초록
<P><B>Abstract</B></P> <P>The use of Raman in Bioprocess development have shown great potential for process understanding and monitoring although there are still some challenges and limitations in performance when conditions such as clone, media or scale are changed during bioprocess development. This study proposes different strategies to balance the different information content of multiple mammalian cell cultivations produced during a bioprocess development program, when several conditions are investigated. The result is a pragmatic approach to PAT monitoring that serves both development and manufacturing stages. Combining risk-assessment techniques with two ways of developing monitoring calibrations (local or general models), we were able to obtain good predictive power from Raman spectroscopy used as PAT tool, when multiple cultivation conditions vary. As an example, the effects of process scale, base powder media and cell-line on Raman spectra are discussed and how using local models specific to some of these cultivation conditions, has a positive impact on calibration performance. It is shown how more accurate calibrations can be obtained using Clone-based local models, which requires less batches than usual approaches (up to 3–9). This study uses thirty-five cultivations of four different types of CHO cell lines, eight different clones, and four different scales – 2 L, 7 L, 15 L and 10,000 L – in two Cultivation Modes – fed-batch and perfusion. The aim is to serve as blueprint to how can PAT approaches be best developed in parallel to bioprocess development.</P> <P><B>Highlights</B></P> <P> <UL> <LI> In mammalian cell cultures, Raman Spect. faces challenges in performance when conditions are changed during development. </LI> <LI> To develop calibrations, RA assays and local model strategies shown to increase the predictive power when conditions vary. </LI> <LI> Process scale, base powder media and cell-line on Raman spectra are conditions with positive impact on model performance. </LI> <LI> Clone-based models produce more accurate calibrations with lower number of batches (up to 3-9), unlike usual approaches. </LI> </UL> </P> <P><B>Graphical abstract</B></P> <P>[DISPLAY OMISSION]</P>