Hints of hidden heritability in GWAS

Gibson, 2010

Although susceptibility loci identified through genome-wide association studies (GWAS) typically explain only a small proportion of the heritability, a classical quantitative genetic analysis now argues that considering together all common SNPs can explain a large proportion of the heritability of these complex traits. A related study provides recommendations for the sample sizes needed in future GWAS to identify additional susceptibility loci.

While GWAS have helped us identify genetic variants associated with many different types of diseases, these associations only explain a few percent of the heritability of complex disease. As I have addressed before, there are a few different reasons that SNPs in GWAS are not capturing heritability of disease: 1) we are improperly estimating heritability of the disease (or phenotype), or 2) the common variants of GWAS are not capable of (statistically significantly) capturing the genetic heritability of these phenotypes. It is important to note that these two explanations are not mutually exclusive.

If the former is true, we need to go back and refine our protocol and understanding of the problem of inheritance and more accurately estimate the proportion of phenotypic variation explained by inheritance.

If the latter is true, it would be an interesting question to examine why it might be true. Many researchers have proposed many different hypotheses, and generally they are not mutually exclusive. Rare variants, epistasis, epigenetics, and geneotype-environment interactions are listed by Greg Gibson as potential sources of heritability. Also noted, is the possibility that complex traits emerge from the interaction of thousands of (common) variants with small effects.

So let the great debate begin; what is the reason that GWAS do not identify causal variants (in most cases). Is it 1) that some rare variants of high impact, 2) or many common variants with small impact are affecting phenotypes. Both of the previous scenarios would lead to a situation of low statistical significance. The former because if a variant is rare it will only be present in a few people in the population. For instance, in a study of 5000 people, if the allele only has a prevalence of 0.2% in the population, no subjects would be expected to be homozygous and ~20 subjects are expected to be heterozygous. The stronger the effect, the more likely it would be for a statistical analysis to pick up the association. But just a few phenotypic outliers could dramatically alter the p-value for the association between the rare variant and the phenotype.
The latter is hard to decipher (ie - find statistically significant associations) because if a common variants only explain a small percentage of the variance, a few ouliers could also change the results. Another problem is that if there are 5000 people in the study and 1000 common variants are affecting the trait, there is the potential problem of multicollinearity caused by not having enough data to fit all of the parameters (for the 1000 genes).

Gibson states, "It is unlikely that GWAS will ever be sufficiently powered to uncover even the majority of the heritability" of complex disease. My reply to that is, what then should we be doing in increase our explanatory power (that is, what tests should we be running to elucidate and assign heritability to genes, regulatory elements or networks of these component parts).

"This [paper by Yang et al] presents an elegant argument that most of the heritability [of height] is hidden rather than missing and hence, that there is no pressing need to invoke more complex genetic mechanisms to explain height."
If that is the case, then I want to know how to uncover the hidden heritability because that is going yield causal variants which will lead to molecular mechanisms for the condition, be it height or complex disease like T2D or CVD.