I'm with a dilemma of whether it is right to measure the mediation effect of variables B, C, and D without a direct link from A to E. I am using the PLS-SEM technique in smartpls.
You require a direct effect so the indirect effect estimates are not biased by the equality that the sum of the indirect effects plus direct effect equate to the total effect. Without it, it is analogous to fitting a regression by forcing the intercept through the origin (ie setting it to zero)