Predicting, controlling and molecular understanding of therapeutic protein self-interactions: a combined experimental and computational approach

Date
2022
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Protein self-interactions (PSI) influence biophysical properties like aggregation, solubility, viscosity, and phase separation of drug products. PSI are a function of protein identity (e.g., sequence, charge, and hydrophobicity distribution) and formulation conditions like temperature, pH, and ionic strength. The main physical forces determining the net protein self-interactions are screened electrostatic attractions and repulsions, non-electrostatic short-range interactions, and steric repulsions. Strongly attractive protein self-interactions can trigger biophysical properties through promoting aggregation rates and poor solubility. In terms of drug product development, this can hinder the developability of a drug candidate. Molecular simulations can play a key role in PSI prediction and screening candidates based on measurable experimental quantities. They can also be used to understand the underlying physics of PSI and aid in redesigning the sequence of poorly behaved proteins toward a more stable protein. ☐ This dissertation includes five chapters and uses computational and experimental methods to facilitate the prediction of PSI through coarse-grained modeling and identify the potential residues that contribute most to PSI. ☐ In the second chapter, a series of coarse-grained (CG) models were used to predict PSI in terms of the second osmotic virial coefficient (B22) for several monoclonal antibodies (MAbs) as a function of total ionic strength. In previous reports, these MAbs have shown a broad range of attractive and repulsive self-interaction in different formulation conditions2–4. The models were also compared based on the computational burden and scalability. Comparing different CG models through predicting B22 values yields guidance for selecting CG models that optimizes prediction accuracy and computational burden. ☐ In the third chapter, the 1bC/D model (One-bead-per-charge-site-and-per-domain) introduced in the second chapter was used to identify the most influential charged residues that could significantly reduce the strong net electrostatic attraction of a poorly soluble MAb. This was done by predicting B22 values after mutating each charge residue via charge-swap and charge-to-neutral mutations. A series of variants that were predicted to have the greatest improvement were made experimentally and tested using light scattering. Then the simulated B22 of variants were compared directly against experimental values. Compared to the experimental results, the model resulted in quantitative/semi-quantitative prediction of protein-self interactions. Also, the results showed that the qualitative pattern of pairwise interactions is not necessarily by itself as an indicator of protein self-interaction behavior and/or which variants will most likely result in improved net protein self-interactions. ☐ In the fourth chapter, a new computational approach was introduced to deal with fluctuating charge residues in protein solutions when the pKa of a titratable charged atom is close to the pH of a solution. The problem arises in molecular and CG molecular simulations as using partial charges is not physical for those charge sites that can be fully ionizable. This is more significant for residues like Histidine, where its pKa (~6) is close to a range of physiological conditions and relevant industrial product conditions. The new computational approach allows fluctuating charge residues based on a given probability governed by the Henderson-Hasselbalch equation. Compared to static charge simulation, the new approach improved quantitatively predicting B22 values against experimental data, particularly when the pH of the solution is close to pKa of titratable residues.
Description
Keywords
Biopharmaceuticals developability, Computational biophysics, Protein characterization, Protein engineering and antibody design, Protein stability, Protein-protein interactions
Citation