Theoretically, TFBS mutations can be buffered at many different levels - starting from the motif itself that may 'absorb' a number of mutations due to a permissive consensus, to the level of CRMs (for example, homotypic motifs and protein interaction partners), cis-regulated genes (involving possible 'backup' by shadow enhancers [55]) as well as further along the regulatory network [56] - which may potentially explain the apparent redundancy that is often observed in the network architecture, both at the level of cooperative TF binding to enhancers and multiple 'cross-talking' pathways [57]. Consistent with previous observations at individual CRMs [58], our observations suggest that much variation is buffered immediately in cis, via the redundancy of TFBS consensus sequences, neighboring homotypic motifs or other factors preserving regulator binding (or at least the overall CRM output). If true, this model may explain two of our preliminary observations that we initially found puzzling: that the levels of tolerated load did not significantly vary depending on the functional annotation of regulated genes (not shown) and that candidate Drosophila enhancers with seemingly very deleterious mutations at Bin,
