To reduce the effect of common environmental factors shared by related individuals, the genetic relationship matrix A = [Aij] for all samples was calculated first. If Aij was larger than a specified threshold, either individual i or j was excluded in the analysis. In our first approach, we chose a threshold of 0.025, which implied that the individuals more closely related than second cousins are excluded in our analysis. Ultimately, we retained 1,838 unrelated individuals. We included age, sex and five principal components as fixed effects. Then we estimated the variance parameters σu2 and σe2 using the Restricted maximum likelihood method (REML) [12,32].
