Does one need the O(epsilon)- and O(epsilon^2)-terms of one-loop amplitudes in an NNLO calculation ?

This article discusses the occurences of one-loop amplitudes within a next-to-next-to-leading order calculation. In an NNLO calculation the one-loop amplitude enters squared and one would therefore naively expect that the O(epsilon)- and O(epsilon^2)-terms of the one-loop amplitudes are required. I show that the calculation of these terms can be avoided if a method is known, which computes the O(epsilon^0)-terms of the finite remainder function of the two-loop amplitude.