Protein interactions, unfolding and aggregation from low to high protein concentrations via coarse-grained molecular modeling and experimental characterization

Roberts, Christopher J.The diverse behavior of protein solutions can be attributed to the collection of microscopic interactions between solvent, protein, buffer and other cosolute molecules. These interactions can dictate different solution properties such as native and non-native aggregation, phase separation (liquid-liquid separation, crystallization, etc.), opalescence and elevated solution viscosity. These are monitored and controlled during the development and manufacturing of protein solutions for different health care, food, and other industrial applications. Better understanding and control of the interactions among solution molecules can help to optimize the development and manufacturing of protein solutions, leading to a decrease in overall product and process development costs. However, the complex nature of the molecular-scale interactions makes it challenging to identify the key contributions to these interactions and resulting solution behaviors. ☐ Protein-protein interactions, as opposed to protein-solvents and protein-cosolute interactions, are the most studied and arguably better understood solution interactions used during the development of protein-based products. Historically, these interactions have been experimentally characterized at low protein concentrations (c2) due to several instrumental and theoretical limitations. This has been done using different interaction parameters such as the protein-protein second osmotic virial coefficient (B22), the diffusion interaction parameter (kD) or their surrogates. These are measured using static and dynamic light scattering (SLS and DLS, respectively) techniques, analytical ultra-centrifugation (AUC) or self-interaction chromatography (SIC), among others. Recent efforts have focused on developing methodologies to measure interactions at much higher c2, where SLS coupled with Kirkwood-Buff (KB) solution theory are used to interpret the high-c2 behavior of protein solutions via the protein-protein KB integral (G22). By contrast, simple colloidal (spherical) models have historically been used to capture protein interactions as a function of solution environment and c2, with larger emphasis on low-c2 conditions. However, these models might lack enough molecular resolution to capture the c2-dependent behavior of protein interactions, as well as enough structural definition to model anisotropic protein molecules, such as monoclonal antibodies (mAb). ☐ The viability of using coarse-grained (CG) molecular models to capture low- to high-c2 protein-protein interactions is examined here for a series of proteins and solutions expanding from globular to mAb proteins. For dilute c2, several different generic molecular descriptions of a given mAb molecule are used to evaluate differences across protein steric interactions (protein excluded volume effects) and the protein molecular volume using advanced Monte Carlo algorithms. This comparison allows one to find models that can self-consistently capture low- and high-c2 packing behavior. These models are later used to evaluate the protein solution osmotic compressibility as a function of c2 using transition matrix Monte Carlo algorithms. The results highlight shortcomings of using spherical models to capture antibody solution interactions, while anisotropic mAb-like models considerably improve consistency between protein packing and molecular volume. For globular proteins, the spherical assumption is expected to accurately represent both protein excluded and molecular volumes, so no additional refinements were performed. ☐ Those CG models that were found to accurately capture the packing behavior of proteins from low to high c2 are coupled with simple potential of mean force (PMF) models to capture short-ranged non-electrostatic attractions (from van der Waals and solvation effects), screened electrostatic interactions (from the protein anisotropic charge distribution) and highly flexible protein regions, when needed. This is done to minimize the number of model parameters yet capture the main contributions to protein-protein interactions. Experimental measurements of excess Rayleigh scattering were performed as a function of solution pH, buffer identity and concentration, and sucrose content, for a series of NaCl concentrations to estimate B22 values and decouple the electrostatic and short-ranged non-electrostatic attractive contributions to B22. The former dominates at low total ionic strength (TIS) and the latter at high-TIS conditions. The results are used to refine model parameters at low c2, which are later used to predict high-c2 excess Rayleigh scattering behavior for similar solution conditions, up to 160 g/L protein concentration. Higher resolution CG models are used to further evaluate strong attractions observed at low TIS, caused by highly anisotropic electrostatic attractions. The results indicate excellent agreement between experimental and predicted values when protein interactions are repulsive to weakly attractive (B22/B22,ST > -3, with B22,ST representing the excluded volume contribution). Strongly attractive interactions can deviate qualitatively from the experimentally observed behavior. The CG models are also used to obtain additional domain-domain potentials of mean force as a function of c2, pH, and TIS to gain insights into preferential interactions across protein domains. ☐ Added cosolutes, such as sugars, surfactants and other stabilizers, can mediate the interactions between proteins, thus affecting the overall solution properties. These cosolutes are expected to affect the solvation shells of proteins, which can be cast in the framework of preferential interactions as a mean to better describe the effects of adding cosolutes to protein solutions. However, current preferential interaction frameworks can only be used to interpret experimental protein-cosolute interactions for ternary systems (three-component solutions) while most protein solutions usually contain a minimum of four components (water, protein, buffer and a cosolute). A new derivation is presented here for protein-solvent and protein-cosolute interactions that can be applied to an arbitrary number and concentration of components for dilute c2. The new framework was used to compare differences of preferential interactions in the absence and presence of buffer, for a series of cosolutes (sucrose, trehalose, sodium phosphate and sodium chloride) and added cosolute concentrations, at fixed solution temperature, pH and pressure. Protein preferential interaction measurements showed that protein-cosolute interactions in the presence (quaternary systems) and absence (ternary systems) of buffer are statistically indistinguishable as the buffer contribution was effectively zero. This would allow to treat quaternary solutions as pseudo-ternary systems, in agreement with the newly derived framework. Additional measurements were performed to identify the effects of adding sucrose to a mAb solution into the protein-protein interactions measured via excess Rayleigh scattering, which showed that the addition of sucrose results in increased protein-protein repulsions. Adding sucrose to a mAb solution resulted in preferential solvation/accumulation of sucrose around the protein surface, in good agreement with the orthogonal measurements of protein-protein interactions and computer simulations. ☐ Finally, experimental measurements and molecular scale simulations were performed for a series of Ala-rich peptides to gain insights on the effects of: (i) modifying side change hydrophobicity and (ii) modifying the chain length into the unfolding and aggregation behavior of peptides. A four-beads-per-amino acid (4bAA) CG model was coupled with Replica Exchange Molecular Dynamics (REMD) to compute unfolding and aggregation transitions and identify intermediate states during the unfolding behavior of five polypeptide sequences with similar chemistry. Circular dichroism (CD) was used to unveil the unfolding and aggregation behavior along with peptide secondary structure of the same sequences as a function of temperature and polypeptide concentration. The results showed a linear increase in thermal stability with chain length, with a decrease in stability with decrease hydrophobicity. This is found to be caused by the increase in solution entropy with the decrease in side-chain hydrophobicity. ☐ Overall, the results in here demonstrate: (i) the effects of solution environment in mediating protein-protein interactions, (ii) how that can be studied within the framework of preferential interactions, (iii) the viability of coupling experimental measurements and computer simulations at low c2 to predict high-c2 interactions and better understand the effect of solution formulations at the microscopic level, and (iv) how more detailed CG models can be used to both capture the anisotropic surface charge distribution of proteins as well as the unfolding and aggregation propensities of polypeptides. The approaches can be generalized to other proteins of interest outside those studied here.University of Delaware, Department of Chemical and Biomolecular EngineeringPh.D