Use of Peptide Microarrays for Fast and Informative Profiling of Therapeutic Antibody Formulation Conditions

Methods to optimize the solution behavior of therapeutic proteins are frequently time-consuming, provide limited information, and often use milligram quantities of material. Here, we present a simple, versatile method that provides valuable information to guide the identification and comparison of formulation conditions for, in principle, any biopharmaceutical drug. The subject protein is incubated with a designed synthetic peptide microarray; the extent of binding to each peptide is dependent on the solution conditions. The array is washed, and the adhesion of the subject protein is detected using a secondary antibody. We exemplify the method using a well-characterized human single-chain Fv and a selection of human monoclonal antibodies. Correlations of peptide adhesion profiles can be used to establish quantitative relationships between different solution conditions, allowing subgrouping into dendrograms. Multidimensional reduction methods, such as t-distributed stochastic neighbor embedding, can be applied to compare how different monoclonals vary in their adhesion properties under different solution conditions. Finally, we screened peptide binding profiles using a selection of monoclonal antibodies for which a range of biophysical measurements were available under specified buffer conditions. We used a neural network method to train the data against aggregation temperature, k_D, percentage recovery after incubation at 25 °C, and melting temperature. The results demonstrate that peptide binding profiles can indeed be effectively trained on these indicators of protein stability and self-association in solution. The method opens up multiple possibilities for the application of machine learning methods in therapeutic protein formulation.