In wireless edge caching, it is usual to cache the contents likely to be requested. Contrary to this traditional approach, we show that under heterogeneous content preference, caching contents that have less probability to be requested but are likely to interfere can increase the achievable rate of users by exploiting interference neutralization. Optimal content placement, which maximizes the average achievable rate, is compared with two extreme content placements: most-requested and most-interfering. Optimal content placement is shown to balance caching interfering contents against caching requested contents.