Explicit Consideration of Solubility and Interaction Specificity in Computational Protein Design Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 20, 2019
  • Jacak, Ronald
    • Affiliation: School of Medicine, Department of Biochemistry and Biophysics
  • Most successes to date in computational protein design have relied on optimizing sequences to fit well for a single structure. Multistate design represents a new approach to designing proteins in which sequences are optimized for multiple contexts usually given by multiple structures (states). In multistate design simulations, sequences that either stabilize the target state or destabilize alternate competing states are selected. This dissertation describes the application of multistate design to two problems in protein design: designing sequences for solubility and increasing binding specificity in protein-protein interface design. Previous studies with the modeling program Rosetta have shown that many designed proteins have patches of hydrophobic surface area that are considerably larger than what is seen in native proteins. These patches can lead to nonspecific association and aggregation. We use a multistate design approach to address protein solubility by disfavoring the aggregated state through the addition of a new solubility term to the Rosetta energy function. The score term explicitly detects and penalizes the formation of hydrophobic patches during design. Designing with this new score term results in proteins with naturally occurring frequencies of hydrophobic amino acids on the surface without large hydrophobic patches. Designing protein-protein interfaces with high affinity and specificity is still a challenge for protein design algorithms. Multistate design is well-suited for addressing the problem of specificity because it can explicitly disfavor off-target interactions. Using a new implementation in Rosetta, multistate design is applied to the orthogonal interface design problem: redesign a protein with many partners to interact with only one of the partners. We use the RalA signaling network as the model system and make our design goal a redesigned RalA that only interacts with the effector RalBP1. Multistate design is able to recover several of the known mutants important for effector binding and predicts many new mutations that alter binding specificity. From in silico predictions, single-state design for Ral/RalBP1 by itself is not sufficient to destabilize RalA's interactions with its other effectors. Only multistate design is able to destabilize both of the negative states and give the desired interaction specificity.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Der, Channing
  • Campbell, Sharon
  • Elston, Timothy
  • Carter, Charles
  • Kuhlman, Brian
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2011

This work has no parents.