Adding vs. removing interaction changes significance of single predictors in lm()

Question

I\'m running two types of ANCOVA\'s in R on the same data. One model includes the interaction term between the two predictors, the other doesn\'t:

model<-l