Life course variations in the heritability of body size

Background: It has been shown recently that whole genome data can
facilitate estimation of genetic contributions to a variety of traits via a mixed
model framework as implemented in GCTA and R/SAS (Yang et al. Nat
Genet 2010, 42:565-9; Zhao & Luan. J Prob Stat 2012, doi 10.1155/
2012.485174). Our aim was to use this approach to investigate the life
course variations in heritability of body size. Methods: We analysed height,
weight and body mass index variables at 11 time-points in 2,452 individuals
(1,225 men, 1,227 women) born in 1946 and enrolled in the MRC National
Survey of Health and Development (NSHD), with genotypes at 147,949
single nucleotide polymorphisms (SNPs) on Metabochips which were subsequently
imputed to 506,255 according to the 1000Genomes project. We
obtained genome-wide kinship matrices using genotypes at SNPs on Metabochips
and genotypes at all SNPs, which were used in mixed models as
implemented in the computer program GCTA. Results were also compared
to those obtained using an alternative procedure of kinship estimation in
PLINK and mixed models in R. Results: In line with earlier findings that
specific genetic variants have variable temporal effects in this cohort (Hardy
et al. Hum Mol Genet. 2010; 19:545-552), we observed age-related variations
in heritability estimates. Estimates based on genotypes at SNPS on
Metabochips and genotypes at all SNPs were comparable but generally
lower than recently reported GCTA estimates with mean(range) being
0.09(0-0.50), 0.11(0-0.20), 0.10(0-0.22) for height, weight and body mass
index, respectively. Variation in estimates was also seen between alternative
procedures. Conclusion: This work supports the utility of large-scale genotype
data in heritability estimation and highlights the age-related variability
in genetic contributions to body size across the life course. Further work
will be to distinguish the effects of established variants and to consider
estimates in a unified longitudinal model including contrast with models
assuming various degrees of temporal homoscedasticity.