PCA projections is carried out by solving least squares equations rather than an orthogonal projection step. This is approriate if PCs are calculated using samples with little missing data but it is desired to project samples with much missing data onto the top PCs.
Next I computed eigenvector averages for all populations in order to make the output more readable. So each symbol represents up to around 20 samples. Corresponding eigenvalues are 13.858230 and 10.064209.