Precalculated interaction vs. MANOVA built interaction

#1
Hi there,

In MANOVA you can build a model yourself and define the interaction
effects that you want to test. However, when I alternatively calculate the interaction myself beforehand and enter it as a covariate in the MANOVA the significance of the main effect (of the variable in the interaction) changes. All the results concerning the other variables remain identical.

My questions: what is causing the difference and what method should I use?

Thanks for your help! Luc.