In:
Science, American Association for the Advancement of Science (AAAS), Vol. 381, No. 6656 ( 2023-07-28)
Abstract:
Protein-protein interactions mediate biological functions important for cell physiology. Interacting proteins coevolve over millennia through sampling of mutations, largely at the protein-protein interface, to achieve the “best fit” for the required function. Protein engineering methods can generate large libraries of amino acids in a protein binding site for screening against other proteins of fixed sequence, mirroring one-half of the evolution process. However, it has been challenging to develop in vitro systems to coevolve two proteins against one another by using “library-on-library” approaches that can recover matched pairs of coevolved proteins. An efficient synthetic system for bidirectional, simultaneous protein-protein coevolution could serve as a platform to simulate natural coevolution. It could also be a way to engineer large numbers of protein-protein complexes with different recognition properties for biotechnology applications. RATIONALE The crux of the problem for library-on-library selections is the loss of connectivity of discrete pairs of interacting proteins during the selection process. We developed an approach for efficient recovery of matched pairs from very large libraries of amino acids on both sides of a protein-protein interface. Our solution was to display the protein as a complex on the surface of yeast. We made libraries of amino acids within the interface of this protein complex representing ~1 billion variants and recovered only the protein complexes. In this fashion, the yeast that we recovered contained the sequences of both mutant interacting proteins. RESULTS Using this strategy, we created several types of coevolution libraries that showed that we could recover interacting pairs of thousands of interface mutants. The mutant complexes displayed a vast diversity of specificities, orthogonalities, and affinities and revealed unanticipated ways that the interfaces structurally compensated to mutations and mediated specificity versus promiscuity. With such a large volume of data, we used systems and network-level analysis of the binding interactions to map evolutionary pathways and the thermodynamic basis for interface evolution. We explored the potential of machine learning to engineer previously unknown interfaces using our large collection of coevolved sequence pairs. Specifically, we investigated if embeddings from protein language models, pretrained on the evolutionary history of extant protein sequences, could be used to model our coevolved protein-protein interfaces. Our objective was to make in silico predictions about mutations not present in our initial library, as well as complexes involving novel amino acids. Through a process known as “transfer learning,” we were able to predict and subsequently validate complexes with amino acid sequences that were not included in the original library. This method allowed us to increase the amino acid diversity of our libraries, surpassing the experimental limits of yeast display. CONCLUSION The integration of a synthetic coevolution platform with machine learning has allowed us to interrogate a protein-protein interaction with exceptional granularity, but also to use this information prospectively. Using this approach, it is possible to revisit basic principles of protein-protein binding with systems-level data depth. We expect the synergy between our experimental coevolution platform and computation to stimulate development of applications in cell engineering such as orthogonal switches and AND gates. Depiction of proteins coevolving in vitro, through energetic connectivity (“lightning bolts”) of amino acids at their binding sites. The relationships between these coevolving amino acids enable computational prediction of protein complexes. Illustration: Chris Garcia and Eric Smith
Type of Medium:
Online Resource
ISSN:
0036-8075
,
1095-9203
DOI:
10.1126/science.adh1720
Language:
English
Publisher:
American Association for the Advancement of Science (AAAS)
Publication Date:
2023
detail.hit.zdb_id:
128410-1
detail.hit.zdb_id:
2066996-3
detail.hit.zdb_id:
2060783-0
SSG:
11
Permalink