Am I missing something completely obvious? Can anyone explain to me how the bias can have an effect on logFC rather than, e.g., abs(logFC)?

Notice that low AU values result in negative logFCs between two technical replicates (fig 2), with high AU values giving positive logFCs. This means that in one of the replicates, probes with high AU values resulted in high signal intensity, while in the other they resulted in low signal intensity, even though these are, to stress the point, Technical Replicates! They are supposed to be identical, and yet probe AU content has diametrically opposite effects in either sample.

To put it another way, imagine the scenario with 10 technical replicates. We split them into two groups of five, and observe the phenomenon shown in fig 2. But, since every sample is supposed to be identical, we could just as easily take any other (arbitrary) grouping of 5-vs-5, which would necessarily destroy the relationship! When we discover a relationship between apparently arbitrary variables and a biological variable, we must be very careful in the interpretation!

The only way I can think of that the symmetry between samples could be broken (resulting in the observed effect) is in the processing order of samples, assuming that they are processed in the same order as the labels (which sounds probable, to be sure), and that either RNA or the RNA dye degrade at different rates depending on the RNA's AU content.

Nevertheless, the effect reported here appears to me too strong to be due to what can surely only be a difference of minutes, if not seconds, in the processing of samples... Anyone else care to comment?