Reshuffled PWMs were generated by ten per-position permutations of the 'real' PWMs. Reshuffled motif matches were detected within the 200 bp proximity of real TF binding sites at the same PWM score thresholds as the real motifs. Position-wise variation data obtained for each permuted motif instance was then 'de-reshuffled' to match the positions of the real PWM to compute the total diversity per permuted motif position. For human motifs, the score thresholds used elsewhere in the study resulted in very low numbers of reshuffled motif instances detected near the corresponding TF binding sites. To overcome this, analyses in Figure 2 used slightly relaxed score thresholds for both real and reshuffled human motifs, adjusted such that the total number of motif instances detected with the 10 reshuffled PWMs was at least 1.5-times higher than the number of real instances for each TF.
