performance was evaluated using receiver operating characteristic (ROC) curves and the corresponding area under the curve (AUC), computed via the pROC R package.⁷¹

Feature selection was further refined by integrating model-specific importance scores and selection frequency across cross-validation iterations. Model-specific importance scores were defined as follows: absolute coefficient values for LASSO, mean decrease in Gini for Random Forest, and feature weights derived from the SVM model. Selection frequency was defined as the proportion of times each feature was selected among the top-ranked features across repeated cross-validation runs. For each model, the top 100 features ranked by importance scores and the top 100 features ranked by selection frequency were intersected to define model-specific feature set. The final set of high-confidence eDMRs was obtained by combining the model-specific feature sets from all three models using a union approach, with duplicate features removed. DMRs were then annotated to their nearest genes using the ChIPseeker R package⁸¹ based on the shortest genomic distance to the transcription start site (TSS). Unless otherwise specified, no maximum distance cutoff was applied. The resulting unique genes were defined as exercise-induced differentially methylated genes (eDMGs).

Identification of epigenetically regulated genes

To identify methylation-driven genes that are most responsive to exercise intervention, we constructed a linear regression model using lm function in R to evaluate the association between changes in promoter methylation and corresponding gene expression. The model was specified as: ΔExpression ∼ ΔMethylation ∗ Group + Age + Sex. Genes showing a significant negative ΔMethylation ∗ Group interaction (β coefficient < 0 and a p-value < 0.05) were considered exercise-responsive methylation-driven genes.

Using the same approach, we further curated the gene panel for focused analysis: A set of CRC-associated oncogenes and tumor suppressor genes (TSGs), and genes involved in key signaling pathways associated with CRC development and progression, based on the COSMIC Cancer Gene Census and recent CRC molecular profiling studies,^48,82,83,84 as summarized in Table S4. These curated genes were screened to identify candidates under potential exercise-induced epigenetic regulation.

RRBS and analyses

Genomic DNA was extracted from paired normal and tumor colonic tissues of 3 patients with LS-associated CRC. Clinical, demographic and molecular characteristics for the 3 patients is provided in Table S5. Reduced representation bisulfite sequencing (RRBS) was performed to profile DNA methylation at single-base resolution, with enrichment for CpG-rich regions, including promoters and CpG islands. Briefly, 100–500 ng of high-quality genomic DNA from each sample was digested with the MspI restriction enzyme, followed by end repair, A-tailing, and ligation to methylated sequencing adapters to prevent amplification bias. Adapter-ligated fragments were size-selected (40–220 bp) and subjected to bisulfite conversion, in which unmethylated cytosines were converted to uracils while methylated cytosines remained unchanged. The converted DNA was PCR amplified, purified, and sequenced on the Illumina HiSeq platform. Sequencing reads were aligned to the hg19 reference genome using Bismark⁷⁴ with default settings. Briefly, duplicated reads were removed with the deduplicate_bismark tool, and methylation levels were determined using bismark_methylation_extractor. Methylated regions were then annotated with annotatePeaks.pl from the HOMER package.⁷³

Functional and pathway enrichment analyses

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed to characterize the biological functions and pathways associated with differentially methylated genes. GO enrichment analysis was performed using enrichGO function in the clusterProfiler R package⁶⁷ with the human annotation database org.Hs.e.g.,.db, using gene symbols as input and including all three GO ontologies (biological process, cellular component, and molecular function). KEGG pathway enrichment analysis was performed using enrichKEGG with Entrez gene identifiers and the human organism annotation (hsa). A p-value cutoff of 0.05 and q-value cutoff of 0.05 were applied. Enrichment analyses were conducted separately for cfDNA-derived hypomethylated eDMGs, tissue-derived hypermethylated eDMGs, and tumor-derived hypomethylated DMGs identified from RRBS data. For visualization, enriched terms were ranked by Benjamini–Hochberg-adjusted p-values, and the top 10 terms from each GO ontology and the top 10 KEGG pathways were displayed. Enriched pathways identified in tissue-derived hypermethylated eDMGs and tumor-derived hypomethylated DMGs were further compared to identify shared pathways exhibiting opposite methylation patterns.

TCR sequencing and diversity analyses

Genomic DNA was extracted from both PBMCs, whole blood, and biopsy tissues using Quick-DNATM Miniprep Plus Kit (ZYMO research, Irvine, CA) following the manufacturer’s instructions.⁸⁵ Library preparation was conducted using the ImmunoSEQ Assay (Adaptive Biotechnologies, Seattle, WA), a multiplex PCR-based method targeting the complementarity-determining region 3 (CDR3) and sequenced at ultra-deep resolution. Sequencing quality control, sequence alignment, V(D)J gene annotation, and clonotype assembly were performed by Adaptive proprietary pipeline. Only productive TCRβ rearrangements (in-frame sequences without stop codons) were retained for downstream analyses.

TCR diversity was assessed between baseline and follow-up time points using rarefaction analysis, Inverse Simpson index, and Gini coefficient. Rarefaction curves were generated using the repDiversity function in the immunarch R package⁶⁵ with interpolation and extrapolation enabled. The inverse Simpson index was used to quantify effective diversity, while the Gini coefficient was used to assess clonality, with lower values indicating more even clone distributions. Group differences in longitudinal changes (post-pre) were evaluated as described in the quantification and statistical analysis section.

Assessment of TCR repertoire similarity

The Morisita overlap index⁸⁶ (MOI) was used to assess TCR repertoire similarities between pre- and post-intervention across groups, as it accounts for both the presence of shared clonotypes and their clonal abundance. MOI values range from 0 to 1, with higher values indicating greater similarity between repertoires, and a lower index reflecting more divergent TCR compositions.

Sequence-level analysis of TCR clonal architecture

To characterize the amino acid sequence-level architecture of the TCR repertoire, pairwise Levenshtein distances⁸⁷ between all CDR3 amino acid sequences were calculated using the stringdist R package⁷⁵ to generate a clonotype-by-clonotype distance matrix.²² This matrix was then converted into a similarity network, where each node represents a unique clonotype and edges connect clonotypes with a Levenshtein distance below a predefined threshold, reflecting similarity at amino acid level. Network density, defined as the mean degree of all nodes, was used to evaluate overall repertoire connectivity, where the degree of a node represents the number of clonotypes with similar CDR3 sequences. A higher network density indicates greater sequence similarity across the repertoire, whereas lower density reflects increased sequence specificity among clonotypes.

TCR degeneracy was defined as the number of unique nucleotide sequences encoding the same CDR3 amino acid sequence.^88,89 For each sample, clonotypes were annotated with their degeneracy level. Clonotype frequency was calculated as the proportional abundance of each unique CDR3 sequence within the repertoire. A public clonotype was defined as a productive TCRβ shared by at least 2 individuals from our cohort by matching the CDR3β, Vβ and Jβ segments at the amino acid level. Clonotypes were then stratified by degeneracy level (e.g., 1, 2–4, ≥5) to assess their distribution across groups and time points. Convergent TCR clones were defined as clonotypes with a degeneracy level >1. For each sample, the proportion of convergent TCR clones was calculated as the cumulative frequency of convergent clonotypes divided by the total repertoire frequency.

Quantification and statistical analysis

Statistical analyses were performed in R unless otherwise specified. Exact sample sizes (n) for each analysis are provided in the corresponding Results, figure legends, and Table S2. Differential methylation analyses were performed using the edgeR framework implemented in MEDIPS, with multiple-testing correction using the Benjamini–Hochberg method. Associations between changes in promoter methylation and gene expression were evaluated using linear regression models as described above. For TCR repertoire analyses, given the small sample size and non-normal data distribution, group differences in longitudinal changes (post-pre) between the exercise and control groups were evaluated using the non-parametric Wilcoxon rank-sum test. Effect sizes were calculated using Cohen’s d based on within-subject changes (post-pre). Standard thresholds were used for interpretation (|d| ≈ 0.2: small, 0.5: medium, 0.8: large, and 1.2: very large), and directionality was interpreted in the context of each metric, with positive values indicating increased diversity (Inverse Simpson index) or increased clonal dominance (Gini coefficient) at follow-up. Statistical significance was defined as p < 0.05 unless otherwise specified.

Additional resources

Biospecimens analyzed in this study were collected as part of the previously published CYCLE-P clinical trial,⁹ which was registered at ClinicalTrials.gov (NCT03495674).