We counted cDNAs from a given gene as showing alternative polyadenylation site usage if 3' UTR length estimates varied by at least 400 bp - smaller variation could be real, but may not be distinguishable from error in our size estimates.