Specificity in Protein-Protein Interactions: High-Throughput Characterization of Rationally Designed and Naturally Evolved Coiled-Coil Networks

As the major effectors of cellular processes, proteins are crucial to all biology. Although proteins are regulated in many fashions, protein-protein interactions are ubiquitous across different classes of proteins. In particular, proteins must interact specifically with certain partners to recapitulate the biology that constitutes life, despite cells containing hundreds of thousands of proteoforms, some fraction of which are highly similar to the intended target. Understanding how specificity in protein-protein interactions occurs has been challenging to investigate because prior techniques were limited to in throughput and ability to pinpoint sequences of interest. We create a high-throughput two-hybrid assay that marries gene synthesis with a next-generation sequencing readout, allowing us to investigate only those interactions of interest with a single experiment providing a quantitative characterization of tens of thousands of interactions. We use this to first to investigate specificity in designed coiled-coils—small alpha-helical proteins which despite a simple hydrophobic interface exhibit high a high-degree of specificity. After validating our assay on a previously published set of coiled-coils, we iteratively find increasingly large sets of orthogonal proteins, proteins where each on-target interaction is specifically preferred to all off-target interactions. In total we screen more than 26,000 interactions in three experiments, and use our data and improve coiled-coil design algorithms while also finding the largest sets of orthogonal proteins to date. While specificity can be designed with large changes to the protein sequence, nature must come by specificity through the slow tinkering of evolution. To investigate the origins of specificity in nature we characterized a bZip family descended from an ancestral homodimer where the extant paralogs do not heterodimerize. We use ancestral reconstruction to trace protein-protein interactions in the coiled-coil domain across the PAR and E4BP4 family, back to the ancestor of humans and cnidarians. We find specificity does not appear once, but rather eight times across our tree, and while the process begins immediately, the final acquisition of specificity takes substantial time. Finally we find that once interactions are lost they never return, and that there is no direct selection for the acquisition of specificity between paralogs