Genome-wide association analyses of physical activity and sedentary behavior provide insights into underlying mechanisms and roles in disease prevention

Abstract

Although physical activity and sedentary behavior are moderately heritable, little is known about the mechanisms that influence these traits. Combining data for up to 703,901 individuals from 51 studies in a multi-ancestry meta-analysis of genome-wide association studies yields 99 loci that associate with self-reported moderate-to-vigorous intensity physical activity during leisure time (MVPA), leisure screen time (LST) and/or sedentary behavior at work. Loci associated with LST are enriched for genes whose expression in skeletal muscle is altered by resistance training. A missense variant in ACTN3 makes the alpha-actinin-3 filaments more flexible, resulting in lower maximal force in isolated type II_A muscle fibers, and possibly protection from exercise-induced muscle damage. Finally, Mendelian randomization analyses show that beneficial effects of lower LST and higher MVPA on several risk factors and diseases are mediated or confounded by body mass index (BMI). Our results provide insights into physical activity mechanisms and its role in disease prevention.

Fig. 1. Overview of the four self-reported physical activity and sedentary traits and correlations with objectively assessed traits.

a, An overview of the four self-reported physical activity and sedentary traits. b, Phenotypic (upper left) and genetic (lower right) correlation coefficients between the four self-reported physical activity and sedentary traits studied here and three accelerometer-assessed traits quantified in UK Biobank participants. AccMod, accelerometer-assessed proportion of time spent in moderate intensity physical activity; AccSed, accelerometer-assessed proportion of time spent sedentary; AccWalking, accelerometer-assessed proportion of time spent walking; SDC, sedentary commuting behavior; SDW, sedentary behavior at work.

Fig. 2. Main results of GWAS and downstream gene prioritization for LST and MVPA.

a, Circular Manhattan plot summarizing the results from European ancestry meta-analyses for LST and MVPA. Outer track, LST; inner track, MVPA. Genome-wide significant variants (P < 5 × 10⁻⁹) are highlighted in orange for loci associated with MVPA and in blue for loci associated with LST. b, Dendrogram showing the 101 independent association signals in LST- and MVPA-associated loci from European ancestry or multi-ancestry meta-analyses. Moving outwards from the center are: (1) chromosome; (2) lead SNP identifiers, in orange for loci associated with MVPA, in blue for loci associated with LST; (3) the most promising gene(s) prioritized in the locus (closest genes are highlighted by filled circles); and (4) the approach(es) by which the gene was prioritized, that is, DEPICT gene prioritization (Dg) or tissue enrichment (Dt); SMR of eQTL signals in blood (Sbl), brain (Sbr) or skeletal muscle (Ssm); credible variants identified by FINEMAP that (i) are coding and likely to have a detrimental effect on protein function (Fcadd) or (ii) show evidence of three-dimensional interactions with the candidate gene in central nervous system cell types (Fcrt); activity-by-contact (ABC) in 26 relevant tissues and cell types; a contribution to enrichment for altered expression in skeletal muscle following a resistance training intervention (RTsm); and/or proximity to an association signal for spontaneous running speed (Ms), time run (Mt) or distance run (Md) in a GWAS of 100 inbred mouse strains.

Fig. 3. Validation of associations with MVPA and LST using PGSs in BioMe participants of three ancestries.

a,c, The best performing PGSs for MVPA (a) and LST (c) were derived using logistic/linear regression analyses; that is, those with the highest incremental R² above and beyond models with only sex, age and the top ten principal components. This was accomplished using inclusion thresholds of P < 0.1101 for MVPA and P < 0.14 for LST. b,d, The association—examined using a logistic regression analysis—of MVPA with the PGSs for MVPA (b) and LST (d) in individuals of African (AA, n = 2,224), European (EA, n = 2,765) and Hispanic (HA, n = 3,206) ancestry in data from the BioMe BioBank. Dots and error bars show OR and 95% CI.

Fig. 4. Genetic correlations of four self-reported physical activity traits with complex traits and diseases.

Results are based on published GWAS with P < 4.6 × 10⁻⁴ for at least one physical activity or sedentary trait. Darker colors reflect higher negative (purple) or positive (red) correlation coefficients. GC, genomic control; HDL, high-density lipoprotein; HOMA-B, homeostasis model assessment of beta-cell function; HOMA-IR, homeostatasis model assessment of insulin resistance; PGC, psychiatric genomics consortium.

Fig. 5. MR analyses between LST, MVPA, BMI and complex diseases.

a, Median causal estimates for MR analyses using the CAUSE method and causal estimates from the MR-PRESSO method after outlier removal and accounting for horizontal pleiotropy. b, The causal effects of LST on complex risk factors and diseases without (in orange) and with (in blue) adjusting for BMI. Dots and error bars show the estimated causal effect sizes and 95% CI. ADHD, attention deficit hyperactivity disorder; T2D, type 2 diabetes.

Fig. 6. Allele p.635Ala in ACTN3 results in a more flexible ACTN3 homodimer.

a, ACTN3 is a homodimer of two antiparallel filaments, with each filament consisting of an N-terminal actin binding domain (ABD, blue), followed by a structural region comprised of four spectrin repeats (gray) with a C-terminal calmodulin (CAM) homology domain (cyan). b, The glutamate residue side chain in position 635 of ACTN3 (p.Glu635) interacts primarily with the arginine in position 638 and the glutamine in position 639. c, The α-helix comprised of residues adjacent to ACTN3 residue 635 (ACTN2 628) exhibits a pronounced kink in ACTN2 (green) at this α-helical turn compared with ACTN3 p.Glu635 (blue) and p.635Ala (orange), decreasing the likelihood of interactions under load with R631, whereas the alanine substitution of ACTN3 p.635Ala precludes any side chain interactions with neighboring residues p.Arg638 or p.Glu639. d, The r.m.s.f. of the spectrin repeat structural region of the ACTN3 dimer for a 150 ns MD simulation for variants p.Glu635 (blue) and p.635Ala (orange, higher MVPA) and ACTN2 (green) (bottom), with the difference in r.m.s.f. between ACTN3 variants shown mapped to the spectrin repeat region (top) with ±0.3 nm difference (red, positive and blue, negative). e, Umbrella sampling of ACTN3 variants p.Glu635 and p.635Ala and ACTN2 with orange, blue and green traces representing the potential of mean force for ACTN3 variants p.635Ala (orange) and p.Glu635 (blue) and ACTN2 (green) ±1 s.d. The reaction coordinate is the distance between the two ABD centers of mass of each dimer, a negative value indicating a shorter distance between the two ABDs. Inset shows the relaxed dimer at reaction coordinate of 0 nm (top) and the direction and effect on the compressive force. f, Single fiber experiments show a higher maximal force and fiber power during isotonic contractions after an eccentric exercise bout in type II_A fibers from an individual homozygous for p.Arg577 and p.Glu635 (blue) compared with type II_A fibers from three p.Arg577 homozygous, p.Glu635Ala heterozygous individuals (orange); and from four p.577Ter homozygous individuals (green).

Extended Data Fig. 1. LST-associated loci are enriched for genes with altered expression in skeletal muscle following resistance training.

Fold-change plot in log scale for the ratio between: (1) the proportion of genes in physical activity-associated loci that showed an altered expression in skeletal muscle (FDR < 0.01) across five categories: inactivity, acute bout of resistance exercise, acute bout of aerobic exercise, resistance training, or aerobic training; and (2) the proportion of all genes that showed an altered expression following such (in)activity in the MetaMex database (PMID: 31980607). Tested loci were MVPA or LST-associated loci. In a given set of loci, we either considered only the genes nearest to the lead SNP, or all genes within 1 Mb of the lead SNP. Only loci harboring at least five genes with altered gene expression levels after intervention were included in this figure. A one-sided Fisher exact test was used to calculate the P-value for enrichment.

Extended Data Fig. 2. A sensitivity analysis shows the analysis of altered gene expression following resistance training is robust to FDR threshold.

We examined the effect of different FDR thresholds on Fisher’s exact test results for the enrichment analysis of alteration in gene expression in skeletal muscle following resistance training. Red square, genes within 1 Mb of the LST lead SNP; green circle, genes within 1 Mb of the MVPA lead SNP; blue triangle, nearest gene LST lead SNP; purple diamond, nearest gene MVPA lead SNP. The horizonal dotted line indicates nominal significance level (P < 0.05), and the vertical dashed line indicates the FDR threshold that was used. FDR thresholds explored range from 0.001 to 0.5.

Extended Data Fig. 3. DEPICT-derived tissue enrichment of MVPA and LST.

a, MVPA. b, LST. SNPs with P < 1 x 10⁻⁵ for association in the European ancestry GWAS of men and women combined were used as input. The dashed line indicates the FDR corrected significance threshold (FDR < 0.05).

Extended Data Fig. 4. Cell type prioritization using CELLECT for MVPA and LST.

a, Prioritization of 115 Tabula Muris cell types across 19 tissues identified two cell types from the brain as significantly associated (stratified linkage disequilibrium score regression) with MVPA (left) and LST (right), namely oligodendrocyte precursor cells and neurons (shown in black; Bonferroni-corrected significance threshold, P < 0.05/115). b, Prioritization of 265 mouse nervous system cell types identified 13 and 45 cell types from 12 distinct brain regions as significantly associated (stratified linkage disequilibrium score regression) with MVPA and LST, respectively (highlighted; Bonferroni-corrected significance threshold, P < 0.05/265.

Extended Data Fig. 5. Protein-protein interactions involving 17 of the 46 candidate genes in GWAS-identified loci prioritized by at least two approaches.

Protein-protein interactions were visualized using String. LONRF2 and CHST10 were prioritized in loci associated with MVPA; the remaining genes were prioritized in loci associated with LST.