Can principal components yield a dimension reduced description of protein dynamics on long time scales?