Evolution and information content of optimal gene regulatory architectures

Abstract

Evolution of gene regulatory sequences is shaped by their genotype-phenotype (GP) map, but this map itself can evolve via changes in the molecular machinery that reads out the regulatory sequences. Using population genetics theory, we derive an optimality principle, implying that the evolved GP maps will associate fit phenotypes with larger numbers of possible genotypes. Mathematically, this builds on analogies between population genetics and statistical physics, as well as optimal coding in information theory. The results are particularly interesting in the context of transcriptional regulation, where optimal values of certain parameters that determine the gene regulatory architecture can be derived even without a complete knowledge of the molecular mechanisms involved. We illustrate the theory using a simplified model of transcription factor (TF) binding to cis-regulatory elements (CREs): the fraction of possible CREs across the entire genotype space that bind a given TF evolves to match the fraction of CREs under selection to bind that TF. Similar "frequency matching" results are expected to emerge generically, because selection implicitly optimizes the entire regulatory architecture that determines the "replicated GP map" for all CREs, while it simultaneously adapts towards individually fit CREs on that GP map. We discuss the subtleties and limitations of the theory related to biophysical constraints and the need for mutational robustness. Lastly, our theory suggests a possible mathematical definition for evolvability with predictive power that can be tested in evolutionary simulations and, ultimately, in comparative genetics studies.