We then considered the genome-wide properties of the mutational load metric. Recombination rates are distributed unevenly along Drosophila chromosomes (Figure 4a, dashed lines) [22,43]; however, we did not observe an association between the TFBS load and local recombination rates (Figure 4a; Figure S5 in Additional file 1). Rather, the analysis of selected 'high-load hotspots' (average load per 100 kb window >5e-3) revealed regions in which motifs with deleterious variation mapped in close proximity to other motifs for the same TF (see Figure 4b for examples). This suggested that TFBS mutations may be partially 'buffered' by neighboring motifs. Consistent with this model, we found that motifs for at least four Drosophila TFs tolerated a significantly lower load when present as 'singletons' compared to sites with two motifs (Figure 4c), particularly for evolutionarily conserved instances. Interestingly, TFs whose binding sites had a higher mean load generally had more motifs per ChIP region (Figure 4d), raising the possibility that a higher number of motifs may allow a TF to tolerate a higher load. The PWM scores of variable motifs were similar to those of 'constant' motifs in their proximity (Figure 4e); it is unlikely, therefore, that these variable motifs are non-functional a priori.
