On the covariance of the Moore-Penrose inverse