A team at Glaxo Smith Kline, GSK, has carried out several studies using Raman spectroscopy and multivariate data analytics in order to monitor upstream cell culture processes. The studies show that Raman spectroscopy is a valuable analytical tool for enabling real-time monitoring and control of production bioreactors and that improved control can lead to improved product quality. Below is a summary of the studies.
Manufacturing of a biopharmaceutical, such as a mammalian or microbial cell culture process, consists of an upstream and a downstream process. The upstream process is basically the cell culture process in the production bioreactor, which is followed by the downstream process of cleaning and purification of the final product. The objective of the studies at GSK was to find out if Raman spectroscopy for real-time monitoring and control of production bioreactors could improve the quality of the monoclonal antibodies, mAb, that were produced.
Advantages of Process Analytical Technology (PAT) for bioreactor monitoring and control
Process Analytical Technology (PAT) tools such as Raman spectroscopy in combination with multivariate data analysis provides the capability to characterize and predict temporal profiles of various biomarkers such as amino acids, vitamins, glucose and lactate. PAT can also be used to control a cell culture process and provide understanding of cell metabolism, nutrient consumption and quality. The value of PAT tools, such as Raman spectroscopy and multivariate data analytics, is that it provides process monitoring at a very high frequency compared to offline measurements that are normally carried out only once a day. When a process can be controlled to this very fine detail it will impact safety, efficacy and robustness, and provides a way to enable Quality by Design, QbD.
Using Raman spectroscopy to control critical process parameters
Raman spectroscopy is based on inelastic scattering of photons from a monochromatic light source, usually from a laser source. When the photons interact with the molecules in a sample, the photons are scattered at a frequency higher or lower than the incident frequency. This is called the Raman effect and can be used to predict the quantities of certain kinds of molecules in the system. The GSK team used a Raman spectrometer from Kaiser (Kaiser Raman RXN2) in the studies.
The Raman spectrometer collects huge amounts of information in the form of spectra. The spectra are used to predict quantities of molecules such as glucose, lactate or impurities in the bioreactor. Multivariate data analytics is a prerequisite for this type of advanced analysis. The team at GSK used SIMCA from the Umetrics Suite of Data Analytics Solutions for multivariate data analysis model building and for PLS-based predictions and control built in the SIMCA-QPp prediction engine.
Feedback control loop for continuous feeding
The studies at GSK were designed to monitor and control the amount of glucose and lactate and were carried out in the following way: Raman spectra were collected from the bioreactor using in-situ probes. The offline measurements (collected once a day) were then matched to the spectrum that was collected closest in time to the offline measurements. The next step was to create calibrated models, where the spectral changes were correlated to the measured components using chemometric methods.
The calibrated models were then applied to future batches to generate process values. Process values, or predictions, were fed to a control system that compared the predictions to the target set point. If the prediction was below the setpoint it activated a pump that was linked to a stock solution of feed (glucose and lactate) that went into the bioreactor until it reached the target setpoint. This way, it was possible to control the process at a high frequency and detail.
Summary of the studies carried out at GSK
The team at GSK conducted several studies with different chemometric approaches that could be useful for the biopharmaceutical industry. Raman spectroscopy for multivariate data analytics was used to build PLS-based models for quantitative monitoring and control of glucose and lactate.
The team started with an initial study where they were able to create a very accurate and robust initial model with only a few batches. The conclusion was that for singular processes, implementation is quite straight forward and a few bathes can give the initial model decent accuracy.
In a second study, the aim was to control only glucose. In this study, the model was not initially calibrated with batches with very low glucose levels. The team found out that extrapolating did not provide correct predictions for low glucose levels. However, when the model was calibrated with batches with very low glucose levels, the new calibrated model became very accurate.
In yet another study, with a different molecule system, a glucose model was created where wavelength range truncation and latent variables were used to calibrate the model. The GSK team found out that truncation improved method accuracy but diminished robustness; although the model was very specific to glucose it was not robust regarding other changes in the process. When the team increased the number of latent variables, they could see improvement in robustness. A conclusion was that for biopharmaceutical processes it is critical to achieve a balance between accuracy and robustness.
The GSK team also tested a hypothesis, namely if feed of lactate could decrease the amount of ammonium in a cell culture process (mAb). Glycosylation plays a major role in the function and quality of a biopharmaceutical – there have been adverse clinical events because of suboptimal glycosylation. Ammonium is known to broadly affect glycosylation. The hypothesis was that controlling ammonium, by feeding lactate in a controlled manner (and thereby reduce the amount of ammonium accumulating in the bioreactor) could be a way to control glycosylation and hence product quality.
The team could demonstrate repeatedly that continuous feed of lactic acid to the cell culture resulted in a reduction of ammonium in comparison to bioreactors with no lactic acid feed. The GSK team could also see that the percent change in glycosylation was significantly increased in bioreactors that were fed lactic acid.
Conclusion: Raman spectroscopy in combination with MVDA is a valuable PAT tool for real-time monitoring and control of production bioreactors
A general conclusion of the studies was that Raman spectroscopy is a valuable PAT tool for enabling real-time monitoring and control of production bioreactors. It can lead to improved product quality and is projected to be integral for implementing future product control strategy at GSK. Another conclusion was that Raman spectroscopy can successfully control the levels of lactic acid in a bioreactor and that lactic feeds can drastically lower the final level of ammonium. The team also saw that lactic acid feeds alter the final glyco-profile of the mAb, which was in line with the hypothesis.
Want to know more?
This webinar from Kaiser provides more in-depth information about the studies carried out at GSK.
The Umetrics Suite of Data Analytics Solutions can also be used for other spectroscopic methods. Learn more here.
Instrument vendors: Find out more about our Partner Program, using SIMCA and how to simplify the embedding process of our multivariate technology (SIMCA-Q)