Raman‐based dynamic feeding strategies using real‐time glucose concentration monitoring system during adalimumab producing CHO cell cultivation

Abstract

The use of Process Analytical Technology tools coupled with chemometrics has been shown great potential for better understanding and control of mammalian cell cultivations through real-time process monitoring. In-line Raman spectroscopy was utilized to determine the glucose concentration of the complex bioreactor culture medium ensuring real-time information for our process control system. This work demonstrates a simple and fast method to achieve a robust partial least squares calibration model under laboratory conditions in an early phase of the development utilizing shake flask and bioreactor cultures. Two types of dynamic feeding strategies were accomplished where the multi-component feed medium additions were controlled manually and automatically based on the Raman monitored glucose concentration. The impact of these dynamic feedings was also investigated and compared to the traditional bolus feeding strategy on cellular metabolism, cell growth, productivity, and binding activity of the antibody product. Both manual and automated dynamic feeding strategies were successfully applied to maintain the glucose concentration within a narrower and lower concentration range. Thus, besides glucose, the glutamate was also limited at low level leading to reduced production of inhibitory metabolites, such as lactate and ammonia. Consequently, these feeding control strategies enabled to provide beneficial cultivation environment for the cells. In both experiments, higher cell growth and prolonged viable cell cultivation were achieved which in turn led to increased antibody product concentration compared to the reference bolus feeding cultivation.

Real‐time amino acid and glucose monitoring system for the automatic control of nutrient feeding in CHO cell culture using Raman spectroscopy

“…However, thereafter, lactate level was accumulated to high levels (> 100 mM) in both experiments (except between cultivation Day 7 and 8 in the amino acid‐based dynamic fed experiment, where lactate dropped from 120 to 80 mM), although the available glucose level was limited at low level (similarly to our previous study). [ 33 ] Unfortunately, the mechanisms that control metabolic shift from lactate production to consumption are influenced by several process parameters, therefore further investigations are required. [ 37 ]…”

“…All generated data was imported into Matlab Version R2017A (The MathWorks), where PLS toolbox 8.2 (Eigenvector Research) was used to build models in a similar fashion to our previously reported efforts. [ 33 ]…”

“…The initial calibration dataset for glucose model was gathered from collected from nine shake flasks (56 points) and three bolus fed bioreactor cultures (50 points). [ 33 ] However, it was experienced that the prediction power (the error of prediction was 5.2 mM) of the model could be further improved by expanding the dataset with new reference measurements from a lower concentration range. Therefore, the dataset was merged with samples collected from seven new independent dynamic fed‐batch bioreactor experiments, where the GLC was controlled at a level of 11 mM using dynamic feeding strategies, which resulted in a total of 204 samples in the final calibration set.…”

“…All generated data was imported into Matlab Version R2017A (The MathWorks), where PLS toolbox 8.2 (Eigenvector Research) was used to build models in a similar fashion to our previously reported efforts. [33] Different pre-processing methods were assessed for the raw Raman spectra (200-1890 cm −1 ) to evaluate the best combination leading to maximum signal-to-noise ratio prior to analysis. First, Automatic Whittaker Filter (AWF) baseline correction (λ = 10000; p = 0.001) was applied to eliminate the baseline shift caused by the fluorescence background.…”

“…Therefore, Raman spectroscopy‐based Partial Least Squares (PLS) calibration models had to be developed and applied to predict the concentration of nutrients and metabolites in real‐time, providing this way an opportunity to increase the information about the metabolic processes of the cells. In our previous study, [ 33 ] we demonstrated that dynamic feeding control based on Raman‐monitored glucose concentration can be successfully used to establish beneficial cultivation environment for the cells. However, this feeding strategy needed to be further improved by taking into consideration that nutrients, added by multi‐component feed medium, are consumed by the cells at a different rate than glucose.…”