Raman spectroscopy as a multi attribute PAT tool for real time monitoring of Critical Quality Attributes in Biologics

With increasing focus on implementation of continuous processing for manufacturing of biopharmaceutical products, there is an urgent requirement for tools that enable inline process monitoring. At present, significant time and resources are invested on monitoring CQAs during in-process stages of downstream manufacturing. In this study, we demonstrate an application of Raman spectroscopy as a multi-attribute PAT (process analytical technology) tool for estimation of protein concentration as well as size/charge based heterogeneities during purification of a monoclonal antibody biotherapeutic product. Pre-processing using standard normal variate (SNV) and Salvitzky – Golay smoothening was able to reduce the fluorescence interference and offset in the data set, that are known challenges while working with Raman spectra. Advanced chemometrics has been applied to the Raman data for measurement of other attributes - the first based on a Bayesian optimised support vector machines (SVM) methodology using the Gaussian kernel and the second on using random forest with feature selection by K-means clustering. Both approaches offered significant improvement over the conventional PLS, especially in aggregate prediction. The tool was successfully validated at all in-process stages with four IgGs for aggregate prediction with R² of 0.93 and RMSE of 0.37 under a broad calibration range of 0-7% high molecular weight species. Other attributes such as charge variants (acidic%/basic%) and protein concentration were also predicted successfully and validated for both trained and untrained IgG molecules. We hope that this study will encourage manufacturers for using Raman as a PAT tool for multi-attribute measurement.