Background: Low or declining handgrip strength, a widely used proxy of muscular fitness and marker of frailty, predicts a range of morbidities and all-cause mortality. We compared different strategies to perform genome-wide association analyses of longitudinal handgrip measurements using sequence data from the NHLBI Trans-Omics for Precision Medicine (TOPMed) program.
Methods: We analyzed 12,337 ethnically diverse participants (77.3% European-Ancestry (EA), 21.6% African-Ancestry (AA), 1.1% other) from six cohort studies (Amish, ARIC, CHS, FHS, HyperGEN, and WHI). Participants had between two and six handgrip measures per exam, from 1-9 separate exams over time, totaling 32,236 observations. Per participant, we selected the maximum observation at each exam and used all exams (ALL), one exam (ONE), or the mean of all exams (MEAN). Linear mixed-effects models were used to conduct association analyses with GMMAT, adjusting for age, sex, height, BMI, study, age×sex, BMI×sex, study×sex, and 11 ancestry principal components, with random effects for study and kinship; and for individual if multiple measures, to account for correlation across exams.
Results: Leveraging multiple measures per participant (ALL) resulted in a 7-13% increase in effective sample size. Genomic control inflation factors were similar and p-values/effect sizes highly correlated (r≥0.85) between analyses. We detected a significant association (lead variant rs4793937, MAF=0.18, P=3.4×10-8, PEA=1.6×10-6, PAA=0.004) in the homeobox B3 (HOXB3) gene, a handgrip GWAS locus (lead variant rs2288278, MAF=0.38, Willems et al., 2017). The TOPMed rs2288278 association was modest (PEA=0.004, PAA=0.92) and the rs4793937 association remained significant after conditioning on rs2288278 (P=2.2×10-6), suggesting two distinct signals (r2EA=0.41; r2AA=0.11). Both variants are expression and DNA methylation quantitative trait loci of HOXB genes in human skeletal muscle tissue (GTEx & FUSION). Validation of these results in the UK Biobank is ongoing.
Conclusion: Leveraging multi-ethnic populations along with longitudinal data in mixed-effects models help identify additional signals in GWAS loci.

Epigenetic clocks have shown that patterns of DNA methylation from blood cells are strongly correlated with chronological age. Those with accelerated methylation age (methylation age that is greater than expected for chronological age) are at higher risk for several diseases of aging and death, but the biological processes underlying such advanced epigenetic age are incompletely understood. Clonal hematopoiesis of indeterminate potential (CHIP) results from somatic mutations in blood stem cells and may be found in ~20% of the elderly. CHIP most commonly arises due to mutations in the DNA methylation altering enzymes, TET2 and DNMT3A, and also associates with increased risk of death, cancer, and cardiovascular disease. Whether CHIP associates with accelerated methylation age is unknown. We used methylation and whole genome sequencing data from several cohorts in TOPMed together comprising thousands of persons to show that CHIP is strongly associated with increased epigenetic age acceleration. The most consistent association is observed for intrinsic aging (2.8 ± 0.36 years, p < 2.5 x 10-14), which measures epigenetic aging that is independent of changes in cell composition, while a more variable association was seen with extrinsic aging (2.5 ± 0.46 years, p < 4.6 x 10-7), which captures epigenetic aging that is driven by changes in cell composition. We also analyzed the gene-specific effects of CHIP mutations on epigenetic aging, dividing our CHIP carriers into DNMT3A, TET2, and all other CHIP mutations. We found that the increase in intrinsic age acceleration seen in CHIP was very consistent across different genes, while the increase in extrinsic aging was lower with DNMT3A mutations and higher in TET2 mutations. The epigenetic clock software we used can also predict cell type composition and leukocyte telomere length (LTL) from methylation. We observed an increased predicted proportion of CD8+/CD28-/CD45RA- T-cells and decreased predicted LTL. Future experiments should seek to determine whether there is a causal relationship between CHIP and epigenetic age acceleration and whether intrinsic or extrinsic age acceleration is predictive of health outcomes in those with CHIP.

Background: Low or declining handgrip strength, a widely used proxy of muscular fitness and a marker of frailty, predicts a range of morbidities and all-cause mortality. We performed association analyses of handgrip strength in 16 targeted GWAS loci (Willems et al, 2017) using longitudinal handgrip measurements and whole-genome sequence data from the Trans-Omics for Precision Medicine (TOPMed) program.
Methods: We analyzed 12,342 ethnically diverse participants from six cohort studies (Amish, ARIC, CHS, FHS, HyperGEN, and WHI) who had between two and six handgrip measures per exam, from one to nine separate exams over time. We selected the maximum observation per participant at each exam and used all exams per participant, totaling 32,266 observations. We conducted association analyses with GMMAT using linear mixed models adjusted for age, sex, height, BMI, study, age×sex, BMI×sex, study×sex, and 11 ancestry principal components. We included random effects for study, kinship and participant to account for correlation of measures across exams. We also performed stratified analyses in European (EA) and African (AA) ancestry (NEA=9,538; NAA=2,663 participants). We defined loci using a 500kb window around the lead 2017 GWAS variants.
Results: Participants had a mean age of 66.6 years (SD: 13.8) and a mean handgrip of 29.4 kg (SD: 10.8). We detected a significant (P≤6.9×10-6) 17q21 association (lead TOPMed variant rs4793937, MAF=0.18, P=1.9×10-7, PEA=4.2×10-6, PAA=0.003) in the homeobox B3 (HOXB3) gene. The association of the lead 2017 GWAS variant (rs2288278, MAF=0.34) was restricted to EA and modest in our sample (PEA=0.005), which may be due to the younger age of the 2017 GWAS participants (mean age (SD): 56.8 years (8.0)). The association of rs4793937 remained significant after conditioning on rs2288278 (P=2.1×10-6), suggesting two distinct signals. The correlation between rs4793937 and rs2288278 was higher in EA (r2=0.41) than in AA (r2=0.11), which may explain why the 2017 GWAS identified only one signal in HOXB3. Using GTEx and FUSION quantitative trait loci (QTLs) results of expression (eQTLs) and DNA methylation (mQTLs) in human skeletal muscle tissue, we found that both variants are eQTLs of HOXB3 and mQTLs of HOXB3 CpG sites.
Conclusion: Leveraging multi-ethnic populations along with longitudinal data can help identify additional signals in GWAS loci associated with complex traits. Further investigation of the function of HOXB3 in muscle cell is needed.

Background: Handgrip strength is a widely used proxy of muscular fitness and a marker of frailty. Low or declining handgrip strength predicts a range of morbidities and all-cause mortality. We compared different strategies to perform genome-wide association study on longitudinal handgrip measurements using sequence data from the Trans-Omics for Precision Medicine (TOPMed) program.
Methods: A total of 12,342 participants from six cohort studies (Amish, ARIC, CHS, FHS, HyperGEN, and WHI) had between two and six handgrip measures per exam from one to nine separate exams over time, totaling 32,266 observations. We selected the maximum observation per participant at each exam and used per participant: 1) all exams (ALL), 2) one exam (ONE), or 3) the mean of all exams (MEAN). We conducted association analyses with GMMAT using linear mixed models adjusted for age, sex, height, BMI, study, age×sex, BMI×sex, study×sex, and 11 ancestry principal components, with random effects for sex, study and kinship. In addition, for the ALL analysis, we included a random effect for participant to account for correlation across exams.
Results: Leveraging multiple measures per individual resulted in a 5-10% increase in effective sample size. The -log(P) from the three analyses were highly correlated (rALLvsONE=0.85, rALLvsMEAN=0.97, rONEvsMEAN=0.88). The genomic control inflation factors for the three analyses were similar; the ALL analysis had the lowest lambda for low frequency and rare variants.
Conclusion: When available, the use of multiple observations per individual using mixed effect models can increase effective sample size and thus power while controlling type I error.