We performed association mapping separately for each of the ten factors in 381 individuals. First, to reduce potential problems due to non-normality, each factor was quantile-quantile transformed to a standard normal distribution (quantile normalized). We then corrected for potential population stratification by controlling each factor for the first two principal components computed by applying SmartPCA [46] to the genotype data (Figure S3). Finally, these residuals were again quantile normalized to form the final 10 factors that were used for association testing.
