Evolution of transcription factor binding sites in mammalian gene regulatory regions: handling counterintuitive results.

The remodeling of transcription factor binding sites is one of the major engines of evolutionary change, yet almost all available examples of this involve sites from regulatory modules brought online during development. Developmental modules are known to enjoy some type of canalization, which allows considerably more cryptic change than would be assumed for nondevelopmental modules. It remains an open question, then, how frequently binding site remodeling occurs in general. There are strong intuitive reasons to expect that regulatory constancy, and hence binding site conservation in general, is the rule, yet little systematic work has been done to verify this. In the present article, we show that the most obvious way of approaching this problem--which is simply to collect experimentally verified binding sites from the literature without further analysis, create multispecies alignments, and apply conservation algorithms--leads to counterintuitive and ultimately inaccurate results. This is because of the low complexity of typical binding sites and, consequently, because of the frequency with which strings resembling legitimate sites occur throughout the genome. In vitro results can easily be confounded by this. Applying one traditional conservation algorithm and two novel algorithms to a data set that ought to be representative of binding sites in general, but which is taken from the literature directly, we find that only 58% of sites appear to be conserved. However, after the data set is carefully vetted against binding site overloading, and after the likelihood of a specific type of compensatory evolution is evaluated, conservation rates as high as 94% appear reasonable.