Background. Much has been written about the “missing heritability” for complex traits. For alcohol and nicotine dependence (AD, ND) there are estimates of heritability of up to 65% from twin studies, yet few causal variants have been replicated from GWAS studies, suggesting that individual effect sizes of SNPs must be small. New statistical genetic techniques have been developed which allow estimation of total variance associated with all SNPs on a GWAS chip, but this has yet to be applied to AD and ND.
Methods. Over 8000 participants in our twin-family studies who had used alcohol or cigarettes at some stage of their lives were individually genotyped with Illumina 370K or 660K chips and 7.034M genotypes were imputed from HapMap 3 and 1000-Genomes data. The GCTA program of Yang, Visscher et al is used first to detect the degree of relatedness between apparently unrelated subjects, based on a set of about 300,000 SNPs pruned for LD. Phenotypic similarity is then regressed on IBS sharing for all possible relative pairs to estimate the total amount of variance due to SNPs on the chip.
Results. Based on GCTA analysis for other complex traits we expect that SNP associated variance accounts for about half the heritability estimated from conventional genetic epidemiology designs.
Conclusions. The gap between the SNP-associated variance estimated by GCTA and twin and family estimates of heritability may be due to several factors. The tag SNPs on the chip are not in perfect LD with the causal SNPs; for other traits, simulation has shown that correcting for imperfect LD raises the SNP “heritability” by about 10%. Commercial chips only interrogate common SNPs so large effects of rare SNPs are simply not captured. Estimates from simulations suggest that this could account for another 20% of variance. There are large sections of the genome containing highly repetitive DNA which are poorly tagged by current chips, and where substantial proportions of variance may be hidden.