Evaluating a Genome-Wide Polygenic Score for Handgrip Strength and Its Interplay with Leisure-Time Physical Activity Across the IGEMS Twin Cohorts

Abstract

Purpose:

Polygenic scores (PGSs) may help assess genetic predisposition to multifactorial traits. We examined whether age, sex, and leisure-time physical activity (LTPA) modify the association between a PGS for handgrip strength (HGS) and measured HGS in older adults.

Methods:

PGS for HGS (PGS_hgs), based on Pan-UK Biobank genome-wide association study data, was calculated for 5103 participants (aged 40–96; 44% women) from eight twin cohorts in Denmark, Sweden, Australia, the United States, and Finland within the IGEMS consortium. Sex-standardized HGS and self-reported LTPA were assessed cross-sectionally. Linear mixed models estimated associations between PGS_hgs and HGS, including interactions with age, country, and LTPA, as well as an association between PGS_hgs and LTPA. Fixed-effect within-pair models were conducted to assess environmental contributions.

Results:

Higher PGS_hgs was associated with greater HGS (β = 2.14, SE = 0.15, P < 0.001), explaining 4.6% of HGS variance overall, with modest variation across countries. In sex-stratified models, PGS_hgs explained 5.2% of the variance in females and 4.3% in males. No statistically significant interaction with age was found. A significant PGS_hgs × LTPA interaction (β = –0.034, P = 0.013) indicated that the association between LTPA and HGS was more pronounced among individuals with lower PGS_hgs. The within-pair models offered limited support for the independent environmental impact of LTPA.

Conclusions:

The PGS_hgs was associated with measured HGS in the meta-analysis, highlighting the potential of PGSs to capture individual differences in strength-related traits across populations. The association of PGS_hgs with HGS was moderated by LTPA, such that the beneficial impact of LTPA on HGS was greater among individuals with a lower genetic propensity for HGS.

The aging process is characterized by an involuntary reduction in muscle mass and strength, resulting from a combination of genetic, environmental, and lifestyle factors, along with their intricate interplay (1). Muscle strength begins to gradually decline from around the age of 30, but significant changes in the aging process occur after the age of 50, with an annual diminishing of muscle strength by 1%–2% (2). Low muscle strength is associated with an increased risk of chronic diseases, disability, and mortality (1,3). Among modifiable factors, physical activity (PA) plays a key role in preserving muscle strength and mitigating age-related decline. Regular PA can counteract losses in muscle mass and function, supporting mobility and reducing the risk of disability (3). At the same time, reduced muscle strength may lead to lower PA engagement, creating a feedback loop that accelerates physical decline (4). Conversely, greater muscle strength can enhance one’s ability and motivation to remain active, reinforcing long-term physical function and well-being (5).

Handgrip strength (HGS) is a widely used, reliable marker of overall muscle strength and an important indicator of functional status and health in aging populations (6). HGS, even when measured in midlife, predicts future adverse outcomes, including disability and mortality (7–9). HGS can, therefore, be considered to signal the individual’s intrinsic physiological capacity to resist functional decline into critical disease and disability levels (10). HGS is a complex, multifactorial trait affected also by internal factors such as age and sex. Males typically exhibit higher HGS throughout the lifespan, and subtle sex differences have been observed in the rate of age-related decline (11,12). Classical twin designs among monozygotic (MZ) and dizygotic (DZ) twins, who share 100% and ~50% of their segregating genes, respectively, have shown that HGS is moderately to highly heritable, with genetic factors accounting for 30%–65% of its variance (13). This heritability appears relatively stable with age (11,12,14), though some findings suggest environmental influences may increase over time (13,15). Evidence regarding sex differences in the genetic architecture of HGS is mixed. While one study reports comparable genetic and environmental contributions across sexes (12), others have found higher heritability in males (11,16) and greater shared environmental influence in females (11). Additionally, the types of environmental factors that contribute to HGS also appear to differ between sexes across adulthood (17).

From a genetic perspective, HGS is a polygenic trait, that is, it is influenced by many genes across different loci, each with a low effect (18,19). A polygenic score (PGS) is a genetic tool used to estimate an individual’s predisposition to a trait or condition, based on the aggregated effect sizes of multiple genetic variants, typically single-nucleotide polymorphisms (SNPs), across the genome (20). However, as with many complex traits, PGSs capture only a fraction of the heritability estimated from family-based designs, leaving a substantial proportion of so-called “missing heritability” that may reflect unmeasured genetic effects, gene–environment interplay, or methodological differences between approaches (21,22). In addition, while the observed sex differences in the genetic architecture of human traits are generally considered to have limited consequences (23,24), studies have shown that the effects of many disease-specific PGS can vary by age and sex (25).

Genetic influences on muscle strength do not operate in isolation but may interact with environmental factors. First, gene–environment interaction refers to situations in which the effect of genetic predisposition on an outcome depends on environmental exposure, or vice versa (26,27). In the context of muscle strength, PA may modify the extent to which genetic predisposition translates into measurable strength differences. Second, gene–environment correlation refers to the possibility that genetic predisposition influences exposure to certain environments (27). For example, individuals genetically predisposed to greater strength might be more likely to engage in PA, which could in turn contribute to higher measured HGS. We have recently constructed PGS for HGS (PGS_hgs) and found it to be a reliable predictor of overall muscle strength among older Finnish females, explaining 6.1% of the variation in the measured HGS and 5.4% of the variation in the knee extension strength (28). Building on this validation, the present study aimed to extend the evaluation of the PGS_hgs to international twin cohorts and investigate whether the association between PGS_hgs and measured strength is modified by age, sex, country of residence, or leisure-time PA (LTPA), either amplifying or attenuating underlying genetic influences. In addition, to examine the association of LTPA with muscle strength, we used within-twin pair analyses (29), which relate differences in LTPA between co-twins to differences in HGS while controlling for genetic and shared family environmental factors. Finally, to evaluate potential pathways suggestive of gene–environment correlation, we examined whether LTPA mediates the association between genetic predisposition to HGS and measured strength, specifically focusing on DZ twins where co‑twins can differ in PGS while still sharing family environment.

METHODS

Study Design and Participants

The PGS for HGS was derived using publicly available summary statistics from the Pan-UK Biobank (Pan-UKBB) genome-wide association study (GWAS) of maximum HGS, which included 418,827 individuals aged 40–69 yr (https://pan.ukbb.broadinstitute.org/). We calculated the PGS_hgs in participants of the Interplay of Genes and Environment across Multiple Studies (IGEMS) consortium (30,31). As an initial step in PGS_hgs validation, we quantified the extent of missing heritability—defined as the proportion of genetic variance not captured by the PGS—by comparing SNP-based and pedigree-based heritability estimates. We then further validated the PGS_hgs by examining its association with measured HGS across the IGEMS cohorts. Subsequently, potential effect modification of the PGS_hgs–HGS association by age, sex, country of residence, and LTPA was evaluated in the full cohort. To assess the role of LTPA, the effect of LTPA on HGS was examined both at the individual level and within-twin pairs. Finally, mediation analyses were conducted in the full sample and in DZ twins. The overall study flow is presented in Figure 1. IGEMS is an international consortium of 21 twin studies focusing on longitudinal twin investigation of adulthood and aging (31). Our analytic sample comprised 5103 twin individuals drawn from eight independent cohorts, spanning five countries: Denmark, Sweden, Australia, the United States, and Finland. All of them had available PGS, ancestry-informative genetic principal components (PCs) (32), and at least one valid HGS measurement (Supplemental Fig. 1, Supplemental Digital Content, https://links.lww.com/MSS/D409). Of those, 4536 participants (2623 DZ twins) had data on LTPA, and 4451 participants had data on both LTPA and body mass index (BMI). A subsample of 1982 complete twin pairs (43% MZ and 57% DZ pairs) was used in within-pair analyses (Supplemental Fig. 2, Supplemental Digital Content, https://links.lww.com/MSS/D409). The specific methods to determine zygosity varied among studies, but in most cases, standard questionnaires were administered, and DNA analysis was conducted to clarify uncertain instances (33). The North West Multi-centre Research Ethics Committee approved the UK Biobank study (21/NW/0157). The IGEMS Consortium was approved by the University of Southern California Institutional Review Board (approval number UP-16-00315). The FITSA data collection was approved by the Ethics Committee of the Central Hospital District of Central Finland (KSSHP 24/2000). All studies obtained informed written consent from all participants and adhered to the principles of the Declaration of Helsinki. Additional cohort-specific details are provided in the Supplemental Digital Content, https://links.lww.com/MSS/D409 under Overview of IGEMS Cohorts Included in the Analysis.

FIGURE 1.

Study design and workflow. The polygenic score for handgrip strength (PGS_hgs) was derived from the Pan-UK Biobank genome-wide association study summary statistics. All subsequent validation, moderation, LTPA association, and mediation analyses were conducted within the IGEMS cohorts (N = 5103), with sample sizes varying by analysis due to the use of covariate data and twin structure (MZ/DZ twins). PGS, polygenic score; HGS, handgrip strength; IGEMS, The Interplay of Genes and Environment across Multiple Studies; LTPA, leisure-time physical activity; MZ, monozygotic; DZ, dizygotic.

Maximum HGS

In the UK Biobank (UKBB), maximal isometric HGS was measured using a calibrated hydraulic hand dynamometer (Jamar J00105; Lafayette Instrument Company, Lafayette, IN) (34). In all IGEMS studies, HGS was measured during in-person assessments conducted by researchers and trained research assistants. Details of HGS measurement protocols for the included cohorts are available in the Supplemental Digital Content, https://links.lww.com/MSS/D409 under Overview of IGEMS Cohorts Included in the Analysis. Due to the differing procedures for measuring HGS among the IGEMS studies, standardized maximum HGS was harmonized following IGEMS guidelines (33). Briefly, the sex- and study-specific standardized values for each wave were computed by normalizing the highest value obtained from all attempts within each wave based on the means and standard deviations (SDs) of maximum HGS from the first available waves in the respective studies. Given the recognized differences in HGS between men and women (11,12), we applied sex-specific standardized HGS values in our analysis. In the validation analysis, we used the first available HGS value for each participant. In analyses involving LTPA, we selected the HGS value closest in time to the LTPA measurement, prioritizing baseline assessments when available.

Leisure-Time Physical Activity

LTPA includes all PAs undertaken during free time, such as exercise, walking, heavy housework, and gardening (35). In the IGEMS studies, LTPA was self-reported using various items, which differed substantially across cohorts. To harmonize the data, continuous T-scores (mean = 50, SD = 10) were calculated, with higher scores indicating greater PA (36). For this study, the first available LTPA score was used for each cohort; however, LTPA was not assessed in the OCTO-Twin cohort.

Covariates

Participant age was defined as age at the time of HGS measurement. BMI was calculated as weight divided by height squared (kg·m⁻²). Weight and height data were collected at each study wave, with measured values prioritized; if unavailable, self-reported values were used. For this study, values closest in time to the HGS assessment were selected. The mean time difference between BMI/height and HGS assessments was 0.18 ± 0.55 yr (range: 0–6.70; N = 5019), with one observation excluded due to an extreme time difference (24.1 yr). Similarly, the mean time difference between LTPA and HGS assessments was 0.74 ± 1.66 yr (range: 0–11.50). Detailed distributions by analytic samples are provided in the Supplemental Tables 1 and 2 (Supplemental Digital Content, https://links.lww.com/MSS/D409). The first 10 genetic PCs were used as covariates to mitigate the effects of population stratification (32). Country of residence was also included as a covariate to account for differences in population characteristics and data collection procedures across the twin cohorts.

Genotype Data

Details of the genotyping procedures are provided in the Supplemental Digital Content, https://links.lww.com/MSS/D409 under Genotyping, quality control, and imputation.

Polygenic Score for HGS

PGS_hgs was calculated for the IGEMS cohorts as a weighted sum of risk alleles, using effect sizes derived from GWAS summary statistics and the pipeline described in our previous study (28). The linkage disequilibrium reference panel, summary statistics, and target sample were restricted to 1,006,473 common variants (MAF > 5%) from the HapMap3 (37) reference panel, excluding the MHC region on chromosome 6 (6p22.1–21.3) in FITSA. These variants are typically well imputed in populations of European and Finnish ancestry. In the remaining IGEMS cohorts, we selected HapMap3 (37) SNPs with good imputation quality (INFO > 0.8) and MAF >1% across all cohorts, resulting in 952,885 SNPs. Using these SNP sets, the summary statistics were processed with the Bayesian method SBayesR (38), employing a linkage disequilibrium reference panel derived from a random sample of 50,000 UKBB (39) individuals to weight the GWAS effects. The final PGS was computed separately in each IGEMS cohort using these SBayesR-derived weights. The distribution of PGS_hgs across the IGEMS cohorts is shown in Supplemental Figures 3 and 4 (Supplemental Digital Content, https://links.lww.com/MSS/D409).

Statistical Analyses

Validation of PGS_hgs.

Missing heritability analysis. We estimated pedigree- and SNP-based heritability of HGS using GCTA-GREML (21,22), with separate genetic relatedness matrices for family structure and SNP genotypes. Missing heritability was initially defined as the difference between these two estimates. Analyses were conducted separately within each country. However, SNP-based heritability estimates showed substantial instability across cohorts (Supplemental Table 3, Supplemental Digital Content, https://links.lww.com/MSS/D409), frequently approaching boundary values (near 0 or 1), likely reflecting limited sample sizes and reduced precision. Because these estimates were not sufficiently reliable, formal estimation of missing heritability was not undertaken. In contrast, pedigree-based heritability estimates were more stable across cohorts and consistent in magnitude with previously reported values (13), although the individual cohort estimates were not statistically significant (P > 0.05). To improve robustness and interpretability, we therefore combined pedigree-based estimates across cohorts using inverse-variance–weighted random-effects meta-analysis (40). Heterogeneity was assessed with I², Cochran Q, and τ² statistics.

Association between PGS_hgs and HGS. First, individuals were divided into PGS deciles to illustrate the linear association between standardized (mean = 0, SD = 1) PGS_hgs and sex-specific standardized HGS. This association was analyzed using linear regression for descriptive purposes. Second, the proportion of variation in HGS explained by PGS_hgs was examined using linear mixed models (41). Model predictors included age, country, the first 10 genetic PCs, and PGS_hgs. Since anthropometric characteristics influence HGS (42,43), we also conducted sensitivity analyses in which the association between PGS_hgs and HGS was re-estimated with additional adjustment for BMI and, in a separate model, for height to evaluate the robustness of the primary findings. Results were reported as the total variance explained by the model (R²) and the change in R² (∆R²) when PGS_hgs was added to the model after the other predictors. Analyses were conducted for the full sample and separately by sex.

Moderation of PGS_hgs effect on HGS by age, country of residence, and LTPA. We first fitted two interaction models to test whether the association between PGS_hgs and HGS varied by age and country of residence. These models included PGS_hgs × age and PGS_hgs × country interaction terms, respectively, along with the main effects of age, country, and the first 10 genetic PCs. To evaluate country-specific differences in genetic associations, the PGS_hgs × country interaction analysis was repeated with rotating reference groups, allowing for estimation of marginal effects within each country. Given evidence of heterogeneity across countries, we also fitted separate models within each country to examine country-specific associations between PGS_hgs and HGS. To investigate whether LTPA moderated the association between PGS_hgs and HGS, we extended the association model to include both the main effect of LTPA and an interaction term PGS_hgs × LTPA, representing a gene–environment interaction (26). This interaction term tested whether the effect of genetic predisposition on HGS varies depending on the level of LTPA, or conversely, whether the effect of LTPA on HGS differs across levels of genetic predisposition. BMI was included as a confounder in this effect modification analysis, given its known association with reduced PA (44) and its possible biological relevance in modulating the genetic expression of muscular strength. In a sensitivity analysis, height was included as an alternative anthropometric covariate. All P values for interaction effects were corrected for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) procedure (q < 0.05) (45), and meta‑analytic pooled estimates were obtained using random‑effects models (40). FDR correction was applied within cohorts and separately for sex-stratified interaction models.

Effect of LTPA on HGS

To investigate the role of the LTPA in muscle strength, we first assessed the association between LTPA and measured HGS with linear mixed models in the full sample. Model predictors included age, country, BMI, and LTPA; in sensitivity analyses, BMI was replaced with height. The association between LTPA and HGS was then examined using fixed-effect within-twin pair regression models (29), conducted for all twin pairs and separately for DZ and MZ pairs. Finding an association between LTPA and HGS, particularly within MZ twins, supports an independent effect of LTPA, beyond shared genetic and environmental influences. Additionally, to explore whether genetic predisposition modifies the relationship between LTPA and HGS, we included a PGS_hgs × LTPA interaction term in the model for DZ twins. For illustrative purposes, we examined marginal effects by categorizing PGS_hgs into low, intermediate, and high levels of genetic liability, based on its tertile distribution. Because MZ twins are genetically identical and lack within-pair variation in PGS, an interaction term between PGS_hgs and LTPA was not estimated in this group. Instead, regression coefficients for the association between LTPA and HGS were plotted across PGS_hgs tertiles among MZ twins to illustrate potential differences in effect size across levels of genetic liability. All fixed-effect within-twin-pair models were adjusted for age and BMI to account for confounding and any residual age differences due to nonsimultaneous assessments. Sensitivity analyses adjusting for height were not conducted in within-pair models because height is highly correlated in sib-pairs, particularly among MZ twins.

Mediation of PGS_hgs Effect on HGS by LTPA

To evaluate LTPA as a potential mediator in the PGS_hgs–HGS relationship, we first tested the association between PGS_hgs and LTPA in the full sample and in DZ twins using linear mixed models adjusted for age at LTPA assessment, sex, country, and the first 10 genetic PCs. In DZ twins, analyses were additionally stratified by PGS_hgs below versus at or above the median. Focusing on DZ twins allowed for within-pair variation in PGS_hgs to analyze whether the PGS_hgs–LTPA association differs depending on genetic liability, once shared family and genetic background were accounted for.

Analyses were conducted using IBM SPSS Statistics, version 30.0.0 (172), R, version 4.2.3, and Stata, version 18.0, software. PGSs and PCs were calculated using Plink 2.0. Statistical significance was set at P < 0.05. In all analyses using linear mixed models, family identifier was included as a random effect to account for within-pair dependency.

RESULTS

Validation of PGS_hgs

Demographic characteristics of participants from the IGEMS studies included in the validation analysis (N = 5103) are presented by country and sex in Table 1. The mean age of participants was 62.9 yr (SD = 10.3), and of the 5103 participants, 44% were females.

TABLE 1.

Sample characteristics by country and sex at the first handgrip strength assessment.

Country	N Individuals	Sex-Specific Std HGS,a Mean (SD)	Age HGS, Range (Median)
All	5103	0.38 (9.80)	40–96 (60)
Males	2858	0.48 (9.65)	45–92 (58)
Females	2245	0.27 (9.98)	40–96 (67)
Denmark	1870	0.48 (9.39)	45–96 (57)
Males	907	0.98 (9.13)	45–89 (57)
Females	963	0.01 (9.61)	45–96 (58)
Sweden	1210	0.92 (10.36)	40–94 (71)
Males	528	1.16 (10.14)	46–85 (70)
Females	682	0.74 (10.53)	40–94 (71)
Australia	271	0.04 (9.80)	70–94 (75)
Males	100	−0.70 (9.86)	70–92 (76)
Females	171	0.47 (9.77)	70–94 (74)
U.S. (Males)	1323	−0.05 (9.76)	51–67 (58)
Finland (Females)	429	0.00 (10.00)	63–76 (69)

HGS was standardized by sex and harmonized across IGEMS cohorts.

^aThe first possible HGS measurement and corresponding age.

HGS, handgrip strength; SD, standard deviation; Std, standardized; U.S., United States.

Heritability analysis. Supplemental Figure 5 (Supplemental Digital Content, https://links.lww.com/MSS/D409) shows the meta-analysis of pedigree-based heritability estimates across the IGEMS cohorts. The pooled heritability estimate was 60% (95% confidence interval [CI]: 45%–78%), and the pooled results indicate a statistically significant heterogeneity of heritabilities between studies (I² = 95%, P < 0.001), heritability estimates ranging from the lowest observed in Sweden (16%) and the highest in Australia (80%), while Finland, Denmark, and the United States showed intermediate values (42%, 61%, and 70%, respectively).

Association between PGS_hgs and HGS. The measured HGS increased linearly across deciles of PGS_hgs in the IGEMS cohorts (R² = 0.092, β = 0.714, SE = 0.052, P < 0.001; N = 5103 participants) (Fig. 2). Overall, PGS_hgs accounted for 4.6% of the variation (ΔR²) in measured HGS (Table 2). When stratified by sex, the explained variance was 4.3% in males and 5.2% in females. Sensitivity analyses adjusting for BMI or height (N = 5019) yielded results consistent with the primary analyses. Adjustment for height attenuated the variance explained by the PGS_hgs to varying degrees across cohorts, but the overall pattern of associations and relative cohort differences remained unchanged; Supplemental Tables 4 and 5 (Supplemental Digital Content, https://links.lww.com/MSS/D409).

FIGURE 2.

Association between polygenic score for handgrip strength (PGS_hgs) deciles and measured HGS. HGS was standardized by sex and harmonized across IGEMS cohorts. Fitted values from a linear regression of HGS on PGS_hgs deciles. Panel A shows the association across all IGEMS cohorts, Panel B in the Danish cohorts, Panel C in the Swedish cohorts, Panel D in the Australian cohort, Panel E in the United States cohort, and Panel F in the Finnish cohort. CI, confidence interval; HGS, handgrip strength; PGS, polygenic score.

TABLE 2.

Associations between a polygenic score for handgrip strength (PGS_hgs) and isometric handgrip strength in the IGEMS cohorts.

	zPGShgs			Full Model
	Beta	SE	P	R2 (%)	P	ΔR2 (%)	N
All	2.14	0.15	<0.001	9.6	<0.001	4.6	5103
Males	2.05	0.19	<0.001	10.2	<0.001	4.3	2858
Women	2.33	0.23	<0.001	10.3	<0.001	5.2	2245
Denmark	2.25	0.23	<0.001	15.3	<0.001	5.6	1870
Males	2.25	0.32	<0.001	19.2	<0.001	5.9	907
Women	2.23	0.33	<0.001	14.1	<0.001	5.3	963
Sweden	1.76	0.32	<0.001	6.4	<0.001	2.9	1210
Males	1.44	0.46	<0.001	6.8	<0.001	2.0	528
Women	2.08	0.42	<0.001	7.8	<0.001	3.8	682
Australia	2.84	0.50	<0.001	34.8	<0.001	7.7	271
Males	2.65	0.83	0.002	56.5	<0.001	6.3	100
Women	2.97	0.67	<0.001	26.9	<0.001	8.6	171
United States (Males)	1.98	0.29	<0.001	5.8	<0.001	4.0	1323
Finland (Females)	2.48	0.54	<0.001	9.39	0.002	6.1	429

The results are shown for the full model including zPGS_hgs as a predictor and adjusted for age and 10 principal genetic components. Analysis in the full cohort was additionally adjusted for country. HGS was standardized by sex and harmonized across IGEMS cohorts. Family number was included in the models as a random factor. ΔR² describes the difference in the coefficient of determination (R²) between the adjusted model with and without PGS_hgs.

HGS, handgrip strength; IGEMS, The Interplay of Genes and Environment across Multiple Studies; PGS, polygenic score; SE, standard error; z, standardized for normal distribution.

Moderation of PGS_hgs effect on HGS by age, country of residence. No statistically significant interaction between PGS_hgs and age was observed in the overall sample (β = –0.016, SE = 0.014, P_interaction = 0.258, N = 5103), nor in males (β = −0.006, SE = 0.021, P_interaction = 0.779, N = 2858) or females (β = −0.038, SE = 0.020, P_interaction = 0.058, N = 2245; Supplemental Table 6, Supplemental Digital Content, https://links.lww.com/MSS/D409). However, when combining interaction estimates across all five countries in a random‑effects meta‑analysis, a modest but statistically significant negative PGS_hgs × age interaction effect was observed (pooled β = –0.049, SE = 0.017, P = 0.003). Heterogeneity was negligible (τ² ≈ 0, I² = 0.26%; Q(4) = 2.546, P = 0.636), indicating that the age‑related attenuation of the genetic effect was highly consistent across cohorts. Sex-stratified meta-analyses suggested that this interaction was present among females (pooled β = –0.066, SE = 0.022, P = 0.002; τ² ≈ 0, I² = 0.03%; Q(3) = 2.142, P = 0.544) but not males (pooled β = –0.031, SE = 0.024, P = 0.206; τ² ≈ 0, I² = 0%; Q(3) = 2.235, P = 0.525).

A statistically significant interaction between PGS_hgs and country of residence was detected in the PGS_hgs × country model, using Australia as the reference group (Supplemental Table 7, Supplemental Digital Content, https://links.lww.com/MSS/D409). Specifically, Denmark (P_interaction = 0.031), Sweden (P_interaction = 0.009), and the United States (P_interaction = 0.009) showed interaction effects, suggesting that the strength of the association between PGS_hgs and measured HGS differs across countries (Supplemental Fig. 6, Supplemental Digital Content, https://links.lww.com/MSS/D409). Finland did not differ significantly (P_interaction = 0.174). No statistically significant interactions were observed when other countries were used as the reference group, reinforcing that the primary contrasts were with Australia. In country-stratified models, PGS_hgs explained between 2.9% and 7.7% of the variation in measured HGS, with the weakest association in Sweden and the strongest in Australia (Table 2). When further stratified by sex, the variance explained ranged from 2.0% (Swedish males) to 6.3% (Australian males), and from 3.8% (Swedish females) to 8.6% (Australian females). Given the unusually large effect estimates observed in the Australian cohort, we conducted an additional sensitivity analysis excluding this cohort. The association between PGS_hgs and measured HGS remained essentially unchanged in the full sample (R² = 0.091, β = 2.07, SE = 0.151, P < 0.001; ΔR² = 4.4; N = 4832) and when stratified by sex (males: R² = 0.096, β = 1.99, SE = 0.193, P < 0.001; ΔR² = 4.3; N = 2758; females: R² = 0.099, β = 2.23, SE = 0.234, P < 0.001; ΔR² = 5.0; N = 2074), indicating that the Australian sample was unlikely to have driven the overall findings.

Moderation of PGS_hgs effect on HGS by LTPA. The demographic characteristics of participants included in the PGS_hgs × LTPA moderation analysis are presented by country and sample in the Supplemental Table 2 (Supplemental Digital Content, https://links.lww.com/MSS/D409). The distributions of LTPA are shown in Supplemental Figures 7 and 8 (Supplemental Digital Content, https://links.lww.com/MSS/D409). A statistically significant interaction between PGS_hgs and LTPA was observed in the full sample (β = −0.034, SE = 0.014, P_interaction = 0.013, N = 4451). In sex-stratified analyses, the interaction was nominally significant among females (β = −0.043, SE = 0.020, P_interaction = 0.035, N = 2013), but did not remain significant after FDR correction (Supplemental Table 8, Supplemental Digital Content, https://links.lww.com/MSS/D409); this finding was consistent with the sensitivity analysis adjusting for height (Supplemental Table 9, Supplemental Digital Content, https://links.lww.com/MSS/D409). At the country level, a robust interaction that survived FDR correction was observed in the Australian cohort overall (β = −0.182, SE = 0.065, P_interaction = 0.006, q = 0.030, N = 203), driven primarily by females (β = −0.240, SE = 0.081, P_interaction = 0.004, q = 0.024, N = 127). Importantly, the interaction between PGS_hgs and LTPA remained statistically significant in the full sample in a sensitivity analysis excluding the Australian cohort (R² = 0.085, β = −0.028, SE = 0.014, P = 0.038, N = 4248). To evaluate the consistency of the PGS_hgs × LTPA interaction across cohorts, we conducted random‑effects meta‑analyses of the cohort‑specific interaction estimates. The pooled interaction effect was statistically significant in the BMI‑adjusted models (pooled β = –0.039, SE = 0.015, P = 0.010; I² = 12.45%), and remained significant when height was used instead of BMI (pooled β = –0.040, SE = 0.015, P = 0.009; I² = 21.03%). Sex‑stratified meta‑analyses showed a stronger interaction among females (BMI‑adjusted pooled β = –0.055, P = 0.024; height‑adjusted pooled β = –0.046, P = 0.025), whereas the interaction was weaker and not consistently significant among males. Full meta‑analytic results are provided in Supplemental Table 10 (Supplemental Digital Content, https://links.lww.com/MSS/D409). Together, these results suggest that the positive effect of LTPA on HGS may be more pronounced among individuals with a genetic predisposition for lower muscle strength (Fig. 3 and Supplemental Fig. 9, Supplemental Digital Content, https://links.lww.com/MSS/D409).

FIGURE 3.

Interaction between polygenic score for handgrip strength (PGS_hgs) and leisure-time physical activity (LTPA) on sex-specific standardized handgrip strength across all cohorts. The figure shows that the positive effect of LTPA on HGS was stronger among individuals with a lower PGS_hgs. The results are based on linear regression models incorporating the main effects of PGS_hgs and LTPA, along with their interaction term (PGS_hgs × LTPA). Models are adjusted for age at HGS measurement, BMI, country of residence, and 10 principal genetic components, with family number included as a random factor. Both PGS_hgs and LTPA were treated as continuous variables in the models. For illustrative purposes, the figure depicts T-score cut-off points, where the lowest value represents the 5th percentile, the highest value represents the 95th percentile, and intermediate values are spaced at 10-point intervals. Panels represent: (A) all cohorts combined; (B) females in all cohorts. Beta coefficients (b) and P values indicate the strength and statistical significance of the PGS_hgs × LTPA interaction term. The plotted lines represent estimated associations, and the error bars indicate 95% confidence intervals (CIs). BMI, body mass index; HGS, handgrip strength; LTPA, leisure-time physical activity; PGS, polygenic score.

Effect of LTPA on HGS

Descriptive characteristics of the total sample are shown in Supplemental Table 2 (Supplemental Digital Content, https://links.lww.com/MSS/D409), and twin pairs included in the within-pair analysis are presented in Supplemental Table 11 (Supplemental Digital Content, https://links.lww.com/MSS/D409), with means and SDs of within-pair differences shown in Supplemental Table 12 (Supplemental Digital Content, https://links.lww.com/MSS/D409). Higher levels of LTPA were associated with greater HGS in the total sample (β = 0.095, SE = 0.014, P < 0.001, N = 4451 adjusted with BMI; and β = 0.089, SE = 0.014, P < 0.001, N = 4451 adjusted with height). Similar associations were observed in the within-pair analyses of the full sample of twin pairs (β = 0.059, SE = 0.018, P = 0.001, N = 1982 pairs) and among DZ twins (β = 0.090, SE = 0.025, P < 0.001, N = 1124 pairs). In contrast, no association between LTPA and HGS was observed among MZ twins (β = 0.004, SE = 0.027, P = 0.894, N = 858 pairs) (Supplemental Table 13, Supplemental Digital Content, https://links.lww.com/MSS/D409). Although the interaction between PGS_hgs and LTPA was not statistically significant in DZ twins (β = –0.008, P_interaction = 0.752), both the marginal effects plot for DZ twins (Supplemental Fig. 10, Supplemental Digital Content, https://links.lww.com/MSS/D409) and the stratified regression coefficients for MZ twins (Supplemental Fig. 11, Supplemental Digital Content, https://links.lww.com/MSS/D409) suggest a weak but consistent trend indicating heterogeneity in the association between LTPA and HGS by genetic predisposition, with modestly larger estimated associations among individuals with lower PGS_hgs.

Mediation of PGS_hgs Effect on HGS by LTPA

PGS_hgs was not statistically significantly associated with LTPA in the full sample (β = 0.141, SE = 0.166, P = 0.392, N =4536), in DZ twins (β = 0.086, SE = 0.208, P = 0.680, N = 2623) or in stratified DZ twin analysis (below median PGS_hgs: β = 0.175, SE = 0.478, P = 0.714, N = 1302; at above/median PGS_hgs: β = −0.020, SE = 0.481, P = 0.967, N = 1321). Therefore, formal mediation analyses were not pursued.

DISCUSSION

In this multinational twin study, we aimed to extend the evaluation of a PGS_hgs to diverse international cohorts and to investigate whether its association with measured strength is modified by age, sex, country of residence, or LTPA. Our findings confirmed that higher PGS_hgs was associated with greater measured HGS, with modest differences across countries. The association remained consistent in sex-stratified models, accounting for slightly more variance in females than in males. Notably, we observed a statistically significant interaction between PGS_hgs and LTPA, suggesting that the association between LTPA and HGS differs according to genetic predisposition, with a more pronounced positive association observed among individuals with lower PGS_hgs. While our within-twin pair analyses offered limited support for an independent environmental effect of LTPA, the observed gene–environment interaction points to the potential of PA to attenuate genetic disadvantage in muscle strength.

Our findings align with existing literature, suggesting that HGS is moderately influenced by genetic factors, with heritability estimates (h²) ranging from 30% to 65% in classical twin studies (13). However, our meta-analysis of pedigree-based heritability estimates across the IGEMS cohorts revealed substantial variability, with estimates ranging from 16% in Sweden to 80% in Australia. The Australian estimate appears relatively high, whereas previous twin studies have reported heritability estimates for HGS in the Swedish cohort ranging from 47% in females to 72% in males (11). Classic twin-based designs may produce higher heritability estimates due to factors such as genetic interactions, including epistasis and dominance, gene–environment interactions, or violations of the equal environments assumption (21,46). Conversely, pedigree-based GCTA-GREML methods typically yield more conservative estimates (21). Yet our dataset included only MZ and DZ twin pairs, limiting the range of genetic relatedness. This constraint may reduce the method’s ability to distinguish additive genetic effects from nonadditive genetic and environmental influences, potentially inflating the heritability estimates. Additionally, the relatively small sample sizes, particularly in the Australian cohort, resulted in low statistical power, likely contributing to inflated point estimates and underscoring the need for caution in interpretation.

While heritability estimates quantify the proportion of trait variance attributable to genetic differences in a certain population, the predictive power of PGS reflects how much of that variance can be captured using the common SNPs identified in GWASs, under the assumption of additive genetic effects only (47). In the present study, the PGS_hgs was significantly associated with measured strength, explaining 4.6% of the variance overall. While this proportion reflects only a fraction of the total heritable variation, it is consistent with the modest predictive power typically observed for PGSs of complex traits (20,47). Extending our previous findings in Finnish females, where the PGS_hgs explained 6.1% of the variation in measured strength (28), the present results demonstrate similar predictive utility across international cohorts, with variance explained ranging from 2.9% to 7.7% in country-stratified models. The strongest associations were observed in Australia, particularly among females (8.6%), while the weakest were in Sweden. These findings reinforce the generalizability of the PGS but may also reflect differences in cohort composition, environmental exposures, or gene–environment correlations across countries (20,48). Notably, the higher predictive power observed in the Australian cohort may be partly attributable to shared genetic ancestry with the discovery sample used to build the PGS, as the original GWAS was based on the UK Biobank, and a substantial proportion of the Australian population has British or Irish ancestry (49). Conversely, the lower estimates in the Swedish cohorts may reflect more selective sampling. For example, the OCTO-Twin study included only like-sexed twin pairs aged 80 yr or older who were both still alive at recruitment (50), and the GENDER study comprised older, unlike-sex twin pairs (51), introducing potential biases due to age, survival, and cohort effects.

Genetic contributions to HGS have generally been found to be similar across sexes (12,13), though some prior research has pointed to a stronger heritable component in males and a greater influence of shared environmental factors in females (11,16). Additionally, one study found that a genetic score based on 16 HGS-associated SNPs predicted the phenotype more effectively in men than in women (19). In contrast, we observed slightly higher variance explained by the PGS_hgs in females (5.2%) than in males (4.3%), even if the overall performance was broadly comparable between sexes across all cohorts. The extent to which genetic effects on HGS are stable across adulthood also remains debated (11–15). While some evidence suggests age-related changes in genetic effects (13,15,25), we found no indication within our cohort that the predictive validity of the PGS diminished with increasing age. However, when interaction estimates were combined across all countries, a modest but statistically significant age‑related attenuation of the PGS_hgs effect emerged, particularly among females. This suggests that although individual cohorts may lack power to detect such interactions, the broader evidence indicates that genetic influences on HGS may weaken slightly with age. This pattern aligns with theoretical expectations that accumulating environmental exposures may weaken genetic influences over time (52). Nevertheless, the magnitude of this attenuation was small, and heterogeneity across cohorts was negligible, supporting the overall stability of genetic effects on HGS across adulthood. Sex-specific exposures, hormonal differences, and differential life expectancy may introduce additional environmental variability, particularly in women, which could obscure genetic effects (17,53). However, we found no strong evidence for meaningful sex differences in the predictive power of the PGS. Rather than reflecting underlying differences in genetic architecture, the modest sex variation observed in our study may result from cohort-specific characteristics, the specific variants included in the PGS, or limited power in sex-stratified analyses. Future research, including sex-stratified genome-wide association studies, may provide deeper insights into the mechanisms underlying sex differences in the genetic basis of HGS (24).

While robust detection of gene–environment interactions is generally difficult (26,54), our findings provide evidence for a gene–environment interaction between LTPA and genetic predisposition for HGS. Specifically, the positive association between LTPA and HGS was more pronounced among individuals with a lower PGS_hgs. This pattern aligns with the differential susceptibility framework, which posits that individuals with certain genetic profiles may be more sensitive to environmental influences, and thus more responsive to lifestyle interventions like PA (55). Our findings also support the previous research suggesting that PA can partially compensate for the genetic risk in various health-related traits (56,57). However, results from our within-twin pair analysis did not support a causal role of PA in promoting muscle strength. While LTPA was associated with greater HGS within DZ twin pairs, this association was not observed among MZ twins. The marked attenuation of the LTPA–HGS relationship in MZ twins implies that the association is at least partly attributable to shared genetic influences rather than an independent environmental effect of LTPA. Nonetheless, the observed trends in stratified and marginal effects, though not statistically significant, indicated that even among genetically identical individuals, higher levels of activity may offer some benefit, particularly for those with lower genetic predisposition. Furthermore, while LTPA moderated the relationship between genetic predisposition and HGS, it did not mediate it. The lack of a significant association between PGS_hgs and LTPA supports our previous notion that genetic predisposition for strength does not substantially influence individuals’ engagement in LTPA (58). However, it is important to note that PA itself is a heritable trait, with genetic factors accounting for a moderate proportion of individual differences in PA behavior (59). This raises the possibility that the association between LTPA and HGS may be confounded by shared genetic influences affecting both traits. Such pleiotropic effects could explain why the association between LTPA and HGS was attenuated in MZ twins, who are genetically identical and typically show high concordance in PA behavior, thereby limiting detectable variation in exposure. Taken together, our results underscore the complex interplay between genetic and environmental influences on muscular strength. Future studies should apply integrative analytical approaches that account not only for the main effects of genetic variants but also for their potential interactions with environmental exposures (54), such as PA, to better elucidate individual variability in response to lifestyle interventions.

This study has several strengths and limitations. A key strength is the relatively large overall sample size (N > 5000) drawn from multinational cohorts, which enhances statistical power and enables replication across samples. HGS and LTPA were harmonized across cohorts using validated methods (33,36), facilitating comparability across studies. However, this harmonization may have reduced some cohort‑specific detail, and the resulting LTPA measure reflects relative rather than absolute activity levels, limiting interpretation in terms of specific activity doses. In addition, LTPA was assessed via self-report questionnaires, which may be subject to recall or reporting bias. Although device‑based measures such as accelerometers could provide more precise estimates of overall activity, their use in large population studies has only recently become feasible (60). LTPA was also assessed at a single time point. While the time gap between HGS and LTPA assessments was generally small, for some individuals, LTPA was measured either several years before or after HGS. Consequently, the harmonized LTPA variable may not fully represent PA at the time of HGS measurement. Nevertheless, studies utilizing comparable self-reported measures suggest moderate stability across extended follow-up periods (61,62), and any resulting measurement error is therefore more likely to have attenuated rather than inflated the observed associations (63). From a practical perspective, current PA guidelines emphasize muscle-strengthening activities performed at least twice weekly as the primary modality for improving muscular strength (35). Because our harmonized LTPA measure aggregates diverse activity types and does not isolate resistance-type exercise, our findings should not be interpreted as evidence that general LTPA causally improves HGS. Rather, the observed interaction may reflect differential responsiveness to specific activity modalities, correlated health behaviors, or shared genetic influences affecting both PA and muscle strength, which could not be disentangled in the present data. Furthermore, the cross-sectional nature of the primary analyses constrains causal inference, though the within-twin pair design partially mitigates this by accounting for shared genetic and environmental confounding (29). In addition, while the use of PGS_hgs derived from the UK Biobank offers a robust genetic proxy, it may limit the generalizability of findings to more diverse populations due to ancestry-related differences (20). Although the overall sample was sizable, individual cohort sizes were relatively small, potentially limiting power in stratified or interaction analyses. Finally, while PGS capture aggregate genetic risk, they do not reflect the full complexity of gene–environment interplay (54) and may miss rare or structural variants contributing to muscular strength (20).

CONCLUSIONS

Our findings confirm that PGS_hgs captures population-level genetic influences on muscle strength. We observed evidence of effect modification by LTPA, such that the association between LTPA and HGS varied by genetic predisposition and was more pronounced among individuals with lower PGS_hgs. However, our results do not support a strong causal effect of general LTPA on muscle strength and should not be interpreted as evidence that nonspecific PA substitutes for muscle-strengthening exercise. These findings highlight the potential value of integrating genetic information to better understand heterogeneity in strength-related traits, while reinforcing the importance of activity modalities specifically targeted to muscular strength.

This study was funded by the Research Council of Finland (grant numbers: 341750, 346509, and 361981 to E. S., and 336823 to J. K.), the Juho Vainio Foundation (E. S.), and the Päivikki and Sakari Sohlberg Foundation (E. S). Funding for the IGEMS studies is detailed in the Supplemental Digital Content, https://links.lww.com/MSS/D409 under Funding for the IGEMS Studies.

The authors have no conflicts of interest to report. The results of the study are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation. The results of the present study do not constitute endorsement by the American College of Sports Medicine.

We thank the IGEMS and UK Biobank study participants for their valuable contribution to science, as well as the researchers involved in the original data collection. We also thank data managers Patricia St. Clair and Orla Hayden from the University of Southern California for their essential support with data access and processing. We are grateful to research assistant Joseph White from the University of Colorado Boulder for his valuable assistance with data analysis.

E. S. and P. H. conceived of the idea for the study, as well as contributed to the design of the study and interpreted the data. P. H., M. N., I. K. K., A. T., and J. W. performed the statistical analysis, assisted by T. P., C. A. R., J. K., and E. S. P. H. and E. S. drafted the first version of the manuscript. All authors reviewed the manuscript and revised it critically for important intellectual content. All of the authors have also approved the conducted analyses and the final version of the manuscript to be published. M. N., I. K. K., A. T., K. A. M., C. A. R., M. S. P., T. R., D. F., M. G., N. L. P., and J. K. contributed to IGEMS data collection and management. E. S. is a guarantor, and as the corresponding author, she attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Comprehensive information on data availability and access procedures is available in the Supplemental Digital Content, https://links.lww.com/MSS/D409 under Data availability for IGEMS studies.

Supplementary Material

msse-58-1914-s001.pdf ^{(1.4MB, pdf)}