A New Pattern in Protein Evolutionary Sequence Information Robustly Identifies Functionally-Important Amino Acid Positions

Abstract

For decades, protein engineers have tried to mimic the natural evolution of functional variation. One approach has been to use the evolutionary sequence information of protein families to predict which positions were mutated to vary the function among homologs. These positions have been previously detected by amino acid conservation, co-evolution between pairs of positions, or patterns of change that parallel phylogenetic trees. In all of these analyses, scores are usually rank-ordered and thresholded to reveal the top pairwise scores. Here, we instead treated un-thresholded, pairwise co-evolutionary scores as weighted networks (using the pairwise scores for all-vs-all positions) and analyzed them with graph theory. Results showed that these calculations can be used to bypass a major complication of co-evolution studies: For a given sequence alignment, alternative co-evolution algorithms usually identify different, top pairwise scores. In contrast, when network centrality scores were compared among five commonly-used, mathematically-divergent co-evolutionary algorithms, agreement was significantly improved (Spearman Rˆ2 values increased by more than 0.2). Further, most of the top central positions had multiple, moderate pairwise scores instead of a few strong scores, which makes them evolutionarily distinct from the top pairwise positions identified by each algorithm. Positions with large centrality scores occurred at key structural locations and/or were functionally sensitive to mutations. We conclude that network centrality calculations reveal a robust pattern of evolutionary constraints - detectable by divergent co-evolutionary algorithms - that occur at key protein locations. Finally, we discuss the idea that multiple patterns of amino acid change appear during evolution that, together, give rise to emergent protein functions.