Looking at the structure of either planar or trench vertical MOSFET, there is clearly an area where only thin oxide separates the gate from the drain. The drain-body junction must be engineered is such a way that under reverse bias, drain-side potential gradient is benign enough that voltage in the uppermost area of the drain near the gate doesn't exceed oxide breakdown limit.
It seems that the worst-case condition (which still doesn't result in failure for other reasons) is gate at -Vgs(max) and drain at +Vds(max), because then the gate-drain oxide is exposed to slightly more than +Vgs(max). Interestingly, this seems to imply that every MOSFET guaranteed to withstand such condition is also certain to withstand slightly more than Vgs(max) on its gate terminal when it is driven normally, i.e. gate positive with respect to source. Hmm.
Note that if you apply HV across drain and gate while leaving the source floating, leakage will presumably charge the source terminal up and cause oxide breakdown.