Metabolomic Characterization of a Perfusion Bioprocess System

Biologics manufacturing typically uses engineered Chinese hamster ovary (CHO) cells to produce antibodies which requires considerable effort to maintain cells at optimum conditions. As metabolites provide the most direct readout of physiology, a quantitative understanding of cell metabolism can enable optimization of growth conditions. Mass spectrometry is the premier tool for metabolite measurement; however, transforming raw data into accurate quantitative measurement requires both expertise and extensive sample preparation. Here we demonstrate the ability of simple sample preparation using universal calibrators and a novel machine learning algorithm to rapidly provide biological insight.
Eight bioreactors expressing a monoclonal antibody were grown for twenty-five days using a perfusion process. Two media compositions, and two target cell densities were compared across the runs. The data was analyzed on an Thermo Scientific Orbitrap Exploris 120TM with polarity switching at Matterworks using a 6.5-minute HILLIC LC method. The raw files were uploaded to the PyxisTM application and absolute concentrations were returned in minutes.
PyxisTM reported absolute concentrations for 82 analytes from a range of annotated KEGG pathways and was able to distinguish between different medias fed to the perfusion reactors by looking at intracellular amino acid levels. Using the PyxisTM data and dimensionality reduction techniques, differences in the reactors were resolved. PyxisTM data showed a change in the reactor's metabolome over the 25-day run with the largest shifts at day 8 and day 15.
In addition to the large metabolomics data set, we tracked reactor performance and fully characterized the purified antibody with product quality methods. With this comprehensive data set we can identify markers of reactor performance and predict cultures which will maintain productivity and product quality.