No abstract

We introduce rugged fitness landscapes called match landscapes for the coevolution of feature-based assortative interactions between P ≥ 2 cognate pairs of tRNAs and aminoacyl-tRNA synthetases (aaRSs) in aaRS-tRNA interaction networks. Our genotype-phenotype-fitness maps assume additive feature-matching energies, a macroscopic theory of aminoacylation kinetics including proofreading, and selection for translational accuracy in multiple, perfectly encoded site-types. We compute the stationary genotype distributions of finite panmictic, asexual populations of haploid aaRs-tRNA interaction networks evolving under mutation, genetic drift, and selection for cognate matching and non-cognate mismatching of aaRS-tRNA pairs. We compared expected genotype frequencies under different matching rules and fitness functions, both with and without linked site-specific modifiers of interaction. Under selection for translational accuracy alone, our model predicts no selection on modifiers to eliminate non-cognate interactions, so long as they are compensated by tighter cognate interactions. Only under combined selection for both translational accuracy and rate do modifiers adaptively eliminate cross-matching in non-cognate aaRS/tRNA pairs. We theorize that the encoding of macromolecular interaction networks is a genetic language that symbolically maps identifying structural and dynamic features of genes and gene-products to functions within cells. Our theory helps explain 1) the remarkable divergence in how aaRSs bind tRNAs, 2) why interaction-informative features are phylogenetically informative, 3) why the Statistical Tree of Life became more tree-like after the Darwinian Transition, and 4) an approach towards computing the probability of the random origin of an interaction network.