Here we have used three different approaches to investigate TFBS functional constraints based on variation data. In the first one, using position-by-position comparisons, we have found that variability at TFBS positions generally correlates with information content, consistent with previous findings based on cross-species comparisons in Drosophila and human for other TFs [35,36] and population studies in yeast [18]. It should be noted that the majority of PWMs used in this study have been derived from comparing the sequences across all binding sites in one genome detected by genome-wide ChIP studies. Variation analyses look at sequence diversity in a different 'dimension': that is, across individuals at a particular point in the genome for each given binding site. That these two dimensions generally correlate with each other (and often also with in vitro biochemical data such as SELEX and protein binding microarrays [46,47]) has been a reassuring confirmation of the general validity of PWM models to describe the sequence 'code' for the analyzed TFs. This, in turn, is an important prerequisite for using PWM scores to compute TFBS mutational load, a per-instance metric that combines the penetrance of a motif mutation with the loss of the PWM match it causes.
