The causal path here is X -> E-> Y A, B, C, D can be referred to as "pretreatment" variables. There is one possible Backdoor path which is: X<-A->B<-D->E->Y However, this path is blocked by B. So we do not have to control for anything to meet the backdoor-criterion. If there is confounding because B or C is adjusted for, you can also choose to control for A or D to meet the backdoor-criterion. Traditional Statistician would control for B and C but this is wrong because it opens up noncausal path (by eliminating collider B) which confounds X and Y.