| dc.description.abstract | The field of pharmaceutical chemistry is currently struggling with the question of how to relate changes in the physical form of a macromolecular biopharmaceutical, such as a therapeutic protein, to changes in the drug's efficacy, safety, and long term stability (ESS). A great number of experimental methods are typically utilized to investigate the differences between forms of a macromolecule, yet conclusions regarding changes in ESS are frequently tentative. An opportunity exists, however, to relate changes in form to changes in ESS. At least once during the development of a new drug, a study is undertaken (at great expense) of the ESS of the drug upon perturbation by multiple manufacturing, formulation, storage and transportation variables. The data acquired is then used to build a model that relates changes in ESS to manufacturing, formulation, storage and transportation variables. It is not common in the pharmaceutical industry, however, to relate changes in comprehensive ESS data sets to comprehensive measurements of changes in macromolecular form. We bridge the gap between physical measurements of a macromolecule's form and measurements of its long term stability, utilizing two data sets collected in a collaboration between our group at the University of Kansas and a group at the Ludwig Maximilians University in Munich, Germany. The long term stability data, collected by the team in Germany, contains measurements of the chemical and conformation stability of Granulocyte Colony Stimulating Factor (GCSF) over a period of two years in 16 different liquid formulations. The short term physical data, collected in our lab, is comprised of spectroscopic characterization of the response of GCSF to thermal unfolding. The same 16 liquid formulations of GCSF were used in each study, allowing us to fit models predicting the long term stability of GCSF from short term measurements. We first apply a novel data reduction method to the short term data. This method selects data in the neighborhood of thermal unfolding transitions, and automates traditional comparative analyses. We then model the long term stability measurements using a linear technique, least squares fits, and a nonlinear one, radial basis function networks (RBFN). Using a Pearson correlation coefficient permutation test, we find that many of the fitted results have less than a 1 percent probability of occurring by chance. | |
