运动改善脂肪肌肉比的全基因组关联分析和效果预判模型构建

杨晓琳; 包大鹏; 李燕春; 晏冰; 梅涛; 何子红

doi:10.16469/J.css.2025KX057

运动改善脂肪肌肉比的全基因组关联分析和效果预判模型构建

Genome-Wide Association Study of Exercise-Induced Fat-to-Muscle Mass Ratio Change and the Construction of Predictive Model

摘要

摘要: 目的：分析高强度间歇训练（high-intensity interval training，HIIT）和抗阻训练（resistance training，RT）干预降低脂肪肌肉比（fat-to-muscle mass ratio，FMR）的效果及个体差异，全基因组筛选效果相关分子标记并分析遗传因素的影响，联合表型和遗传因素构建运动改善FMR效果预判模型。方法：组织身体活动不足成年人进行12周HIIT（n=260）或RT（n=177）运动干预，分析12周干预前后FMR改善效果差异。采用全基因组关联分析（genome-wide association studies，GWAS）筛选12周HIIT或RT干预改善FMR效果差异分子标记（P＜1.0×10⁻⁵）。基于GWAS结果计算多基因评分（polygenic scores，PGS），分析PGS与运动降低FMR效果关系，通过随机森林、支持向量机、分布式梯度增强和逻辑回归4种方法构建运动降低FMR效果预判模型。结果：1）12周HIIT或RT干预均引起FMR显著下降（P＜0.05），HIIT降低FMR效果优于RT（P＜0.05），但2种干预降低FMR效果均存在个体差异（无效率为34.0%和40.1%）；2）14个单核苷酸多态性（single nucleotide polymorphisms，SNPs）与HIIT改善FMR效果显著相关（P＜1×10⁻⁵），12个SNPs与RT干预改善FMR效果显著相关（P＜1×10⁻⁵），其中rs2277512显著性达P＜5×10⁻⁸，解释度为14.1%；PGS对HIIT和RT改善FMR效果的解释度分别为53.9%和52.6%；3）基于性别、年龄、PGS、初始BMI、初始肌肉质量、初始脂肪质量和运动方案构建的随机森林模型能够有效预判运动改善FMR效果（AUC=0.935，cut-off=0.339），模型特征中PGS重要性最高。结论：HIIT和RT运动干预后FMR改善效果均存在个体差异，且2种运动方式的PGS解释度均不低于52%；联合表型和遗传因素构建的随机森林预测模型可在运动前预测运动改善FMR效果，为精准健身提供方案选择依据。

Abstract: Objective: To evaluate the effects of high-intensity interval training (HIIT) and resistance training (RT) on reducing the fat-to-muscle mass ratio (FMR), explore individual variability, identify genetic markers by genome-wide association study (GWAS), and construct predictive models combining phenotypic and genetic factors for precision exercise. Methods: Adults with insufficient physical activity were subjected to 12 weeks of HIIT (n=260) or RT (n=177) exercise intervention, and the difference in FMR improvement before and after 12 weeks of intervention was analyzed. GWAS identified molecular markers associated with FMR improvement (P＜1.0×10⁻⁵). The polygenic scores (PGS) were calculated based on the GWAS results, and the relationship between PGS and the effect of exercise-induced FMR reduction was analyzed. The prediction model of exercise-induced FMR reduction was constructed by four methods: random forest, support vector machine, eXtreme gradient boosting and logistic regression. Results: 1) Both HIIT and RT significantly reduced FMR (P＜0.05), with HIIT showing superior efficacy (P＜0.05). Individual variability was noted, with 34.0% (HIIT) and 40.1% (RT) showing no response. 2) Fourteen and 12 single nucleotide polymorphisms (SNPs) were significantly associated with FMR improvement after HIIT and RT, respectively (P＜1×10⁻⁵). Among these, rs2277512 reached genome-wide significance (P＜5×10⁻⁸) and explained 14.1% of the variance. PGS explained 53.9% and 52.6% of the variability in FMR decrease for HIIT and RT, respectively. 3) The random forest model based on gender, age, PGS, initial BMI, initial muscle mass, initial fat mass and exercise program can effectively predict the effect of exercise on FMR improvement (AUC=0.935, cut-off=0.339). Among the model features, PGS is the most important. Conclusions: There were individual differences in the improvement effect of FMR after HIIT and RT exercise intervention, and the PGS explained over 52% of the variability in FMR under two exercise modes. The random forest prediction model constructed by combining phenotypic and genetic factors can predict the effect of exercise on improving FMR before exercise, and provide a basis for scheme selection for precise fitness.

HTML全文

参考文献(83)

施引文献

资源附件(0)