Healthy lifestyle practice correlates with decreased obesity prevalence in individuals with high polygenic risk: TMM CommCohort study

The study addresses whether adopting healthy lifestyle behaviors can mitigate obesity risk among individuals with high genetic susceptibility quantified by polygenic scores (PGS). Obesity, a key component of metabolic syndrome, substantially elevates risk for lifestyle-related diseases and has risen globally. Heritability of BMI is estimated at ~0.28–0.30, and PGS provides a quantitative measure of genetic risk. While prior work has examined interactions of lifestyle and genetic risk for specific diseases (e.g., coronary disease), comprehensive evidence for obesity is limited. This study aims to build and validate an obesity PGS in a large Japanese community cohort and assess how lifestyle factors, particularly physical activity and sodium intake, relate to obesity risk across PGS strata, using a dichotomized outcome (BMI ≥ 25 kg/m²) to facilitate clinical interpretability.

Prior literature establishes the health burden and rising prevalence of obesity. Genetic contributions to BMI are nontrivial (heritability ~0.28–0.30), and PGS methods have enabled quantification of genetic risk. Previous studies have linked favorable lifestyle behaviors to reduced risk even among those with high polygenic risk for specific diseases such as coronary artery disease, stroke, dementia, breast and lung cancers, and type 2 diabetes. However, fewer studies have evaluated lifestyle modification effects stratified by PGS for intermediate, broadly relevant risk factors like obesity or blood pressure, particularly in East Asian populations. This gap motivates assessing the practical utility of obesity PGS in the context of lifestyle behaviors.

Design and cohorts: Cross-sectional analysis using four datasets from the Tohoku Medical Megabank Community-Based Cohort (TMM CommCohort): TMM10K (n ≈ 9,811; model selection), and three validation datasets TMM67K (n = 50,195), TMM18K (n = 11,947), and TMM8K (n = 6,796). Participants in TMM10K and TMM67K were recruited at specific government health checkups; TMM18K and TMM8K participants visited TMM assessment centers. Overlaps were retained in TMM10K and excluded from other datasets. Obesity definition: BMI ≥ 25 kg/m² (Japanese criterion). Lifestyle exposures: - Sodium intake estimated from spot urine via Tanaka’s formula. - Physical activity from self-administered questionnaires; leisure-time exercise (LTE: slow/fast walking, light-to-moderate, strenuous exercise) and daily life activity (DLA: occupational sitting, standing, walking, strenuous work) quantified in MET-hours/day and categorized into quintiles (Q1–Q5). Ethics: Informed consent obtained; IRB approval (Iwate Medical University HG H25-2). Genotyping and imputation: Genotyping per prior protocols; pre-phasing with Eagle v2.4.1; imputation with Minimac3 v2.0.1 using 1000 Genomes Phase 3 East Asian reference. Exclusions: individuals with call rate <0.95, non-Japanese ancestry, or sex mismatch; variants with call rate <0.95, HWE p < 1e-6, MAF < 0.01, or imputation R2 < 0.3. PGS construction: SNP weights from BioBank Japan BMI GWAS summary statistics (n = 158,284). Variants with imputation R2 < 0.3 in GWAS summary excluded. Two approaches evaluated in TMM10K: - Pruning and thresholding (P+T): 24 models across 6 p-value thresholds (1, 0.5, 0.05, 5e-4, 5e-6, 5e-8) and 4 LD r2 thresholds (0.2, 0.4, 0.6, 0.8). - LDpred v1.0.11: 7 models with fraction of causal variants p ∈ {1, 0.3, 0.1, 0.03, 0.01, 0.003, 0.001}. Individual PGSs computed using PLINK2 (v2.00a2LM). PGSs normalized within each validation dataset (mean 0, SD 1). Model selection and validation: In TMM10K, models were evaluated by AUC for predicting obesity adjusting for age and sex; best model (LDpred p = 0.03) selected and applied to validation datasets. Statistical analysis: - AUCs computed using ROCR; 95% CIs via pROC. - Meta-analysis across validation datasets using fixed-effect model; random-effects used when heterogeneity significant (p < 0.05). - Trend tests for LTE, DLA, and sodium intake quintiles using logistic regression adjusted for age and sex, stratified by PGS categories. - PGS categories: low (1st–10th percentile), intermediate (11th–90th), high (91st–100th). - Spearman correlations between PGS and continuous obesity-related measures (BMI, body fat percentage by bioelectrical impedance [BFP], visceral fat area by CT [VFA], waist circumference) after rank-based inverse-normal transformation. - Power analysis with metapower under varying heterogeneity (I2) and effect sizes. Sensitivity analyses: Sex-stratified analyses; normalization of PGS after pooling all validation datasets.

- PGS model performance: In TMM10K, LDpred model with p = 0.03 achieved best performance for obesity (BMI ≥ 25), AUC ≈ 0.63; pROC AUC 0.62 [95% CI: 0.61–0.63]. Correlation with continuous BMI (adjusted) Spearman’s ρ ≈ 0.246. Comparable to published East Asian models (Privé PGS002161 AUC 0.61 [0.60–0.62]; Tanigawa PGS001228 AUC 0.60 [0.59–0.62]). - Validation and risk stratification: Normalized PGS distributions were similar across validation datasets. Obesity prevalence rose with PGS percentile; increase from baseline (1st percentile) to 50th percentile by ~17.6–19.1%, and to 100th percentile by ~39.5–48.5%. Meta-analysis ORs (reference = low PGS, 1st–10th percentile): • Intermediate PGS (11th–90th): OR 2.27 [95% CI: 2.12–2.44] • High PGS (91st–100th): OR 4.83 [4.45–5.25]. - Alternative obesity metrics: PGS showed consistent predictive ability across obesity definitions by waist circumference, BFP, and VFA with AUCs 0.58–0.63; significant correlations with quantitative metrics (ρ: 0.131–0.201 for BFP/VFA/waist; ρ: 0.242–0.254 for BMI). - Physical activity: Across all PGS strata, higher LTE quintiles associated with lower obesity risk; meta-analysis trend OR per quintile increase ≈ 0.90 [0.87–0.94], including high PGS group. DLA showed no significant association overall in any PGS category; sex-stratified sensitivity suggested a modest protective association of higher DLA among males with medium to high PGS. - Sodium intake: Higher sodium intake quintiles associated with increased obesity risk across all PGS strata; meta-analysis trend ORs per quintile: • Low PGS: 1.29 [1.21–1.38] • Intermediate PGS: 1.30 [1.25–1.35] • High PGS: 1.24 [1.17–1.31]. - Overall, healthy lifestyle practices (higher LTE, lower sodium intake) correlate with lower obesity prevalence even among individuals with high genetic risk, though high PGS still confers substantial residual risk.

The study directly addresses whether healthy lifestyle behaviors can attenuate obesity risk among individuals with elevated genetic susceptibility. The constructed and validated obesity PGS effectively stratified risk, with high PGS conferring ~4.8-fold higher odds of obesity relative to low PGS. Importantly, increased leisure-time exercise was associated with reduced obesity risk across all genetic risk categories, including the highest PGS decile, indicating modifiable behaviors can mitigate, though not eliminate, genetic risk. Conversely, higher sodium intake correlated with increased obesity risk in all PGS strata, suggesting dietary sodium as a potentially actionable target irrespective of genetic risk. Despite these associations, individuals with high PGS maintained higher obesity risk even at the highest LTE levels, underscoring the strong genetic influence and the need for tailored, possibly intensified prevention strategies in this subgroup. The PGS also demonstrated utility across multiple obesity definitions and modest correlations with quantitative adiposity measures, supporting its robustness. These findings support integrating PGS with lifestyle counseling to personalize obesity prevention and management.

A polygenic score for obesity derived from Japanese GWAS data was developed and validated, stratifying individuals into markedly different obesity risk categories. Across genetic risk strata, greater leisure-time exercise was associated with reduced obesity risk, while higher sodium intake was associated with increased risk. These results suggest that healthy lifestyle practices can beneficially influence obesity risk even among those with high polygenic susceptibility, informing personalized preventive strategies. Future work should include longitudinal validation, evaluation in diverse populations, exploration of mechanisms (e.g., sodium-adiposity link), and enhancement of PGS performance through larger ancestry-matched GWAS.

- Population and generalizability: Community-based cohort from specific regions in Japan; unmeasured geographic or cohort-specific biases may limit generalizability; external replication required. - Study design: Cross-sectional; cannot infer temporality or predictive performance over time; longitudinal studies needed. - PGS normalization: PGS was normalized within each dataset, which could bias risk estimation if distributions differ substantially between datasets; sensitivity analyses suggest limited impact, but residual bias is possible. - PGS performance constraints: Accuracy limited by GWAS training sample size and ancestry; larger GWAS could improve predictive performance (discussed by authors).