January 29, 2012
ERVs are sexed-up bivariate heritabilities

Complex diseases such as major depression show considerable heritability but linkage and genome-wide association studies have so far not identified a sufficient number of genetic variants to account for the observed genetic variance. One problem might be that the observed phenotype of interest, such as incidence of major depression, is too far removed from the underlying genetic variants to produce a strong enough signal to detect given the power of current techniques. To aid both genetic studies and to understand the underlying biology and physiology of a disease, the search is now on for endophenotypes: heritable, biological markers that are associated with the disease but that do not depend on disease state (Gottesman & Gould 2003). I got on to this topic after a presentation by @anamariafernand to our journal club.

To speed the identification of endophenotypes for mental illness, Glahn et al (2012) present the concept of an endophenotype ranking value (ERV)

$$\mathrm{ERV}_{ie} = \left | \sqrt{h^2_i} \sqrt{h^2_e} \rho_g \right |$$

between an illness \( i \) and endophenotype \( e \) where \( h^2_i \) and \( h^2_e \) are the heritabilities and \( \rho_g \) is the genetic correlation between \( e \) and \( i \). The ERV is useful as it goes in that it allows Glahn et al to detect several potential neurocognition, brain structure, and gene expression endophenotypes.

However, in quantitative genetic terms, the ERV is not anything new. Michelle pointed out that if you carry through the equation and drop the absolute value signs, the ERV formula reduces to

$$h_i h_e \rho_g$$

which is the same as the bivariate heritability (Falconer & Mackay 1996)

$$h_x h_y r_G$$

between two traits \( x \) and \( y \). So the main innovation is in using this standard quantity as part of a ranking scheme for identifying which phenotypes merit further exploration.