264816 Use of Aggregation Prediction As a Function of Protein and Excipient Properties for Lyophilized Formulation Design

Thursday, November 1, 2012: 12:30 PM
326 (Convention Center )
Brock C. Roughton1, Lavanya K. Iyer2, Haider S. Tarar3, Anthony I. Pokphanh3, Elizabeth M. Topp2 and Kyle V. Camarda3, (1)Bioengineering Graduate Program, University of Kansas, Lawrence, KS, (2)Department of Industrial and Physical Pharmacy, Purdue University, West Lafayette, IN, (3)Department of Chemical and Petroleum Engineering, University of Kansas, Lawrence, KS

Protein stability is a major concern during development of a protein drug product, with physical and chemical instability playing major roles. Lyophilization, or freeze-drying, is often used to help stabilize protein formulations. However, protein aggregation may still occur at significant levels in lyophilized formulations. Protein aggregation is undesirable, resulting in product deactivation and an increased potential for immunogenic response in patients. Predicting the aggregation propensity  of lyophilized proteins is desirable for aiding formulation decisions, such as whether lyophilization should be pursued and whether stabilizing additives would be required. Such additive molecules are referred to as excipients and often comprise the bulk of a drug product. If lyophilization of a protein is necessary, prediction of the effect of added excipients on aggregation would provide a useful tool for formulation design. Several excipients, such as sucrose and glycine, are commonly used to prevent aggregation in lyophilized formulations but are not effective for all proteins and exhibit different protective effects from protein to protein.

A predictive model for aggregation of lyophilized proteins as a function of protein descriptors and excipient molecular descriptors is presented. The protein descriptors are comprised of physical/structural descriptors as well as computationally determined descriptors developed in previous literature for predicting aggregation based on primary or tertiary structure. Connectivity indices describing atomic and bonding configurations were used as molecular descriptors for the excipients. The protein and excipient descriptors were used to correlate percent native protein retained after lyophilization and reconstitution, as determined experimentally by size-exclusion chromatography (SEC). Losses in percent native protein correspond to protein aggregation. Selection of descriptors was performed using an exhaustive search method to provide sufficient fit while avoiding overfitting, as evaluated by R2 and the Akaikie Information Criterion (AIC). The predictive power of the resulting correlation was evaluated using leave-one-out cross-validation (LOOCV). Initial results showed that use of protein descriptors alone was sufficient to predict aggregation on a formulation-by-formulation basis but was not able to predict aggregation across formulations. Fifteen proteins were considered under sucrose, buffer, urea, and glycine formulations.

From the set of proteins used in the initial work, four proteins were selected based on diversity of descriptor values. An exhaustive search was performed to identify the four proteins that best spanned the descriptor space, specifically for the descriptors that were selected for the sucrose and glycine formulations. The four selected proteins were lyophilized with 15 different excipients, resulting in 60 unique formulations. The excipients spanned the structural space between glycine and sucrose. After lyophilization and reconstitution, SEC analysis was used to quantify protein lost to aggregation. Combination of protein and excipient descriptors was successful in predicting aggregation across all 60 formulations through use of a single correlation. The resulting correlation was used in a computational molecular design (CMD) framework to optimally design an excipient for a given protein that minimized aggregation. A stochastic optimization method was used to solve for optimal excipient molecular structures and prediction intervals were calculated for the solutions at a 95% confidence level. The correlation was also used to search for proteins that would show minimal aggregation for a given excipient commonly used in lyophilized formulations.  Use of the predictive model provides a basis for rational lyophilized formulation design, using either approved molecules or optimally-designed excipients for a given protein.

Extended Abstract: File Not Uploaded