On Cartesian skeletons of graphs