(241d) Use of Aggregation Predictions in the Design of Protein Formulations

A major concern in the development of therapeutic protein formulations is protein aggregation, and excipients are often used to increase the stability of such formulations. An ideal excipient binds with aggregation prone regions on the protein to limit interaction of that region with other protein molecules. The goal of this project is to predict aggregation-prone regions and use this information within a computational molecular design formulation to develop excipients which interact favorably with these regions. To build a predictive model based both on the protein's structural characteristics and on the molecular structure of an excipient, aggregation was evaluated for a set of model proteins and excipients using computational estimation techniques, and these results were compared to experimental values in the literature. Two algorithms were evaluated for the prediction of protein aggregation: Aggrescan (http://bioinf.uab.es/aggrescan/) and SAP (Spatial Aggregation Potential). Aggrescan assigns each amino acid an aggregation propensity score. An aggregation prone region, or hot spot, is identified by a sequence of amino acids with high propensities. SAP uses molecular simulation to determine hot spot regions that are hydrophobic and solvent accessible. In this algorithm, each residue is scored and the results are mapped to the three-dimensional protein structure. Both SAP and Aggrescan predicted similar hot spot regions for each protein, with SAP being slightly more accurate when compared to experimental data. However, the computational effort required to use the SAP algorithm is not compatible with a computational screening tool. A simple correlation was also developed linking the percent hydrophobic surface area of each protein to experimental aggregation values. These results demonstrate that aggregation potential can be estimated effectively using structural parameters of a given protein which are easily accessed. The effect of excipient structure also needs to be quantified in order to design excipients to stabilize specific protein systems. A docking simulation algorithm was employed using model proteins together with the excipients trehalose, poly(vinylpyrrolidone) and guanadine hydrochloride. The SAP algorithm was used to identify hot spot regions, and the docking algorithm was used to measure binding energies for each excipient-hot spot pair. Smaller binding energies suggest more favorable protein-excipient interactions. Correlations were then developed relating topological indices to the binding energy for each protein-excipient pair. The correlations are used within a molecular design framework to determine novel excipient structures that exhibit minimized binding energies. Results are analyzed using hydrogen-deuterium exchange and FTIR analysis to verify the extent of protection by the excipients.