Robust inference of positive selection on regulatory sequences in human brain

A long standing hypothesis is that divergence between humans and chimpanzees might have been driven more by regulatory level adaptions than by protein sequence adaptations. This has especially been suggested for regulatory adaptions in the evolution of the human brain. There is some support for this hypothesis, but it has been limited by the lack of a reliable and powerful way to detect positive selection on regulatory sequences. We present a new method to detect positive selection on transcription factor binding sites, based on Orr’s sign test applied to a machine learning model of binding. Unlike other methods, this requires neither defining a priori neutral sites, nor detecting accelerated evolution, thus removing major sources of bias. The method is validated in flies, mice, and primates, by a McDonald-Kreitman-like measure of polymorphism vs. divergence, by experimental binding site gains and losses, and by changes in expression levels. We scanned the signals of positive selection for CTCF binding sites in 29 human and 11 mouse tissues or cell types. We found that human brain related cell types have the highest proportion of positive selection. This is consistent with the importance of adaptive evolution on gene regulation in the evolution of the human brain.