On the role of secondary motions in turbulent square duct flow