Computational protein design with multiple functional and structural constraints.

摘要

In this work, a series of computational tools to predict protein sequences compatible with a given three-dimensional protein structure and a set of structural or functional constraints are presented. First, a "multi-constraint" protocol to design protein sequences optimized for multiple criteria is presented. For a number of multi-specific signaling and structural proteins, interface sequences are computationally designed to bind multiple interaction partners and design predictions are compared to naturally occurring amino acid sequences. In many cases, the multi-constraint design algorithm successfully "added up" diverse sequence preferences seen among several characterized binding partners, resulting in the prediction of highly native-like interface sequences. Multi-constraint designed sequences were also found to have overall weaker predicted binding scores than sequences designed to bind only a single interaction partner, suggesting that multi-specificity may come at a cost of affinity. This section concludes by discussing two distinct mechanisms for maintaining multi-specific binding, and providing examples of how the design protocol presented here might be used to rationally design proteins with multiple functional roles.;A method to predict sets of amino acids tolerated at protein-protein interfaces is presented next. By incorporating a flexible backbone move, termed "backrub", computational predictions of amino acid tolerances at a model interface, human growth hormone with its receptor, are found to closely mimic sequences observed in an experimental phage display dataset. The importance of incorporating backbone flexibility when predicting amino acid tolerance to substitution is discussed and an automated method to computationally predicting sequence libraries to enable challenging protein engineering problems is given.;Finally, a protocol for predicting single amino acid substitutions tolerated for a protein of great biological relevance, HIV-1 protease is presented. In this work multiple constraints present on the HIV-1 protease fold and function are integrated and a reduced set of amino acid mutations (able to be reached by a single mutation at the nucleotide level) was considered. Despite the simplifications inherent in the model, ∼80% of amino acid substitutions that occurred in clinical HIV-1 protease sequences were predicted as tolerated. This work further demonstrates that use of a single, fixed backbone as a structural template for design results in overall poorer predictive performance than designing on an ensemble of either crystallographically determined or computationally generated backbone structures.