Several model-to-model transformation languages have been primarily designed to easily address the syntactic and semantic translation of read-only input models towards write-only output models. While this approach has been proven successful in many practical cases, it is not directly applicable to transformations that need to modify their source models, like refactorings. In this paper we investigate the application of a model-to-model transformation language to in-place transformations, by providing a systematic view of the problem, comparing alternative solutions and proposing a transformation semantics to address this problem in ATL.