Bioreactor Integration and PAT for Real-Time Control of Upstream Bioprocesses

PAT-enabled scaling and optimization of upstream bioprocesses Process Raman spectroscopy and bioreactor application note compendium Compendium Introduction Scaling up a bioprocess is a complex, multi-variable challenge. Designing processes with scalable parameters is essential for ensuring confidence when transitioning from the research laboratory to clinical and production operations. The application notes in this compendium show how modern bioreactors and process Raman technology can enhance bioprocess development and enable advanced process control strategies from pilot to production scale. Raman spectroscopy as a PAT tool in a continuous perfusion run 5 Real-time metabolite monitoring with process Raman 14 Monitor and maintain glucose concentrations in real-time 17 Table of Contents Additional upstream processing resources Streamlining seed train scale-up through utilization of the turndown ratio of a single use bioreactor 23 Part 1: Intuitive bioprocess scale-up from bench scale to pilot scale 29 Demonstration of scalability for multiple cell lines 36 Chemometric model transferability across process Raman instruments 43 Measure key variables with process Raman for enhanced control 47 Product solutions 51 PAT-enabled advanced process control for cell culture PAT-enabled advanced process control for cell culture Using Raman Spectroscopy as a Process Analytical Technology Tool in a 50-Day Continuous Perfusion Run Application note Authors Mike Bates¹, Kevin Broadbelt¹, Jan Ott², Vivian Ott², Regine Eibl², Lin Chen¹, David Kuntz¹, Lin Zhang¹, Mathew Zustiak¹, and Sue Woods¹ (Thermo Fisher Scientific)1 School of Life Sciences and Facility Management, Institute of Chemistry and Biotechnology, ZHAW Zurich University of Applied Sciences², 8820 Wädenswil, Switzerland Highlights • Implementation of Process Raman as a PAT tool for in-line and real-time monitoring of critical process parameters, including titer, cell densities, and nutrient concentrations in mAb-producing, high-cell-density perfusion CHO cell culture, is discussed. • The Thermo Scientific™ MarqMetrix™ All-In-One Process Raman Analyzer with Thermo Scientific™ Lykos™ PAT Software, with 21 CFR Part 11 compliance, is used for in-line, real-time, calibration-free monitoring of a bioreactor during a 50-day continuous perfusion run. Summary The biopharmaceutical industry is driven by the need to increase production and reduce costs while maintaining product quality. One effective way to achieve this goal is to streamline the monitoring process of biologics production, thus allowing more effective control of production parameters. In this application note, we introduce a case study using Process Raman for in-line, real-time monitoring of specific critical process parameters (CPPs) in high-density mammalian perfusion cell cultures reaching 100–130 million cells mL-1. The Thermo Scientific™ MarqMetrix™ All-In-One Process Raman Analyzer System is shown to enable in-line measurements of CPPs, including glucose, lactate, ammonium, product titer, and cell viability, over the course of a 50-day continuous perfusion bioreactor run. Background In recent years, Raman spectroscopy has gained popularity as a process analytical technology (PAT) tool that enables real-time monitoring and control of critical bioprocessing parameters that are key to the successful production of therapeutic drugs. The product portfolio of biologics is broadening, and implementation of different spectroscopic PAT tools can address the limitations of traditional off-line analytical methods for such products. The production of biotherapeutic products requires high process efficiency while ensuring product quality and minimized manufacturing costs¹. Implementing PAT tools in biopharmaceutical manufacturing is a critical priority identified by the FDA, with the goal of allowing rapid development and access to novel therapeutics and existing medications without compromising product quality², ³. Cell culture processes are labor intensive due to the frequent sample analyses required. When operating bioreactors in fed-batch mode, nutrients are periodically supplemented via a bolus feeding strategy using predetermined volumes of concentrated nutrients. While batch production is a well-vetted manufacturing method, this production strategy can be slow and inefficient. Recent advances in biomanufacturing processes involve continuous processing. Continuous bioprocessing technologies, whether upstream or downstream, can increase speed while decreasing the costs of producing these essential biologics⁵. Perfusion cell culture is a process that uses filters to keep cells in a bioreactor while continuously exchanging culture medium. Fresh medium replenishes nutrients and carbon sources, while cellular waste and medium depleted nutrients are removed. Key advantages of bioreactors operated in perfusion mode include flexibility, lower cost, improved quality, and greater speed. In this study, we focused on the continuous perfusion operation mode for suspension cell culture. The goal of continuous perfusion is to develop a process that maintains a steady state in which productivity and product quality can be sustained long-term with minimal variability, with bioreactors running for 30 – 90 days. Raman spectroscopy is a laser‐based method for generating a chemical fingerprint of a sample², ³. A key advantage of Raman spectroscopy is its ability to measure numerous analytes in a non‐destructive manner, in situ, and with low interference from water. Fortuitously, numerous analytes with distinct Raman fingerprints enable monitoring of CPPs such as nutrient feed levels, cell metabolites, cell growth profiles, product levels, and product quality attributes¹, ². The MarqMetrix All-In-One Process Raman Analyzer with Thermo Scientific Lykos PAT Software is designed to offer accurate, reliable, real-time identification and quantification of numerous CPP analytes. Process Raman is extremely advantageous when adopted into a continuous perfusion culture system. Utilizing an immersed Raman optical sensor provides real-time process information about the CPPs, unlike traditional monitoring systems, which require manual sampling and off-line analysis¹, ². Noted advantages include an increase in data acquisition frequency, the opportunity for rapid correction of any detected process parameter deviations, and reduction of contamination risk due to a reduced offline sampling. This application note describes the integration of the MarqMetrix All-In-One Process Raman Analyzer with the 50L Thermo Scientific™ DynaDrive™ bioreactor to perform in-line measurements of CPPs in a continuous perfusion run (See Figures 1-3). Here, we highlight the integration of the MarqMetrix All-In-One Process Raman Analyzer System to perform in-line measurements of glucose, lactate, ammonium, viable cell density (VCD), total cell density (TCD), and titer. This PAT tool provides accurate prediction models for several parameters and metabolites and shows high correlations with offline measurements. The simultaneous measurements of metabolites, product titer, and protein concentration allow for real-time process control, demonstrating the effectiveness of process Raman spectroscopy as a PAT tool in biopharmaceutical industries. Materials and methods Cell line and medium For the two 50 L perfusion cultivations, a trastuzumabproducing CHO K1 cell line was used. The basal medium was 0.66x concentrated High-Intensity Perfusion CHO medium (Gibco™), whereas 1x concentrated medium was used as the feed medium. Both media were supplemented with 4 mmol L-1 l-glutamine (Gibco™) and 1% Anti-Clumping Agent (Gibco™). Inoculum production Inoculum production was carried out over a period of 10 days. The first three passages were carried out at a shake flask scale. A 10 L wave-mixed bioreactor (Cellbag 10 L, Cytiva) with a working volume of 5 L was used for the last passage. Perfusion cultivations at 50 L scale For the 50 L perfusion cultivations, the ATF version of the DynaDrive S.U.B. (Thermo Scientific) with the corresponding G3Pro Bioprocess Controller was used. Cell retention was realized with Repligen’s XCell ATF6 single-use version and the corresponding XCell LS controller at an average ATF flow rate of 17 L min-1. The cultivation was started with a VCD of 3x105 cells mL-1. Perfusion was started after an initial 3-day batch phase. The perfusion rate was limited to 1 d-1. The bleed was controlled with the permittivity probe IncyteArc (Hamilton Bonaduz) to keep a viable cell volume of 100–130 mm3 mL-1 constant. To keep the bioreactor volumes constant at 50 L, the harvest pump was controlled by the bioreactor weight. Glucose concentration was either kept constant at 2 g L-1 with a CIT Sens Bio glucose sensor (C-CIT) for the first run or was manually controlled for the second run. To prevent foaming, a 1:10 diluted antifoam C emulsion (Sigma Aldrich) was added quasi-continuously. In all cultivations, the pH was controlled to a value of ≤7.15 by adding CO₂ via the drilled hole sparger. DO was controlled at 40% by adding N₂, air and O₂ using the corresponding sparger. An overlay gassing rate of 0.05 vessel volumes min-1 with air was chosen. The stirrer speed was set to 136.5 rpm (corresponding to 40 W m-3) in the DynaDrive. Analytics Once a day, cell-specific parameters, such as the VCD, viability, and cell diameter, were determined with a Cedex HiRes analyzer (Roche Diagnostics, Basel, Switzerland). The Cedex Bio analyzer (Roche Diagnostics) was used to check the substrate and metabolite concentrations of glucose, lactate, glutamine, and ammonium as well as the IgG titer. Figure 2. 50L Thermo Scientific DynaDrive S.U.B. for cell culture applications. Figure 1. Scheme of perfusion process configuration and integration of an in-line Raman sensor (MarqMetrix All-In-One Process Raman Analyzer with Lykos PAT Software). High-Intensity Perfusion CHO Medium (Gibco) 50L Thermo Scientific DynaDrive S.U.B. bioreactor XCell LS Lab controller Thermo Scientific™ MarqMetrix™ All-In-One Process Raman Analyzer with Thermo Scientific™ Lykos PAT Software Fresh Media Spent Media Additionally, an augmentation approach was used to improve the models. Data from two ZHAW perfusion runs was collected. One run was used to augment training data to improve the prediction accuracy for the other. The run used to augment the training data was weighted equally to the rest of the training data despite having far fewer samples because this data was assumed to have greater similarities to the prediction run. In this manner, the model was able to learn from the bulk data in the regular training set but was fine-tuned specifically to give the best predictions on ZHAW data. This approach optimizes the benefits of using both a broad general dataset and a specifically targeted but much smaller dataset. Results To analyze the chemometric modeling results, let us first define the key figures of merit: bias and RMSEP. Bias is the average difference between predicted values and reference values. The Root Mean Squared Error of Prediction (RMSEP) is the combined error of bias and precision, where precision is the randomness (noise) around the mean of the predicted values, assuming there is no bias. Furthermore, the quality of a chemometric model is typically evaluated using the Q-residuals and Hotelling’s T-squared values. The Q-residuals are used to quantify how well the model fits the raw data. Q-residuals should typically be less than 1, which indicates that the model accounts for all the variance in the spectra. A high Q-residual value indicates an observation not well explained by the model, suggesting it may be an outlier. Hotelling’s T-squared values measure the distance of each observation from the model’s center in the space of the retained components. A high T-squared value expresses an observation far from the model’s center, which could also suggest it as a potential outlier. Both Q-residuals and Hotelling’s T-squared values are important tools for model diagnostics in chemometrics. While the results of applying generalized chemometric models, based on bolus-fed CHO cell lines, produced prediction errors of approximately 1 g L-1 for glucose and lactate, these generalized models produced poor results when predicting ammonium, titer, VCD and TCD. In contrast, the augmentation of the generalized models using one ZHAW run to predict the other ZHAW run resulted in prediction errors of 0.36 g L-1 for glucose and 0.37 g L-1 for lactate. Furthermore, the augmentation of the generalized model enabled accurate predictions for ammonium (RMSEP = 0.95 mmol L-1) and titer (RMSEP = 0.36 g L-1) while also providing prediction accuracy for VCD & TCD of ≈10 million cells mL-1, which translates to +/- 10% of the stationary phase concentrations of ≈100 million cells mL-1. Thermo Scientific MarqMetrix All-In-One Process Raman Analyzer Measurements Measurements were performed using the MarqMetrix All-In-One Process Raman Analyzer System, with the optical Bioreactor ball probe directly immersed in the bioreactor (50L). Each Raman spectra resulted from an average of 60 measurements, with an integration/exposure time of 1 sec and a laser power setting of 450 mW. The total acquisition time per data spectra was 2 minutes, with a timestamp matched between the MarqMetrix All-In-One Process Raman Analyzer and offline instrument analysis to build the model. Chemometric Model building Independent data from multiple MarqMetrix All-In-One Process Raman Analyzers and numerous bioreactors were used to create models for each analyte. The training datasets were collected from 12–24 samples per bioreactor to create each chemometric model. In-line and at-line measurements were aligned using timestamps between the MarqMetrix All-In-One Process Raman Analyzer and the at-line instrument, the Nova Flex II. All data was reviewed before building the models. In addition, an algorithm was implemented to remove data spikes in the spectra caused by cosmic rays. The spectral region of interest was selected, and multiple spectra were averaged to increase signal-to-noise ratios such that each measurement corresponded to a ten-minute read-time. The spectra were pre-processed to remove differences in the baseline due to fluorescence and other effects. Spectra were also normalized to remove differences in absolute intensity between various bioreactor types. Partial Least Squares (PLS) models were created for each analyte of interest, and leave-out-one-run cross-validation was performed to test the optimization of each model. Analytes of interest include glucose, lactate, ammonium, VCD, TCD, and titer. Figure 3. MarqMetrix All-In-One Process Raman Analyzer with optical MarqMetrix Bioreactor BallProbe™ immersed in a 5L glass bioreactor. Residual # Occurrences –0.5 0.0 0.5 12.0 D. Residuals Reference Predicted 0.0 2.5 5.0 7.5 10.0 12.5 C. Gluc (g/L) Predicted vs. Reference Panel C in each figure demonstrates the linearity of the CHO-cell training data, augmented with a weighted ZHAW run (green) and a ZHAW run used as a prediction data set (purple). One reason for applying the weighted-ZHAW augmentation to the independent training data set was the very small linear range for many analytes in these perfusion runs. In contrast to the small linear range of CPPs such as glucose, lactate, ammonium, and titer, the cell density values, VCD and TCD, exhibited a wide range from 20–130 million cells mL-1. The results demonstrate excellent correlations between the model prediction data obtained from the Thermo Scientific MarqMetrix All-In-One Process Raman Analyzer and the offline data for numerous CPPs (Table 1). Figures 4-9 show the correlation between offline data and predictions made by applying the chemometric models to the spectra collected using the Thermo Scientific MarqMetrix All-In-One Process Raman Analyzer. In each Figure, the predicted vs. offline data is plotted in panel A. These figures demonstrate the ability to monitor changes in various CPPs in real-time with a high degree of accuracy. Panel B in each figure speaks to the quality of the chemometric model, as the Q-residuals and Hotelling’s T-squared values for each analyte model exhibit very low values; this indicates a good fit of the model with negligible unaccounted variance or outliers. Figure 4. Modeling results for glucose monitoring in the 50-day perfusion run of the bioreactor. (A) shows the modeling results generated from the process Raman spectra compared to the results of offline analysis. (B) shows the Q-residuals vs. Hotelling’s T² values. (C) shows the linearity of the predicted vs. offline values. (D) shows the histogram of residuals between predicted and offline values. 0.0 0.2 0.4 0.6 0.8 1.0 T² A. Glucose – Predicted vs. Offline Q 0.0 0.5 1.0 0.0 0.5 Glucose (g L-1) 2023-07-15 2023-07-22 2023-08-01 2023-08-08 2023-08-15 2023-08-22 1.0 1.5 2.0 2.5 B. Gluc Q vs. T² 0.0 2.5 5.0 7.5 10.0 12.5 2.5 5.0 7.5 10.0 12.5 0.0 Model Timeseries Offline Model ZHAW 1=1 Predicted Offline A. Ammonium – Predicted vs. Offline A. Lactate – Predicted vs. Offline Figure 6. Modeling results for ammonium monitoring in the bioreactor’s 50-day perfusion run. (A) shows the modeling results generated from the process Raman spectra compared to the results of offline analysis. (B) shows the Q-residuals vs. Hotelling’s T² values. (C) shows the linearity of the predicted vs. offline values. (D) shows the histogram of residuals between predicted and offline values. Figure 5. Modeling results for lactate monitoring in the bioreactor’s 50-day perfusion run. (A) shows the modeling results generated from the process Raman spectra compared to the results of offline analysis. (B) shows the Q-residuals vs. Hotelling’s T² values. (C) shows the linearity of the predicted vs. offline values. (D) shows the histogram of residuals between predicted and offline values. Residual # Occurrences –0.6 –0.4 –0.2 0.0 D. Residuals Reference Predicted 0 1 2 3 C. Lac (g/L) Predicted vs. Reference 0.0 0.2 0.4 0.6 0.8 1.0 T² Q 0.0 1.0 2.0 0.5 Lactate (g L-1) 2023-07-15 2023-07-22 2023-08-01 2023-08-08 2023-08-15 2023-08-22 1.0 1.5 B. Lac Q vs. T² 0 1 2 3 2 4 6 8 10 0 Model Timeseries Offline Model ZHAW 1=1 Predicted Offline 0.0 0.5 1.5 0.2 Residual # Occurrences 0 1 D. Residuals Reference Predicted 0 5 10 15 20 25 C. NH₄+ (mmol/L) Predicted vs. Reference 0 4 6 8 10 12 T² Q 0.0 1.0 2.0 1 NH₄+ (mmol L-1) 2023-07-15 2023-07-22 2023-08-01 2023-08-08 2023-08-15 2023-08-22 2 3 B. NH₄+ Q vs. T² 0 10 20 25 2.5 5.0 7.5 10.0 12.5 0.0 0 0.5 1.5 2 Model Timeseries Offline Model ZHAW 1=1 Predicted Offline 4 2 5 15 Figure 8. Modeling results for Total Cell Density (TCD) monitoring in the bioreactor’s 50-day perfusion run. (A) shows the modeling results generated from the process Raman spectra compared to the results of offline analysis. (B) shows the Q-residuals vs. Hotelling’s T² values. (C) shows the linearity of the predicted vs. offline values. (D) shows the histogram of residuals between predicted and offline values. Figure 7. Modeling results for Viable Cell Density (VCD) monitoring in the 50-day perfusion run of the bioreactor. (A) shows the modeling results generated from the process Raman spectra compared to the results of offline analysis. (B) shows the Q-residuals vs. Hotelling’s T² values. (C) shows the linearity of the predicted vs. offline values. (D) shows the histogram of residuals between predicted and offline values. A. TCD – Predicted vs. Offline A. VCD – Predicted vs. Offline Residual # Occurrences 0 20 D. Residuals Reference Predicted 0 50 100 C. VCD Predicted vs. Reference 0 1 2 3 T² Q 0.0 1.0 2.0 40 VCD (10⁶ cells mL-1) 2023-07-15 2023-07-22 2023-08-01 2023-08-08 2023-08-15 2023-08-22 60 80 B. VCD Q vs. T² –25 25 75 125 2 4 6 10 12 0 20 0.5 1.5 Residual # Occurrences 0 D. Residuals Reference Predicted 0 50 100 C. TCD Predicted vs. Reference 0 1 2 3 T² Q 0.0 1.0 2.0 40 TCD (10⁶ cells mL-1) 2023-07-15 2023-07-22 2023-08-01 2023-08-08 2023-08-15 2023-08-22 60 80 B. TCD Q vs. T² 0 50 100 125 5.0 7.5 10.0 12.5 15.00 0.0 20 0.5 1.5 20 100 25 75 100 Predicted Offline Model Timeseries Offline Model ZHAW 1=1 0 50 100 8 Model Timeseries Offline Model ZHAW 1=1 Predicted Offline 2.5 -20 Analyte Units RMSEP Bias Glucose g L-1 0.36 -0.15 Lactate g L-1 0.37 0.32 Ammonium mmol L-1 0.95 -0.71 VCD million cells mL-1 10.78 0.47 TCD million cells mL-1 10.59 0.62 Titer g L-1 0.36 -0.27 Table 1. Summary of Chemometric Modeling Results for this 50-day Perfusion Run. Figure 9. Modeling results for titer monitoring in the 50-day perfusion run of the bioreactor. (A) shows the modeling results generated from the Process Raman spectra compared to the results of offline analysis. (B) shows the Q-residuals vs. Hotelling’s T² values. (C) shows the linearity of the predicted vs. offline values. (D) shows the histogram of residuals between predicted and offline values. A. Titer – Predicted vs. Offline Residual # Occurrences 0.00 0.50 D. Residuals Reference Predicted 0 2 4 C. Titer Predicted vs. Reference 0.0 0.4 0.6 1.0 T² Q 0.0 1.0 2.0 0.5 IgG Titer (g L-1) 2023-07-15 2023-07-22 2023-08-01 2023-08-08 2023-08-15 2023-08-22 1.0 1.5 B. Titer Q vs. T² –1 1 3 5 2 4 6 10 12 0 0.0 0.5 1.5 2.0 0 2 4 8 Conclusion The study in this application note highlights an approach to using Raman spectroscopy, specifically Process Raman, for in-line, real-time monitoring of a high cell density mammalian cell culture run in a bioreactor operated in perfusion mode The MarqMetrix All-In-One Process Raman Analyzer provides exceptional data quality, which, when combined with multivariate PLS modeling, enables in-line, real-time monitoring of CPPs glucose, lactate, ammonium, titer, and cell density measurements. The ability to obtain accurate cell density measurements in a wide range from 20-130 million cells mL-1 showcases the robustness and utility of these chemometric models and the efficacy of utilizing the MarqMetrix All-In-One Process Raman Analyzer to collect high-quality spectral data. Great data lead to robust and accurate chemometric models, thereby unlocking the potential of Process Raman as a critical PAT tool in biopharmaceutical processes. The simultaneous measurements of metabolites and product titer are of special interest as they will allow for greater process control of cultivation and purification parameters within continuous biomanufacturing processes. Additionally, Process Raman greatly enhances the operator’s understanding of the CHO cell perfusion process as these CPPs are measured repeatedly in short intervals in a non-destructive manner. This study demonstrates the capability of Process Raman spectroscopy as a PAT tool that can pair seamlessly with automation systems to improve yields and enhance product quality. The results of the perfusion cultivations discussed in this application note have been published in detail, https://www.mdpi.com/2227-9717/12/4/806.⁶ Predicted Offline Model Timeseries Offline Model ZHAW 1=1 0.2 0.8 –0.25 0.25 0.75 References 1. Research, C. for D. E. and. PAT — A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance. U.S. Food and Drug Administration. https://www.fda.gov/regulatory-information/ search-fda-guidance-documents/pat-framework-innovative-pharmaceuticaldevelopment-manufacturing-and-quality-assurance (accessed 2023-03-21). 2. Real time monitoring of multiple parameters in mammalian cell culture bioreactors using an in-line Raman spectroscopy probe. Nicholas R Abu-Absi ¹, Brian M Kenty, Maryann Ehly Cuellar, Michael C Borys, Sivakesava Sakhamuri, David J Strachan, Michael C Hausladen, Zheng Jian Li. Biotechnol Bioeng. 2011 May;108(5):1215-21. doi: 10.1002/bit.23023. Epub 2010 Dec 22 3. Quick generation of Raman spectroscopy based in-process glucose control to influence biopharmaceutical protein product quality during mammalian cell culture. Brandon N Berry ¹, Terrence M Dobrowsky ¹, Rebecca C Timson ¹, Rashmi Kshirsagar ¹, Thomas Ryll ¹, Kelly Wiltberger ² Biotechnol Prog 2016 Jan-Feb;32(1):224-34. doi: 10.1002/btpr.2205. Epub 2015 Dec 21. 4. Modernizing the Way Drugs Are Made: A Transition to Continuous Manufacturing | FDA, Sau (Larry) Lee, Ph.D., Deputy Director of the Office of Testing and Research, and Chair of the Emerging Technology Team, Office of Pharmaceutical Quality, CDER 5. Integrated Continuous Pharmaceutical Technologies – A Review. András Domokos, Brigitta Nagy, Botond Szilágyi, György Marosi, and Zsombor Kristóf Nagy. Organic Process Research & Development 2021 25 (4), 721-739 DOI: 10.1021/acs. oprd.0c00504 6. Scaling Fed-Batch and Perfusion Antibody Production Processes in Geometrically Dissimilar Stirred Bioreactors. Vivian Ott, Jan Ott, Dieter Eibl, Regine Eibl. Processes. 2024, 12(4), 806. Learn more at thermofisher.com/marqmetrixAIO For research use only. Not for use in diagnostic procedures. For current certifications, visit thermofisher.com/certifications © 2024 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. MCS-AN1068-EN 5/24 Real time metabolite monitoring using the MarqMetrix All-In-One Process Raman Analyzer and the 500L Dynadrive Single-Use Bioreactor (S.U.B.) Application note Authors Juan Villa, Matthew Zustiak, Elizabeth Amoako, David Kuntz, Lin Zhang, Kevin Broadbelt and Sue Woods (Thermo Fisher Scientific) Background The capability of Raman spectroscopy to reflect small changes in complex aqueous systems has expanded the application of this technique to analyze biopharmaceutical processes such as cell growth in bioreactors. When the Raman spectrometer is utilized as a continuous process analyzer of these complicated chemical systems, this technology can monitor biopharmaceutical production processes real-time, in-situ and non-destructively. The ability of Raman Spectroscopy to detect changes of numerous metabolites during bioreactor processes has elevated this technology to a robust Process Analytical tool. Thermo Scientific™ MarqMetrix™ Single-Use Bioreactor BallProbe™ Sampling Optic with TouchRaman™ immersion technology Reusable Raman Spectroscopy analysis optical probes offer benefits such as improved process repeatability and reliability by reducing run-to-run variability. With the MarqMetrix All-In-One Process Raman Analyzer, there are a wide range of probes available. The MarqMetrix Single-Use Bioreactor BallProbe Sampling Optic is designed to meet the requirements of the bioprocess industry and can be used with the MarqMetrix All-In-One Process Raman Analyzer. These probes are quick and easy to swap and connect, are durable and can handle sterility practices including offline autoclaving. Thermo Scientific™ DynaDrive™ Single-Use Bioreactor (S.U.B.), for perfusion cell culture applications The DynaDrive Single-Use Bioreactor (S.U.B.), the latest advancement in S.U.B. technology, offers improved performance and scalability for large volume bioproduction. The cuboid-shaped tank offers several key advantages over legacy S.U.B. designs including superior mixing and mass transfer capabilities as well as improved scalability. This application note describes the integration of the MarqMetrix All-In-One Process Raman Analyzer with the 500L DynaDrive Single-Use bioreactor to perform in-line measurements of critical process parameters (CPPs). Utilizing continuously generated spectral data throughout a cell growth culture run, accurate prediction models for several parameters and metabolites were developed using this integrated system. Process Raman analysis Materials and methods Cell culture and feeding strategy Cell culture was performed in a 500L DynaDrive S.U.B, containing a working volume of approximately 320L of cell culture medium, and inoculated with 0.5*106 cells/mL at a temperature of 36.5 °C, pH= 6.9+/- 0.3, DO = 50%). The pH level was controlled by CO2 gassing and sodium carbonate additions, as needed. The cells were grown in a chemically defined medium and fed daily with a two-step feeding process, starting at day 3. The first feed media was added at 4% by weight based on the starting volume and the second feed media was added at 0.4%. The temperature was shifted to 33 °C on day 6. The run terminated after 14 days. The bioreactor was covered to protect from stray light. After autoclaving, the MarqMetrix Single-Use Bioreactor BallProbe Sampling Optic was inserted into the DynaDrive S.U.B. during the run for in-line, real-time spectral Raman data generation. Many pre-processing techniques were tested, including the Savitzky Golay filter with derivatives, Automatic Whitaker Smoothing, Extended Multiplicative Scatter Correction, SNV, and mean centering. The best pre-processing techniques used varied, based on which specific parameter of interest was modelled. Partial Least Squares (PLS) models were created for each property of interest and cross-validation was performed to test the optimization of each model. Properties of interest included glucose, lactate, glutamine, glutamate, TCD, VCD, and other common metabolites generated during the bioreactor culture run. Results In this work, continuous in-line Raman spectroscopy was applied to a fed-batch CHO cell culture process. The in-line spectral data was correlated to the offline analytical data acquired for parameters of interest. The use of Raman spectroscopy to monitor process parameters first requires chemometric model building with an externally calibrated data set (independent offline data). To assess the accuracy of the MarqMetrix All-In-One Process Raman Analyzer predicted values, bioreactor samples were collected daily and analyzed for comparison. The root mean square error of calibration (RMSEC), root mean square error of cross validation (RMSECV) and root mean square error of prediction were calculated for each parameter (RMSEP). The error was averaged based upon the prediction of the model to identify the RMSECV which is used to construct the model. The RMSEP is used to test the model against “new” data that the model has not seen. The coefficient of variation, R², was recorded for each PLS model. The value is used to determine the amount of variation of the Y variable which the model predictors (X variables) can explain. It is important to note that the combined use of several large, independent data sets from bioreactor runs of the same CHO culture process produced predictive chemometric models that are more accurate and robust. For this study, five independent datasets from previous bioreactor runs were combined to train a large chemometric model. The calibration model was then applied to the spectral data obtained during this DynaDrive S.U.B run. The data indicates that the model was able to accurately predict this new dataset, and that model predictions were highly correlated with data measurements collected offline for numerous metabolites as shown in Table 1. Figure 2. Thermo Scientific MarqMetrix All-In-One Process Raman Analyzer. Figure 1. 500L Thermo Scientific DynaDrive S.U.B. for cell culture applications. MarqMetrix All-In-One Process Raman Analyzer measurements Measurements were performed using the MarqMetrix All-InOne Process Raman Analyzer, with the MarqMetrix Single-Use Bioreactor BallProbe Sampling Optic of the MarqMetrix All-In-One Process Raman Analyzer directly immersed in the bioreactors (500L) D. Each Raman spectra was the result of an average of 20 measurements with an integration/exposure time of 3 sec, and laser power setting at 450 mW. The total acquisition time per data spectra was 2 minutes, with a timestamp matched between the MarqMetrix All-In-One Process Raman Analyzer and off-line instrument analysis to build the model. Chemometrics, model building Independent data from multiple MarqMetrix All-In-One Process Raman Analyzer instruments, probes, and bioreactor types were used to create models. The training datasets were collected from 45 samples per bioreactor to create each chemometric model. The spectral data was reviewed, and outlier spectral spikes caused by cosmic rays were removed. The spectral region of interest was selected, and the spectra were preprocessed to remove the baseline and maximize signal to noise. For research use only. Not for use in diagnostic procedures. For current certifications, visit thermofisher.com/certifications © 2024 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. TS_MCS_AN_DynaDrive_MMAIO_0824 Learn more at thermofisher.com/marqmetrixAIO Table 1. Correlation of model prediction with offline data analysis. Metabolite Predicted R2 Predicted RMSEC RMSECV RMSEP Glucose (g/L) 0.98 0.43 0.49 0.40 Lactate (g/L) 0.92 0.15 0.18 0.25 Glutamine (mmol/L) 0.92 0.42 0.48 0.58 Titer 0.92 0.21 0.25 0.37 Cell Viability (%) 0.94 1.72 2.29 1.83 Figure 3. Thermo Scientific DynaDrive S.U.B. Chemometric Model Plots- Comparison of Raman Model vs Offline Analytical Data for Important Bioreactor Parameters. Conclusions The MarqMetrix All-In-One Process Raman Analyzer provides accurate in-line, real-time measurements of the critical process parameters glucose, glutamine, and lactate as well as total and viable cell densities in the 500L DynaDrive S.U.B. using the reusable MarqMetrix Single-Use Bioreactor BallProbe Sampling Optic with TouchRaman immersion technology. The correlation analysis shows excellent agreement between the model prediction data and the offline analytical data, indicating the robustness of the model applied to the parameters shown in Table 1. 0 2 4 6 8 2022 Glucose DynaDrive S.U.B. Gluc (g/L) RMSEP:0.4 Gluc (g/L) 8/25 8/27 8/29 8/31 9/01 9/03 9/05 9/07 Predicted Offline 0.0 0.5 1.0 1.5 2.0 2022 Lactate DynaDrive S.U.B. Lac (g/L) RMSEP:0.25 Lac (g/L) 8/25 8/27 8/29 8/31 9/01 9/03 9/05 9/07 Predicted Offline 0.0 1.0 2.0 3.0 3.5 2022 Titer DynaDrive S.U.B. Titer RMSEP:0.37 Titer 8/25 8/27 8/29 8/31 9/01 9/03 9/05 9/07 Predicted Offline 0 1 2 3 6 2022 Glutamine DynaDrive S.U.B. Gln RMSEP:0.58 Gln 8/25 8/27 8/29 8/31 9/01 9/03 9/05 9/07 Predicted Offline 80 85 90 95 100 2022 Cell Viability DynaDrive S.U.B. Viability RMSEP:1.83 Viability 8/25 8/27 8/29 8/31 9/01 9/03 9/05 9/07 Predicted Offline 0.5 1.5 2.5 4 5 Use of Lykos and TruBio Software Programs for Automated Feedback Control to Monitor and Maintain Glucose Concentrations in Real Time Application note Authors Juan Villa, Matthew Zustiak, David Kuntz, Lin Zhang, Nimesh Khadka, Kevin Broadbelt and Sue Woods (Thermo Fisher Scientific) Summary In this study, the Thermo Scientific™ MarqMetrix™ All-In-One Process Raman Analyzer with Thermo Scientific™ Lykos™ PAT Software and Thermo Scientific™ TruBio™ 6.0 Bioprocess Control Software was used to provide in-line feedback control of glucose in both Fed-Batch and Perfusion CHO cell culture processes without the need for operator intervention. This ability to monitor and maintain the desired glucose concentration in real time leads to improved process consistency and product quality, supporting crucial mAb post-translational modification of the product (e.g., glycosylation). An OPC-UA (Open Platform Communications United Architecture) connection between the Lykos PAT software and the TruBio 6.0 bioprocess control software was used for feedback control of one of the integrated pumps on the G3Lab controller supplying a concentrated glucose solution. Both Fed-Batch and Perfusion cultures were maintained automatically without operator process intervention. Introduction Process analytical technology (PAT) enables manufacturers to measure and control a process based on the product’s critical quality attributes (CQAs) in real time. Enhanced control of critical process parameters (CPPs) optimizes quality while reducing the cost and time of product development and manufacturing. With PAT, these CPPs and, in some instances, CQAs can be measured in real time, therefore leading to gains in essential real-time process information and the ability to create a quality-by-design workstream. In recent years, process Raman spectroscopy has gained popularity as a PAT tool that enables real-time monitoring and control of critical bioprocessing parameters which are key to the successful production of therapeutic drugs. Successful production implies high process efficiency, high and consistent product quality, and minimized manufacturing costs¹. Implementing PAT tools in biopharmaceutical manufacturing continues to receive much interest, allowing rapid development and access to therapeutics and existing medications without compromising high quality⁵, ⁶. Complex, multivariate, and univariate instrument data are interpreted, and based on this information, critical process parameters are predicted and, where necessary, adjusted to optimize the outcome of the process. Analytical results make it possible to predict the quality of the end material and understand how altering CPPs will affect the process and product. In turn, by executing experiments with real-time quality predictions, the relationships between the CPPs and CQAs can be established to develop true process understanding. Armed with this knowledge, it is possible to ‘close the loop’ and control the process using those quality predictions. Integration of continuous monitoring into a bioprocess and the application of analytical data is crucial for understanding the process and proactively addressing challenges. A key challenge is in-line monitoring of critical quality attributes such as glycosylation that affect the bioproduct’s stability, immunogenicity, safety, and potency. Maintaining the glucose concentration at a steady level in the bioreactor is extremely important to ensure consistent glycosylation of the product while optimizing product yields¹, ². This online continuous monitoring and control also allow for the reduction of manual sampling and feeding of the bioreactor, which are costly and inefficient and increase the risk of contamination each time the sterile system is accessed. This application note describes the use of the MarqMetrix All-In-One Process Raman Analyzer with Lykos PAT Raman Software to monitor glucose levels in a bioreactor and to trigger automated addition of feed so as to maintain the desired concentration. Communication between Lykos and the TruBio 6.0 control software was executed using OPC-UA⁹. A system that incorporates a sensitive Raman Process Analyzer and an established complex feedback loop allows for real-time monitoring of a critical parameter while maintaining stable concentrations. This system frees the operator from manual sampling and feeding, thereby eliminating the risks of contamination to the health of the cells. Materials & Methods Bioreactor Control System and Process Parameters Process monitoring and control were performed using the MarqMetrix All-In-One Process Raman Analyzer and TruBio 6.0 Bioprocess Control Software powered by the Emerson DeltaV Distributed Control System. Cell culture was performed in a HyPerforma Glass reactor and a HyPerforma G3Lab Controller. Fed-Batch strategy A 3L reactor was prepared with an initial working volume of 1.5L of SPM (ExpiCho stable production media + 6mM glutamine + 1 g/L pluronic) culture medium and inoculated with ExpiCho cells at 0.6*10⁶ cells/mL. Bioreactor environmental control parameters were set at a temperature of 36.5°C, a pH of 7.0, and a dissolved oxygen content of 40% air saturation. The pH level was controlled as needed using CO₂ gassing and 1M sodium carbonate additions. Feeding Strategy The cells were grown in a chemically defined medium and were fed using a continuous feeding process starting on day 3. The feed media (2X EFFICIENT FEED C+) was added at 3% reactor weight daily. The run was terminated after 14 days. The bioreactor was covered to protect it from stray light. Perfusion strategy Fresh basal medium addition was initiated on day 3 at 0.5 reactor volumes/day while spent media was removed through a proprietary perfusion device at the same rate, thereby maintaining the reactor volume at a constant level. The perfusion rate was increased stepwise up to 2 volumes per day by day 5 and held at that rate for the duration of the culture. Glucose control Strategy Glucose feedback control was initiated on day 3 for the fed-batch culture and day 0 for the perfusion culture, using the glucose value provided by Lykos to the TruBio 6.0 Bioreactor Control Software, to control the glucose concentration in the bioreactors at 3.0 g/L for the fed-batch and 4.0 g/L for the perfusion cultures. MarqMetrix All-In-One Process Raman Analyzer Monitoring Measurements were performed using the MarqMetrix All-In-One Process Raman Analyzer, with the optical BioReactor Ball Probe of the MarqMetrix All-In-One Process Raman Analyzer directly immersed in the bioreactors⁷, ⁸. Each Raman spectrum was the result of an average of 20 measurements with an integration/exposure time of 3000 milliseconds and laser power set at 450 mW. The total acquisition time per data spectra was 2 minutes (1 min dark spectrum correction and 1 min sample spectra) with a timestamp matched between the MarqMetrix All-In-One Process Raman Analyzer and offline instrument analysis to align the online and offline results. Measurements were taken every 2 minutes, and based on these measurements, the pump rate was adjusted automatically. Table 1. Media and Cell Line Summaries. Bioreactor Run Summaries Run Mode: Fed-Batch Run Mode: Perfusion • 1 Bioreactors (5L glass vessel) • Inoculation: 0.6*10⁶ cells ExpiCHOs • Initial media: ExpiCHO SPM • Feed media, EFC+ • Run time: 14 days • 1 Bioreactor (3L glass vessel) • Inoculation: 0.6*10⁶ cells ExpiCHOs • Initial media: HipCHO • Feed media: HipCHO (perfusion) • Run time: 10 days LyKos PAT Software The Lykos provided user access to the MarqMetrix All-InOne Process Raman Solution and data access for the stored Raman spectra. As part of this application note, two new capabilities were developed: A process analysis engine that applies a chemometric model to generate a process value, and an OPC-UA server which allows automated data access. Process Control Integration Glucose addition rate control by the TruBio 6.0 Process Control System was based on process measurements from the Lycos PAT Software via OPC-UA. Chemometric Model building Independent data from multiple MarqMetrix All-In-One Process Raman Analyzers and 3 bioreactors were used to create models. The training datasets were collected from 12–24 samples per bioreactor to create each chemometric model. In-line and at-line measurements were aligned using timestamps between the MarqMetrixAll-In-One Process Raman Analyzer and the at-line instrument, a Flex2 cell culture analyzer from Nova Biomedical. All data was reviewed before building the models. In addition, an algorithm was implemented to remove data spikes in the spectra caused by cosmic rays. The spectral region of interest was selected, and each measurement corresponded to ten minutes. The spectra were pre-processed to remove the baseline and, maximize the signal-to-noise ratio, and correct for path length differences. Partial Least Squares (PLS) models were created for each property of interest, and leave-out-one-run cross-validation was performed to test the optimization of each model. Properties of interest included glucose, lactate, and titer generated during the bioreactor culture run. Lykos PAT Software communicated directly with the bioreactorcontrolling software to send the glucose concentration measured in the bioreactor. A MarqMetrix All-In-One Process Raman Analyzer and Lykos PAT Software were integrated with TruBio 6.0 Biorprocess Control Software (powered via the DeltaV Distributed Control Platform from Emerson). Raman spectroscopy was used to determine the glucose concentration in the cell culture. Figure 1A. Process Control Diagram. Figure 1B. Process Monitoring & Control. Glucose Feed Solution TruBio 6.0 BioProcess Control System Process Value via OPC-UA Raman Spectra via Ethernet MarqMetrix All-In-One Process Raman Solution MarqMetrix BallProbe Sampling Optic BioReactor Feed Pump Control LASER Signal via Fiber Optic Cable Lycos PAT Software • LASER Pulse Injection • Raman Spectra Detection • Generate the spectra • Read the spectra • Calculate glucose concentration • Send the glucose value to TruBio 6 software control • Read the glucose concentration sent by Lykos PAT Raman Software • Adjust the pump rate for feed (containing glucose) Ethernet BioReactor LASER and Raman Signal Feed Control Results During the cell culture process, glucose was consumed by the cells, and concentration levels stabilized at around 3 g/L for the fed-batch culture and around 4 g/L for the perfusion reactor. Figures 2A & B show that once the target glucose concentration was reached, the glucose concentrations were precisely maintained at 3 g/L and 4 g/L for the fed-batch and perfusion cultures, respectively. Traditional offline samples were collected to measure the glucose and lactate concentrations as controls to compare the accuracy of the Raman analyzer measurements. Figure 2A. Fed-Batch Data from real-time control of glucose and monitoring of lactate and titer for a fed-batch culture are presented here. Figure 2B. Perfusion Data from real-time control of glucose and monitoring of lactate for a perfusion culture are presented here. Glucose Lactate Titer Correlation of Model Prediction with Offline Data Metabolite/Variable Predicted R² Predicted RMSEP Glucose (g/L) 0.890 0.315 Lactate (g/L) 0.929 0.223 Titer (g/L) 0.980 0.200 Correlation of Model Prediction with Offline Data Metabolite/Variable Predicted R² Predicted RMSEP Glucose (g/L) 0.95 0.34 Lactate (g/L) 1.00 0.36 Glucose Lactate Discussion In the above-described experiments, process data was monitored by Lykos PAT software and easily integrated with TruBio bioprocess control software via an OPC-UA protocol. The control loop demonstrated a very stable glucose concentration, which was achieved along with accurate measurements by the MarqMetrix All-In-One Process Raman Analyzer. Automating the process, as demonstrated, significantly reduced the intervention of the operator, who typically must sample the bioreactor for offline or at-line analyses. Because of this automation, the risk of contamination and batch rejection relating to operator error and other deviations is significantly reduced. In this application note we demonstrated a culture environment that can ensure maximum productivity in a reproducible manner using an automated feedback control loop. With real-time process monitoring by Thermo Fisher Scientific’s MarqMetrix All-In-One Process Raman Analyzer with Lykos PAT Raman Software and feedback control enabled by TruBio 6.0 Bioprocess Control Software, we have demonstrated a real-world application of PAT and the ability to implement a process developed with a quality-by-design approach. References 1. Ghaderi D, Zhang M, Hurtado-Ziola N, et al. Production platforms for biotherapeutic glycoproteins. Occurrence, impact, and challenges of non-human sialylation. Biotechnol Genet Eng Rev. 2012; 28:147–75. 2. Durocher Y, Butler M. Expression systems for therapeutic glycoprotein production. Curr Opin Biotechnol. 2009; 20:700–7. 3. Swiech K, Picanco-Castro V, Covas DT. Human cells: new platform for recombinant therapeutic protein production. Protein Expr Purif. 2012; 84:147–53. 4. Research, C. for D. E. and. PAT — A Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance. U.S. Food and Drug Administration. https://www.fda.gov/regulatory-information/ search-fda-guidance-documents/pat-framework-innovative-pharmaceuticaldevelopment-manufacturing-and-quality-assurance (accessed 2023-03-21). 5. Real time monitoring of multiple parameters in mammalian cell culture bioreactors using an in-line Raman spectroscopy probe. Nicholas R Abu-Absi, Brian M Kenty, Maryann Ehly Cuellar, Michael C Borys, Sivakesava Sakhamuri, David J Strachan, Michael C Hausladen, Zheng Jian Li. Biotechnol Bioeng. 2011 May;108(5):1215-21. doi: 10.1002/bit.23023. Epub 2010 Dec 22. 6. Quick generation of Raman spectroscopy based in-process glucose control to influence biopharmaceutical protein product quality during mammalian cell culture. Brandon N Berry , Terrence M Dobrowsky , Rebecca C Timson , Rashmi Kshirsagar, Thomas Ryll, Kelly Wiltberger Biotechnol Prog 2016 Jan-Feb;32(1):224-34. doi: 10.1002/btpr.2205. Epub 2015 Dec 21. 7. Real-time metabolite monitoring using the Thermo Scientific™ Ramina™ Process Analyzer System. Juan Villa, Mathew Zustiak, Elizabeth Amoako, David Kuntz, Lin Zhang and Kevin Broadbelt (Thermo Fisher Scientific). 8. Real-time metabolite monitoring using the Thermo Scientific™ Ramina™ Process Analyzer System and the Thermo Scientific™ 500L HyPerforma™ Dynadrive™ Single-Use Bioreactor (S.U.B.). Juan Villa, Mathew Zustiak, Elizabeth Amoako, David Kuntz, Lin Zhang and Kevin Broadbelt (Thermo Fisher Scientific). 9. Lange, J et al., Frank Iwanitz, and Thomas Burke, OPC From Data Access to Unified Architecture, 4th rev. Ed., OPC Foundation – Softing (VDE Verlag GMBH, 2010. Learn more at thermofisher.com/marqmetrixAIO For research use only. Not for use in diagnostic procedures. For current certifications, visit thermofisher.com/certifications © 2024 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. MCS-AN1012-EN 2/24 Additional upstream processing resources Streamlining seed train scale-up through utilization of the turndown ratio of the DynaDrive S.U.B. Single-use technologies Application note | DynaDrive Single-Use Bioreactors Introduction Upstream bioproduction has seen a substantial movement in the industry toward single-use systems. This has been driven primarily by the need to reduce contamination risk and cleaning requirements when compared to stainless steel systems, and to allow for faster changeover of equipment between batches. At the same time, bioprocessing manufacturing processes have matured significantly, and intensification of cell culture processes has pushed the limits of these legacy single-use systems. In recent years, Thermo Fisher Scientific has brought enhancements to the Thermo Scientific™ HyPerforma™ Single-Use Bioreactor (S.U.B.) platform with 5:1 S.U.B. and enhanced S.U.B. options, which have allowed for higher turndown ratios and optimized mixing and gassing strategies for intensified processes. Now Thermo Fisher Scientific has launched a next-generation bioreactor for the biopharmaceutical industry with the Thermo Scientific™ DynaDrive™ S.U.B. (Figure 1). Keywords DynaDrive S.U.B., single-use bioreactors, scalability, turndown Figure 1. DynaDrive S.U.B.s in 50 L, 500 L, 3,000 L, and 5,000 L sizes. S.U.B. seed train options During seed train expansion, cell cultures are sequentially increased in volume and cell population to provide enough starting material to enter batch, fed-batch, or perfusion production unit operations. Small-scale cell cultures, usually in less than 1 L of working volume, are typically maintained in flasks on an orbital shaking platform. Shake flask culture requires careful handling within a biosafety cabinet in a separate, higherclassification cleanroom to limit the risk of contamination. Due to the heightened risk associated with these open-unit operations, it is desirable for bioprocess engineers to move the cell culture into a closed system as early as possible in the scale-up process. A commonly chosen closed system for expansion steps after shake flasks is the rocking-motion bioreactor. This single-use closed system employs a platform to mix cells and gases to maintain viable cultures. After expanding in the rocker, cultures are transferred to stirred-tank S.U.B.s for further expansion prior to inoculating the N-stage production vessel (Figure 2A). Enhancements to our legacy S.U.B. systems, with the launch of the 5:1 and the enhanced S.U.B. systems, allowed for working volumes as low as 20% of the final working volume (e.g., 10 L in a 50 L vessel), which has allowed for scale-up of the seed train within the vessel and elimination of some seed train vessels (Figure 2B). With the innovative design features of the 50 L Building on our extensive experience and nearly two decades of end users’ feedback, the DynaDrive S.U.B. employs a new agitator drive technology with carefully engineered hardware that enables exceptional performance. The DynaDrive S.U.B. allows for a higher turndown ratio of at least 10:1 and up to 20:1 in the larger sizes, with a reliable power input of up to 80 W/m³, and scalability through each size. The innovative impeller design enables a higher turndown ratio than previously seen in single-use systems, offering working volumes as low as 5 L in the 50 L S.U.B. These higher turndown ratios open a new paradigm of what is possible with seed trains, potentially eliminating multiple vessels and reducing logistical and operating costs dramatically while increasing the efficiency of the seed train through reduced connections and transfer losses. DynaDrive S.U.B., cell cultures in working volumes as low as 5 L can be grown in a stirred-tank vessel, allowing for scale-up processes to happen at even earlier stages. This extremely high turndown ratio eliminates the need for rockers and extra steps in the seed train process further on, with the ability to seed the 5,000 L DynaDrive S.U.B. at 20:1 directly from the 50 L scale. This essentially means that the seed train process can be limited to two reactors, with each having scale-up steps that take place within them. For example, the 5 L minimum working volume of the 50 L DynaDrive S.U.B. enables cell culture scale-up to transfer directly from shake flasks to the S.U.B. After the 5 L culture reaches a viable cell density suitable for expansion, additional culture medium can be added to the vessel as a second passage within the 50 L DynaDrive S.U.B. Once that culture reaches a viable cell density suitable for expansion, it can then be used to seed the 5,000 L DynaDrive S.U.B. at a 250–500 L working volume, which subsequently can be passaged within the reactor for inoculation of the production run (Figure 2C). Alternatively, because of the flexibility of the DynaDrive S.U.B. systems, the 500 L DynaDrive S.U.B. can be used in the seed train in place of the 5,000 L DynaDrive S.U.B., increasing the required reactors to three, which allows for increased overall bioreactor production and throughput. In this scenario, the seed train would still utilize the 5 L minimum working volume of the 50 L DynaDrive S.U.B., but after reaching a viable cell density suitable for expansion, these cells would be used to seed the 500 L DynaDrive S.U.B. at a 25 L or 50 L volume. When ready, this would be expanded within the 500 L vessel and then used to inoculate the 5,000 L DynaDrive S.U.B. production run (Figure 2D). Since every additional aseptic connection is a potential vector for contamination, the high turndown ratios of the DynaDrive S.U.B. systems enable risk reduction for bioprocess scale-up operations. Performing expansions in the same vessel saves time in setup and reduces the quantity of sterile connections needed for the overall seed train expansion. 2 Figure 2. Improving facility efficiency with increased S.U.B. turndown ratios and careful logistical planning. Using bioreactors capable of high turndown ratios, such as the DynaDrive S.U.B.s, can enable a streamlined seed train for cell expansion. Benefits of high turndown ratios include increased facility space through reduction of bioprocess units, reduction of operator labor (e.g., setups, takedowns, and sterile or aseptic transfers), and reduction of logistical risks. (A) Example of a 2:1 turndown ratio operation that incorporates a rocking-motion bioreactor and stirred-tank S.U.B.s. (B) With the removal of at least one or more bioprocess units (e.g., rocker and S.U.B.), the cell expansion requires 20–25% less operator intervention and up to 50% fewer single-use Thermo Scientific™ BioProcess Containers, compared to a 2:1 turndown ratio operation. The high-intensity portions of operator labor (dark blue) of both the 5:1 and the DynaDrive S.U.B. seed trains are spaced at intervals of 2–4 days, reducing risk of operator error through fatigue or haste. (C) Using the 5,000 L DynaDrive S.U.B. at a turndown ratio of 10:1 as the N-1 bioreactor allows for a smaller footprint in the facility, less setup, and fewer sterile or aseptic transfers. (D) Alternatively, utilizing the 500 L DynaDrive S.U.B. in the seed train allows the 5,000 L DynaDrive S.U.B. to remain a dedicated production vessel. Maximum split ratio = 10; minimum cell density = 0.3 x 10⁶ cells/mL. Phase Day 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Incubator Rocker Incubator Incubator Incubator Final production phase 2,000 L Final production phase Final production phase Fed-batch production phase Fed-batch production phase Fed-batch production phase Fed-batch production phase 2,000 L Seed train 250 L 2:1 50 L 2:1 2:1 turndown maximum S.U.B. count: 4 Seed-train cell expansion phase Phase Seed-train cell expansion phase Phase Seed-train cell expansion phase Phase Seed-train cell expansion phase DynaDrive 10:1 turndown S.U.B. count: 3 Operator labor 10:1 50 L 10:1 500 L 10:1 5,000 L Final production phase 5,000 L 10:1 Seed train 50 L 10:1 DynaDrive 10:1 turndown S.U.B. count: 2 Operator labor 5:1 Operator labor 10:1 50 L 5:1 5:1 turndown maximum S.U.B. count: 3 Operator labor 2:1 500 L 5:1 A B C D 3 Application and benefits By utilizing DynaDrive S.U.B.s in seed trains, not only are there ideal turndown ratios of scaling within vessels, but the system has been optimized to provide ideal mixing and mass transfer, even at low turndown ratios, as shown in Table 1. Also shown in Table 1 is a comparison of the monitoring capabilities of each system. In comparing the single-use closed system options for seed train expansion, stirred-tank S.U.B.s allow for better monitoring and control than rocker systems. Due to design limitations of the rocker bioreactors, there are limited online process monitoring capabilities, whereas the 50 L DynaDrive S.U.B. can support temperature, pH, and DO probes as well as an additional probe at the 10:1 turndown ratio, with space for more probes at higher volumes (Table 1). This capability provides greater design space for bioprocess monitoring and enhanced process analytical technology (PAT) capabilities for quality by design (QbD) approaches. Table 1. Highlighted performance comparisons and monitoring capabilities of the HyPerforma, enhanced HyPerforma, and DynaDrive S.U.B. systems. S.U.B. Rocker HyPerforma Enhanced HyPerforma DynaDrive Generation Rocker 2:1 5:1 Enhanced fed-batch Enhanced perfusion DynaDrive Vessel size 10 to 50 L 50 to 2,000 L 50 to 500 L 50 to 5,000 L Scalable volumes 1 to 25 L 25 to 2,000 L 10 to 2,000 L 12.5 to 500 L 25 to 500 L 5 to 5,000 L Turndown ratio 5:1 2:1 5:1 4:1 2:1 10:1 or 20:1 Sparge Overlay only Frit and drilled-hole sparger (DHS) Enhanced DHS kLa at 20 W/m³ with DHS only No data ≤10 hr–1 ≤20 hr–1 ≤25 hr–1 20–30 hr–1 Max kLa No data  ≤15 hr–1 30–40 hr–1 >40 hr–1 Max P/V No data 40 W/m³ 100 W/m³ 80 W/m³ Dissolved oxygen (DO) sensor Must be singleuse, limited options available Standard or single-use available, multiple options pH sensor Must be singleuse, limited options available Standard or single-use available, multiple options Temperature control Electric, heat only Jacket, temperature control unit (TCU) Additional sensor port options None available 2 probe belts, no options below first probe belt 2 probe belts plus 1 DO and 1 pH option at low turndown ratio Additional port options at low turndown ratio 4 0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 0 3 6 9 12 15 18 VCD (x 106 cells/mL) Time (day) 50 L, VCD 500 L, VCD 5,000 L, VCD 50 L, viability 500 L, viability 5,000 L, viability Viability (%) Figure 3. VCD and viability of ExpiCHO-S cell culture from N-1 through N-stage production culture in the 50 L, 500 L, and 5,000 L DynaDrive S.U.B.s. Each N-1 culture was seeded at a 10:1 volume. The low viability seen in the N-1 culture in the 50 L S.U.B. occurred in a prototype BPC with known effects on cell health at low working volumes. Production BPCs have been shown to have reduced impact on cell health and growth across multiple cell clones tested. The following case study was done to evaluate the performance of the DynaDrive S.U.B.s across all scales from 50 to 5,000 L in a fed-batch process using CHO cells (Table 2), as well as to evaluate the performance of the cells during the seed train steps within these bioreactors. Methods Cells were expanded in shake flasks, the 50 L DynaDrive S.U.B., or both through the N-2 stage. The N-1 stage for each seed train was carried out at a 10:1 turndown ratio within the N-stage vessel. Fresh production medium was added to the S.U.B. after 3 days, resulting in the culture starting at the proper N-stage production volume and initial seed density. Operating conditions for both the N-1 and N-stage production vessels are described in Table 3. During the production run, daily bolus feeds of 2X concentrated EfficientFeed C+ AGT Supplement were added from days 3 to 13 through either a subsurface (5,000 L S.U.B.) or top feed (50 and 500 L S.U.B.) line. Following addition of EfficientFeed C+ AGT Supplement, glucose was supplemented in the same way on an as-needed basis after taking glucose measurements to bring the final glucose concentration to >4 g/L. Cell counts, viability, gases, nutrients, and metabolites were measured offline daily. All three DynaDrive S.U.B. volumes were tested in N-stage production runs. The 500 L and 5,000 L production runs came from the same seed train, with the 500 L N-1 being inoculated with cells from the 5,000 L N-1 stage reactor, prior to the start of that 5,000 L N-stage production run. Case study Seed train evaluation using ExpiCHO-S cell line in 14-day fed-batch run Table 2. Cell line evaluated in 50–5,000 L DynaDrive S.U.B.s. Cell type Gibco™ ExpiCHO-S™ Cells Production medium Gibco™ ExpiCHO™ Stable Production Medium Feed supplement Gibco™ EfficientFeed™ C+ AGT™ Supplement Titer range ~3 g/L Table 3. Operating parameters for evaluation of ExpiCHO-S cells in the 3 scales of DynaDrive S.U.B.s. S.U.B. 50 L 500 L 5,000 L Target starting volume 5 L, 35 L* 50 L, 350 L* 500 L, 3,500 L* Seeding density 0.3 x 10⁶ cells/mL, 0.7 x 10⁶ cells/mL* Temperature 37°C (N-1 and days 0–5), 34°C (days 5–14) pH 6.8–7.2 pH control Acid control: sparged CO₂ Base control: 1 N NaOH Acid control: sparged CO₂ Base control: 1 N NaOH Acid control: sparged CO₂ through the macro DHS Base control: 1 N NaOH Agitation 140 rpm, 120 rpm* 60 rpm 26 rpm for N-1 37 rpm (days 0–3) 33 rpm (days 3–14) DO 40% Air crossflow/headspace (slpm) 1 6 10–20 DO cascade Air supplemented with O₂ through DHS Air supplemented with O₂ through DHS Air supplied to both macro and micro DHS O₂ supplemented through micro DHS Feeding strategy Daily bolus of 1.05 L EfficientFeed C+ AGT Supplement and glucose (as needed) Daily bolus of 10.5 L EfficientFeed C+ AGT Supplement and glucose (as needed) Daily bolus of 105 L EfficientFeed C+ AGT Supplement and glucose (as needed) * The N-1 culture conditions are listed first, followed by the N-stage culture conditions. 5 For Research or Further Manufacturing. Not for diagnostic use or direct administration into humans or animals. © 2021, 2022 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. EXT3848 1022 Learn more at thermofisher.com/dynadrive Results Viable cell density (VCD) and viability (Figure 3) for the N-1 and production cultures show consistent growth profiles among the cultures, with similar cell density and viability trends at each step in the seed train. The viability at the end of the run was above 80% in all systems, which was within expectations. Conclusion Overall, the ability of the DynaDrive S.U.B. systems to operate at a 10:1 or 20:1 turndown ratio enables more efficient cleanroom utilization, reduces risk of contamination, and simplifies seed train expansion operations. This innovative product offering from Thermo Fisher Scientific enables consolidation of unit operations into fewer vessels and helps provide a more flexible manufacturing system for upstream bioprocessing. Application note | DynaDrive S.U.B. Part 1: Intuitive bioprocess scale-up from bench scale to pilot scale A comparative study of Thermo Scientific single-use bioreactors Introduction The transition from bench scale to pilot scale is often viewed as a critical step in bioproduction process development, generally because of physical differences of the separate systems. The ability to generate scalable parameters is critical to enable confidence when progressing to clinical and production scales. An example of scaling up a CHO-based bioprocess from bench scale to pilot scale using the Thermo Scientific™ DynaDrive™ Single-Use Bioreactor (S.U.B.) is outlined here, highlighting similarities and differences in control parameters and outcomes. In general, the culture carried out in the DynaDrive S.U.B. displayed key performance indicators that matched or exceeded those of the bench-scale bioreactor, including IgG titer, specific productivity, and cell density. Keywords Single-use bioreactor, DynaDrive, CHO, fed-batch, scale-up, Efficient-Pro, bench scale, pilot scale Single-use bioprocessing Methods IgG-producing CHO-K1 cells were thawed and propagated following standard procedures to establish a suitable seed train. The reactors used in this study included the 3 L Thermo Scientific™ HyPerforma™ Glass Bioreactor, the Thermo Scientific™ 5:1 HyPerforma™ S.U.B. (50 L), and the DynaDrive™ S.U.B. (50 L). Cell expansion was performed in a series of shake flasks until a sufficient number of cells were generated to inoculate the bioreactors. Control parameters used for the n-stage operation of each bioreactor are displayed in Tables 1–3. Table 1. Operation and control parameters. Parameter 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B. n-1 stage seed volume — 10 L 5 L Target initial/ final volume 1.7/2 L 36/50 L 36/50 L Seed density (x10⁶ cells/mL) 0.3 0.3 0.3 Temperature set point (°C) 37 37 37 Agitation (rpm) 350 183 105 Power input per volume (W/m³) 100* 20 20 Tip speed (m/sec) 1.01 1.06 0.59 Impeller configuration 2 down-pumping pitched-blade impellers 1 down-pumping pitched-blade impeller 3 down-pumping pitched-blade impellers Sparger configuration Drilled pipe sparger 7 x 800 µm holes Drilled-hole sparger (DHS) 360 x 178 µm pores DHS 1,448 x 80 µm pores Target glucose conc. (g/L) 3 3 3 Foam control High-threshold output 45 45 45 Foam alarm delay (sec) 60 60 60 Splash delay (sec) 5 5 5 * Approximate. Power number for the impeller configuration used has not been determined at time of publication. Table 2. Dissolved oxygen (DO) control gassing strategy. Parameter 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B. DO set point (%) 40 40 40 DO PID Gain 0.07 0.10 0.10 Reset 200 200 200 O₂ Controller output 15  100% 15  100% 15  100% MFC scaling 0  0.5 slpm 0  5 slpm 0  3 slpm N₂ Controller output 0  40% 0  30% 0  30% MFC scaling 0.15  0 slpm 1  0 slpm 1  0 slpm Air Overlay 0.2 slpm 5 slpm 5 slpm Sparge controller output — — 15  30  45% MFC scaling — — 0  0.5  0 slpm Table 3. pH control strategy. Parameter 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B. pH set point 7.15 7.15 7.15 pH PID Gain 0.04 0.04 0.04 Reset 200 200 200 pH deadband Not enabled Not enabled Not enabled CO₂ Controller output –100  0% –100  0% –100  0% MFC scaling 0.08  0 slpm 2  0 slpm 2  0 slpm Base Not enabled Not enabled Not enabled 2 Bioprocess scale-up from bench scale to pilot scale thermofisher.com/sut Although a power input per volume (P/V) of 20 W/m³ was targeted for the 50 L S.U.B.s, the P/V was approximately 100 W/m³ in the 3 L glass bioreactor culture. While the difference in P/V between scales is large, this is not a significant concern as P/V is frequently an impractical or unmeaningful scaling parameter at the benchtop scale. In this case, an agitation rate was selected for the 3 L glass bioreactor culture that provided for tip speed comparable to that of the 50 L HyPerforma S.U.B. Gibco™ Efficient-Pro™ medium was used throughout the expansion and in the bioreactors for the n-stage production process. Gibco™ Efficient-Pro™ Feed 1 was used to supplement the production run cultures from day 3 onward, feeding 2.25% of the current vessel volume daily (see Equation 1). A 2 M glucose solution was also used to supplement the cultures, targeting a Equation 1: Daily feed rate Feed rate (mL/min) = (Current volume (L) ) ( ) ( ) ( ) 1,000 mL 1 L 0.0225 day 1 day 1,440 min 3 g/L glucose concentration upon starting the fed-batch phase of the process. Gibco™ FoamAway™ Irradiated AOF Antifoaming Agent was used to control excess foam during operation via the use of the foam probe and pump triggered by Thermo Scientific™ TruBio™ automation software “Foam Hi Lim Out” as a remote set point for the pump. Reactors were sampled daily, and measurements were recorded for cell count, cell size, cell viability, metabolites, and protein titer. 3 Bioprocess scale-up from bench scale to pilot scale thermofisher.com/sut 3 Results Culture growth was similar among all reactors, with peak viable cell densities (VCD) between 60.4 x 10⁶ cells/mL and 72.0 x 10⁶ cells/mL (Figure 1). Comparing cell viability, some discrepancies between scales were apparent (Figure 2). The 50 L HyPerforma and DynaDrive S.U.B. cultures had comparable cell viability profiles, with the viability of the 3 L glass bioreactor culture trending slightly higher throughout the latter half of the process. Protein production was also similar among the conditions tested, with peak protein titer reaching between 3.48 and 3.73 g/L (Figure 3). Productivity behavior is further elucidated by evaluating specific productivity (QP), which is shown in terms of pg/cell per day (Figure 4). Excluding a couple of minor deviations, the specific productivity lies within a similar range for each of the vessels during the fed-batch portion of the process, with the overall trend being similar across all conditions. Figure 1. Viable cell density (VCD) profiles for CHO-K1 cells in the 3 L glass bioreactor, 50 L HyPerforma S.U.B., and 50 L DynaDrive S.U.B. 0 10 20 30 40 50 60 70 80 0 2 4 6 8 10 12 14 VCD (x 106 cells/mL) Day 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B. Figure 2. Viability profiles for CHO-K1 cells in the 3 L glass bioreactor, 50 L HyPerforma S.U.B., and 50 L DynaDrive S.U.B. 0 10 20 30 40 50 60 70 80 90 100 Viability (%) 0 2 4 6 8 10 12 14 Day Lorem ipsum 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B. Figure 3. IgG titer profiles for CHO-K1 cells in the 3 L glass bioreactor, 50 L HyPerforma S.U.B., and 50 L DynaDrive S.U.B. 0 2 4 6 8 10 12 14 Day Lorem ipsum 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B. 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 lgG titer (g/L) Figure 4. Specific productivity (QP, a 3-point moving average) profiles for CHO-K1 cells in the 3 L glass bioreactor, 50 L HyPerforma S.U.B., and 50 L DynaDrive S.U.B. QP is expressed in terms of picograms IgG per cell per day. 2 4 0 6 8 10 12 14 0 2 4 6 8 10 12 14 QP (pg/cell per day) Day 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B 4 Bioprocess scale-up from bench scale to pilot scale thermofisher.com/sut The lactate concentration in each reactor increased over the first few days, with a primary peak on day 4, after which the concentration dropped for 4–6 days before increasing again until process termination (Figure 5). While the lactate concentration in the 3 L culture was lower than that of both 50 L cultures during the primary peak, the lactate accumulation rate was higher in the vessel from day 10 and beyond, resulting in a higher final concentration. The relatively larger increase in lactate concentration in the 3 L glass bioreactor culture coincided with a drop in pCO₂ near the end of the process, falling as low as ~25 mmHg, whereas pCO₂ in both 50 L cultures was maintained at 40–80 mmHg from day 3 onward. Scaled total gas flow rates (vessel volume per minute, VVM) show a lower O₂ sparge rate in the glass reactor and the DynaDrive S.U.B. compared to the HyPerforma S.U.B. (Figure 6). Nitrogen flows are also shown in the figure, and any sparged air flow rates were treated as 21% O₂ and 78% N₂. A direct comparison of the gassing requirements of the glass bioreactor and S.U.B.s is somewhat difficult due to the relatively high P/V in the glass reactor and varying rates of sparged N₂ in each condition. Even so, it can be seen from the similarity in scaled flow rates that the 50 L DynaDrive S.U.B. serves as a straightforward option when considering scalability from bench to pilot scale. Inspection of the CO₂ sparge rates (Figure 7) and referring to Figure 5 show that acceptable pCO₂ concentrations were maintained in the 50 L cultures even while the CO₂ flow rate was insignificant. Figure 5. Lactate and pCO₂ profiles for CHO-K1 cells in the 3 L glass bioreactor, 50 L HyPerforma S.U.B., and 50 L DynaDrive S.U.B. 0 2 4 6 8 10 12 14 Day Lorem ipsum 3 L HyPerforma Glass Bioreactor lactate 50 L HyPerforma S.U.B. lactate 50 L DynaDrive S.U.B. lactate 50 L DynaDrive S.U.B. pCO2 3 L HyPerforma Glass Bioreactor pCO2 50 L HyPerforma S.U.B. pCO2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 Lactate (g/L) 0 20 40 60 80 100 120 140 160 pCO2 (mmHg) Figure 6. Oxygen and nitrogen gas flow through the drilled-hole sparger for each vessel in terms of vessel volume per minute. 2 L HyPerforma Glass Bioreactor O2 50 L HyPerforma S.U.B. O2 50 L DynaDrive S.U.B. O2 0 2 4 6 8 10 12 14 Day Lorem ipsum 3 L HyPerforma Glass Bioreactor O2 50 L HyPerforma S.U.B. O2 50 L DynaDrive S.U.B. O2 3 L HyPerforma Glass Bioreactor N2 50 L HyPerforma S.U.B. N2 50 L DynaDrive S.U.B. N2 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 Gas flow (VVM) Figure 7. pH profile and carbon dioxide gas flow through the drilled-hole sparger in terms of vessel volume per minute. 0 2 4 6 8 10 12 14 Day Lorem ipsum 3 L HyPerforma Glass Bioreactor pH 50 L HyPerforma S.U.B. pH 50 L DynaDrive S.U.B. pH 3 L HyPerforma Glass Bioreactor CO2 50 L HyPerforma S.U.B. CO2 50 L DynaDrive S.U.B. CO2 0.025 0.020 0.015 0.010 0.005 0.000 CO2 flow rate (VVM) 7.30 7.25 7.20 7.15 7.10 7.05 7.00 6.95 6.90 6.85 6.80 pH 5 Bioprocess scale-up from bench scale to pilot scale thermofisher.com/sut 5 Totals for feeds and supplements at the time of run termination are provided in Table 4. Antifoam requirements are significantly lower in the 3 L culture compared to the 50 L cultures due to much smaller sparged gas flow rates, larger pore sizes in the drilled-hole sparger, and the foam sensor's proximity to the liquid level in the smaller culture. Table 4. Total feed and supplement (from calibrated pump totalizer) added by the end of process. Parameter 3 L HyPerforma Glass Bioreactor 50 L HyPerforma S.U.B. 50 L DynaDrive S.U.B. Total feed 442 mL 10.0 L 9.9 L Total 2 M glucose solution 160 mL 3.2 L 2.6 L Total FoamAway agent 4.7 mL 350 mL 280 mL 6 Bioprocess scale-up from bench scale to pilot scale thermofisher.com/sut For Research Use or Further Manufacturing. Not for diagnostic use or direct administration into humans or animals. © 2024 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. COL028238 0724 Learn more at thermofisher.com/sut Author Jace Parkinson, Engineer II, Systems Design, Thermo Fisher Scientific, Logan, UT Discussion and conclusion Considerations for scale-up of cell culture processes frequently center around agitation and gassing strategies. The control parameters used in this study were found to produce comparable results when scaling from benchtop glass bioreactors to Thermo Scientific™ S.U.B.s at the 50 L scale. Cell growth profiles, metabolism, and productivity were maintained in similar ranges in each of the bioreactors. While an aggressive CHO-K1 clone was used, with VCD reaching 65 x 10⁶ cells/mL in the DynaDrive S.U.B., the cell cultures were easily maintained within the design constraints of the reactors, with mixing and gassing demands being met in the glass bioreactor and the S.U.B.s without difficulty. The final protein titer at the time of harvest was 3.48 g/L in the glass bioreactor and 3.73 g/L in the DynaDrive S.U.B., with productivity being maintained during the duration of the culture. Ordering information Product Cat. No. Efficient-Pro AGT Medium A5322303 Efficient-Pro AGT Feed 1 A5209102 FoamAway Irradiated AOF Antifoaming Agent A1036902 HyPerforma G3 Lab Controller F100-2695-001 HyPerforma G3 Lite Controller F100-2701-001 HyPerforma G3 Pro Controller F100-2961-001 HyPerforma Glass Bioreactor (3 L) F100-2680-002 HyPerforma 5:1 Single-Use Bioreactor (50 L) SUB0050.9500 Bioprocess Container for HyPerforma 5:1 S.U.B. (50 L) SH31073.01 DynaDrive Single-Use Bioreactor (50 L) DDB0050.1011 DynaDrive Bioprocess Container (50 L) SH31192.01 Scale-up evaluation of the DynaDrive S.U.B.s Part 1: ExpiCHO-S Cells and CHO-S Cells (cGMP banked) Single-use technologies Application note | DynaDrive S.U.B. Introduction As a molecule approaches commercial launch and more is known about the potential market demand, companies are often faced with the decision to scale up or scale out their manufacturing processes. Generally, when scale-up production vessels of more than 2,000 L are required, this decision also involves moving from single-use bioreactors (S.U.B.s) to traditional stainless-steel systems. Additionally, more recently developed process intensification methods have allowed manufacturers increased product output, pushing titers past 10 g/L in some cases. These output achievements require increased production efficiency and input, pushing many bioreactor systems past their limits. S.U.B. quality requirements, robustness, and functional performance can all become constraints, especially at scales up to 2,000 L. For example, as oxygen transfer rate (OTR) becomes a limiting factor, most traditional S.U.B.s rely primarily on increased sparging flow to increase oxygen mass transfer. Keywords Single-use bioreactor, DynaDrive S.U.B., scalability, fed-batch Maintaining a dissolved oxygen (DO) target in high-demand cell cultures can be increasingly difficult due to limitations in the amount of mixing power that can be distributed effectively through the drivetrain of traditional S.U.B.s. Sparging through a micro-sparger has become a widely used strategy to improve OTR in traditional S.U.B.s and typically requires a secondary sparger to facilitate removal or stripping of dissolved CO₂ (measured as the partial pressure of CO₂, or pCO₂). Some cell lines, however, are sensitive to the higher shear produced by micro-sparging, and process scale-up cannot depend on this method alone to ensure sufficient O₂ delivery or CO₂ removal. A next generation of S.U.B., the Thermo Scientific™ DynaDrive™ S.U.B., with vastly improved mixing and mass transfer performance, is now enabling scale-up to 5,000 L and process intensification. Previous limits are no longer a burden for the DynaDrive S.U.B., and it continues to leverage known and acknowledged benefits of legacy units. DynaDrive S.U.B.s are multifunction reactors for a range of applications, including intermediate-scale production of preclinical, clinical, and commercial material, as well as perfusion for both production and N-1 seed processes. Additionally, each DynaDrive S.U.B. includes features that are improved over legacy and alternative S.U.B. options: • Each system is equipped with a Thermo Scientific™ BioProcess Container (BPC) load assist device, reducing handling and setup time, increasing safety, and providing consistent BPC loading. BPC loading can be accomplished in less time at the 50 L and 500 L scales compared to legacy S.U.B.s, and in less than 45 minutes at the 3,000 L and 5,000 L scales. • Best-in-class enhanced drilled-hole sparger (DHS) provides repeatable and reliable performance that users of S.U.B.s have embraced due to its linear scale-up benefits. • Revolutionary drivetrain design with multiple impellers allows increased power input and efficiency while offering reduced shear rates. • Cuboid design contributes to better BPC fit and increased baffled-like mixing efficiency, and allows more productive use of facility footprint. • 10:1 or better turndown ratio reduces facility requirements and investment costs while increasing flexibility in seed train applications and all aspects of scale-up. • Continuous mixing during harvest and minimal hold-up volume (<1%) after drain. • Improved exhaust system for the 3,000 L and 5,000 L S.U.B.s, allowing for increased gas flow rates and utmost reliability typically required for production-scale cultures. These major design changes have enabled a power-to-volume (P/V) ratio of up to 80 W/m³ in all sizes, t95 mixing times of less than 60 sec, and kLa performance of at least 40 hr–1 at all scales (Table 1). Additionally, the DynaDrive S.U.B. allows for process scale-up and transfer from legacy S.U.B.s, offering benefits of consistent BPC film, assurance of supply, robust quality controls, BPC integrity, and industry-leading BPC customization options. End users can continue using previously qualified traditional and single-use sensing options as well as inlet and exhaust filters and other peripheral components integrated through high-strength porting and line sets. Goal The goal of this study was to evaluate the performance of the DynaDrive S.U.B. across 50 L–5,000 L scales using two different cell lines (Table 2) together with previously developed processes specific to those cell lines for manufacturing up to a 2,000 L scale. These experiments were designed to demonstrate that the DynaDrive S.U.B. could be successfully implemented for use with multiple cell lines across scales with standard scale-up criteria. Both cell lines were subjected to a 14-day fed-batch run at full working volume for each scale. Table 1. Comparison of DynaDrive S.U.B. capabilities. Parameter 50 L S.U.B. 500 L S.U.B. 5,000 L S.U.B. Maximum volume 50 L 500 L 5,000 L Turndown ratio 10:1 20:1 20:1 kLa >50 hr–1 >50 hr–1 40 hr–1 t 95 mixing times <30 sec <40 sec <60 sec Maximum P/V ratio 80 W/m³ 80 W/m³ 80 W/m³ Table 2. Cell lines evaluated in 50 L–5,000 L DynaDrive S.U.B.s. Parameter Cell line 1 Cell line 2 Cell type Gibco™ ExpiCHO-S™ Cells Gibco™ CHO-S™ Cells (cGMP banked; part of the Gibco™ Freedom™ CHO-S™ Kit) Production medium Gibco™ ExpiCHO™ Stable Production Medium (SPM) Gibco™ Dynamis™ Medium Feed supplement Gibco™ EfficientFeed™ C+ AGT™ Supplement Titer range Medium: ~3 g/L Low: ~1 g/L Cell line characteristics Platform cell line Legacy cell line Case study 1 2 2 DynaDrive S.U.B. thermofisher.com/dynadrive DynaDrive S.U.B. thermofisher.com/dynadrive Case study 1 Scale evaluation using ExpiCHO-S Cells in a 14-day fed-batch run Methods Cells were expanded in shake flasks or pilot-scale S.U.B.s through the N-2 stage. The N-1 stage for each seed train was performed at a 10:1 turndown ratio. Fresh production medium was added to the S.U.B. after 3 days, resulting in the culture starting at proper N-stage production volume and initial seed density. Operating conditions are described in Table 3. Daily bolus feeds of 2X concentrated EfficientFeed C+ AGT Supplement were added from day 3 to 13 via either a subsurface (5,000 L S.U.B.) or top feed line (50 L and 500 L S.U.B.s). Glucose was supplemented in the same manner on an as-needed basis after taking a glucose measurement following addition of EfficientFeed C+ AGT Supplement to bring the final glucose concentration to >4 g/L. Cell counts, viability, dissolved gases, nutrients, and metabolites were measured offline daily. Titer samples were filtered and frozen daily starting on day 6 for batch testing at the culmination of the run. Table 3. Operating parameters for evaluation of ExpiCHO-S Cells in 3 scales of DynaDrive S.U.B.s. S.U.B. 50 L 500 L 5,000 L Target starting volume 35 L 350 L 3,500 L Seeding density 0.7 x 10⁶ cells/mL Temperature 37°C (days 0–5) 34°C (days 5–14) pH 6.8–7.2 pH control Acid control: sparged CO₂ Base control: 1 N NaOH Acid control: sparged CO₂ Base control: 1 N NaOH Acid control: sparged CO₂ through macro DHS Base control: 1 N NaOH Agitation 120 rpm 60 rpm 26 rpm (days 0–3) 33 rpm (days 3–14) DO 40% Air headspace 1 slpm 6 slpm 10–20 slpm DO cascade Air supplemented with O₂ through DHS Air supplemented with O₂ through DHS Air supplied to both macro and micro DHS; O₂ supplemented through micro DHS Feeding strategy Daily bolus of 1.05 L EfficientFeed C+ AGT Supplement, and glucose (as needed) Daily bolus of 10.5 L EfficientFeed C+ AGT Supplement, and glucose (as needed) Daily bolus of 105 L EfficientFeed C+ AGT Supplement, and glucose (as needed) Figure 1. VCD and viability comparison of ExpiCHO-S Cells in the 3 DynaDrive S.U.B.s. 0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 0 2 4 6 8 10 12 14 Viability (%) VCD (x 106 cells/mL) Time (day) 50 L, VCD 500 L, VCD 5,000 L, VCD 50 L, viability 500 L, viability 5,000 L, viability Figure 2. Titer results for ExpiCHO-S Cells in the 3 DynaDrive S.U.B.s. 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0 2 4 6 8 10 12 14 IgG (g/L) Time (day) 50 L 500 L 5,000 L DynaDrive S.U.B. thermofisher.com/dynadrive 3 Figure 3. pCO₂ profile measurements for ExpiCHO-S Cells in the 3 DynaDrive S.U.B.s. 0 20 40 60 80 100 120 0 2 4 6 8 10 12 14 pCO2 (mm Hg) Time (day) 50 L 500 L 5,000 L Figure 4. Lactate and NH₄⁺ profiles for ExpiCHO-S Cells in the 3 DynaDrive S.U.B.s. 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 1 2 3 4 5 6 7 8 9 10 0 2 4 6 8 10 12 14 NH4+ (mM) Lactate (g/L) Time (day) 50 L, lactate 500 L, lactate 5,000 L, lactate 50 L, NH4+ 500 L, NH4+ 5,000 L, NH4+ Figure 5. DHS gas flow rates for ExpiCHO-S Cells in the 3 DynaDrive S.U.B.s. 0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035 0.040 0.045 0 2 4 6 8 10 12 14 DHS gas flow (VVM) Time (day) 5,000 L, O2 VVM 5,000 L, micro DHS, air VVM 500 L, O2 VVM 500 L, air VVM 50 L, O2 VVM 50 L, air VVM 5,000 L, macro DHS, air VVM Results Viable cell density (VCD) and viability for the cultures show consistent growth profiles among the cultures, with similar cell density and viability trends (Figure 1). Peak VCDs were similar at ~20 x 10⁶ cells/mL, and the end-of-run viability was above 80% in all systems. IgG concentration measured 2.7–3.0 g/L on day 14 and is within ranges observed historically (Figure 2). While CO₂ levels in the 5,000 L culture were within expected conditions, levels in the 50 L and 500 L cultures were slightly lower than anticipated, indicating possible over-stripping of CO₂ (Figure 3). Metabolite data collected offline indicated healthy cultures with maintained glucose and low levels of metabolic byproducts, including lactate and ammonium, staying within ranges observed historically (Figure 4). Minimum glucose concentrations were maintained according to the culture protocol by feeding with EfficientFeed C+ AGT Supplement daily, and a glucose solution when needed. The gas flow rates were controlled based on culture oxygen demand (Figure 5). For the 50 L and 500 L cultures, DO was maintained by first sparging air, then supplementing with O₂ through the single DHS. For the 5,000 L culture, DO was maintained by sparging air through the macro DHS at a constant flow rate and sparging an air–O₂ mix through the micro DHS. Using these strategies, DO was maintained at 40 ± 5% with no major fluctuations during the duration of the run. Importantly, gas flow requirements for the cultures remained very low at all scales, never reaching past 0.035 vessel volumes per minute (VVM) for either the 50 L or 500 L S.U.B. For the 5,000 L S.U.B., gas flow requirements were below 0.01 VVM for the micro DHS with a constant 0.005 VVM for the macro DHS. These low gas flow rates in conjunction with the relatively low power inputs used at low rpm represent only about 30% of the available performance capacity of the system. The sparge strategy of the 5,000 L S.U.B. was adjusted on days 3–6 to successfully keep pCO₂ levels below 80 mm Hg for the balance of the cell culture run. 4 DynaDrive S.U.B. thermofisher.com/dynadrive DynaDrive S.U.B. thermofisher.com/dynadrive Case study 2 Scale evaluation using CHO-S Cells (cGMP banked) in a 14-day fed-batch run Methods Cells were expanded in shake flasks or pilot-scale S.U.B.s until seeding into the production vessels at either a 10:1 or 20:1 turndown ratio. The 50 L and 500 L S.U.B.s were brought to full volume with fresh production medium after 1–2 days, while the 5,000 L S.U.B. was filled to a 5:1 turndown ratio after 3 days and then full production volume after another day of growth. Operating conditions for each bioreactor are described in Table 4. A 2X concentration of EfficientFeed C+ AGT Supplement was added continuously from day 3 to 11 through either a subsurface (5,000 L S.U.B.) or top feed line (50 L and 500 L S.U.B.s). Glucose was supplemented in a continuous drip as needed depending on culture demands, to maintain glucose concentrations of 1–3 g/L. Cell counts, viability, dissolved gases, metabolites, and nutrients were measured offline daily. Titer samples were filtered and frozen daily starting on day 6. Titer samples were batch tested at the end of the run. Results VCD and viability for the cultures show similar growth profiles among the cultures, with similar peak VCD and viability trends (Figure 6). The 5,000 L culture exhibited slightly slower growth and lower peak VCD compared with the 500 L culture (VCD of 23 x 10⁶ and 26 x 10⁶ cells/mL, respectively) but was still within expected ranges. Productivity throughout the run was within expected ranges (Figure 7). The 50 L culture exhibited the lowest harvest titer, likely due to low pCO₂ in the latter half of the run. Metabolites and gas flow demands were similar in concentration and magnitude compared to the ExpiCHO-S cell cultures. Glucose was maintained in the S.U.B.s at around 1–5 g/L, while lactate and ammonium concentrations were within ranges observed historically (data not shown). Table 4. Operating parameters for evaluation of CHO-S Cells in 3 scales of DynaDrive S.U.B.s. S.U.B. 50 L 500 L 5,000 L Target starting volume 42.5 L 425 L 4,250 L Seeding density 0.3 x 10⁶ cells/mL Temperature 37°C pH 6.8–7.2 pH control Acid control: sparged CO₂ Base control: not applicable Acid control: sparged CO₂ Base control: not applicable Acid control: sparged CO₂ through macro DHS Base control: not applicable Agitation 120 rpm 68 rpm 31–41 rpm DO 30% Air headspace 3 slpm 6 slpm 10–20 slpm DO cascade N₂ and O₂ through the DHS N₂ and O₂ through the DHS N₂ and O₂ through the micro DHS Feeding strategy 7.5 L of EfficientFeed C+ AGT Supplement added on continuous drip from day 3 to 11, and glucose as needed 75 L of EfficientFeed C+ AGT Supplement added on continuous drip from day 3 to 11, and glucose as needed 750 L of EfficientFeed C+ AGT Supplement added on continuous drip from day 3 to 11, and glucose as needed Figure 6. VCD and viability comparison of CHO-S Cells in the 3 DynaDrive S.U.B.s. 0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 30 35 40 45 50 0 2 4 6 8 10 12 14 Viability (%) VCD (x 106 cells/mL) Time (day) 50 L, VCD 500 L, VCD 5,000 L, VCD 50 L, viability 500 L, viability 5,000 L, viability Figure 7. Titer results for CHO-S Cells in the 3 DynaDrive S.U.B.s. 50 L 500 L 5,000 L 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0 2 4 6 8 10 12 14 lgG (g/L) Time (day) DynaDrive S.U.B. thermofisher.com/dynadrive 5 Discussion As cell cultures are scaled to large-volume S.U.B.s, several factors can become more difficult to manage and control. Often, decisions related to culture parameters and operation thresholds must be balanced to provide the culture with the best opportunity for success. The DynaDrive S.U.B. assists with ensuring culture success due to specifically designed aspects of the S.U.B., including the geometrically scalable agitator drivetrain and the uniquely linear sparge performance of the DHS, which both directly benefit mass transfer and mixing abilities. The agitator drivetrain of the DynaDrive S.U.B. with multiple impellers at optimal locations allows for optimal and scalable power input. The cultures performed in these case studies were operated at a power input of approximately 20 W/m³, far below the maximum recommended 80 W/m³. Mass transfer in this design is more proportional to the power input of a system due to the mixer’s ability to both disperse and retain sparged gas. Therefore, if a culture were to demand more mass transfer, increased mixer speeds above those tested here would be an option for improved performance. The gas flow rates observed in these studies were far below the maximum rated gas flow rates for each system. As with power input for each system, mass transfer is shown to be linearly proportional and scalable to gas flow rate in the DynaDrive S.U.B.s. For the 50 L and 500 L ExpiCHO-S cell cultures, total gas flow rate through the DHS was never higher than 0.035 VVM, less than 25% of the maximum 0.15 VVM limit. For the 5,000 L ExpiCHO-S cell culture, total gas flow rate through the micro DHS reached only 0.007 VVM (30 slpm) for most of the culture. Additionally, the 5,000 L culture used a minimal macro DHS flow rate of only 15 slpm of air to maintain CO₂ stripping and to provide additional O₂ mass transfer. These chosen gas flow rates resulted in easily maintained O₂ mass transfer to support cultures of light to moderate demand with cell densities at 20–30 x 10⁶ cells/mL. More demanding clones are expected to require higher gas flow rates or agitation but still remain within design limits of the DynaDrive S.U.B.s. It is important to note that the DHSs were designed with specific pore quantities and sizes to provide these high levels of mass transfer at low gas flow rates, removing the need for other sparger types such as the legacy sintered sparger. One of the biggest concerns when scaling to larger S.U.B. sizes is the ability to control dissolved CO₂ concentrations within ideal physiological ranges (60–100 mm Hg). While spargers in smaller S.U.B.s typically provide sufficient CO₂ stripping capability due to their often oversized bubbles, shorter liquid column heights, and favorable overlay surface-to-volume ratios, spargers for larger S.U.B.s can be limited in providing sufficient CO₂ stripping while maintaining required O₂ mass transfer at reasonable gas flow rates. The 5,000 L DynaDrive S.U.B. is equipped with 3 separate DHSs to provide optimal gassing to drive mass transfer for both gases of interest. The two larger macro DHSs provide an amount of O₂ mass transfer while also providing more CO₂ stripping capability due to the larger bubbles created, while the single micro DHS provides higher O₂ mass transfer due to the smaller bubbles created. Balancing these spargers in tandem with agitation in these cultures allowed pCO₂ to remain within desired ranges for each culture, thus providing a very generous operating design window in anticipation of future process intensification that may be requested by the end user. Culture conditions in a S.U.B. are highly variable and must be balanced with multiple inputs and outputs. Through experience and tracking online and offline readings, setpoints can be balanced for gas flow rates, agitation, pH, DO, and feed flow rates. Generally speaking, metabolite buildup such as lactate and respired CO₂ lead to acidic conditions in the reactor that require the addition of base to balance the culture pH. Additionally, especially in larger vessels, buildup of pCO₂ can be detrimental to cell health. The S.U.B.s in this study were able to maintain pCO₂ and pH conditions through the employed gassing strategies, leading to pCO₂ levels maintained within physiological conditions. The 50 L and 500 L cultures actually showed a propensity to exhibit too much CO₂ stripping. While the pH in each culture was balanced sufficiently, more optimal gassing and pH control strategies could be employed in the future to provide more optimal growth and production conditions. Finally, each reactor was seeded at a low working volume: 10:1 turndown ratio in the 50 L S.U.B. and 20:1 turndown ratio in the 500 L and 5,000 L S.U.B.s. This has been shown to reduce the complexity of seed train requirements by eliminating intermediate vessels, consumables, and operation space. While using a low turndown ratio in a production vessel such as the 5,000 L DynaDrive S.U.B. is not ideal for some commercial applications seeking to maximize daily productivity, this feature allows for flexibility in manufacturing spaces not previously available. 6 DynaDrive S.U.B. thermofisher.com/dynadrive DynaDrive S.U.B. thermofisher.com/dynadrive For Research Use or Further Manufacturing. Not for diagnostic use or direct administration into humans or animals. © 2021, 2022 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. EXT3458 0822 Learn more at thermofisher.com/dynadrive Authors: Jordan Cobia, Product Manager; Ben Madsen, R&D Manager Conclusions The DynaDrive S.U.B.s were able to support both clones in this study and provide controlled conditions to achieve target VCDs and titers while maintaining high viability. This was achieved by choosing simple scale-up parameters among the S.U.B. sizes, including consistent power input and gas flow rates. The low gas flow rates and agitation rates tested provided sufficient mass transfer to maintain DO setpoints while maintaining pCO₂ levels at or below maximum limits. Additionally, the ability of each S.U.B. to be seeded at low volume allowed reduction in seed train complexity by using the S.U.B. at the N-1 culture stage. Fewer manipulations, media preparations, and fluid transfers, and less overall consumption of resources have the potential to noticeably mitigate risk, reduce waste, and lower operating costs. Overall, the different scale-up processes demonstrated in these studies show the versatility of the DynaDrive S.U.B. to maintain culture setpoints compared to historical processes with simple scale-up criteria. The modified drivetrain and sparging, when compared to legacy S.U.B. products, did not have adverse effects on the cell cultures, thus enabling a much larger design space for process development and production than previously available. Additionally, the DynaDrive S.U.B. provides consistent, scalable performance from 50 L to 5,000 L. With this demonstrated consistency, the DynaDrive S.U.B. will enable users to scale up their process with minimal process changes to meet demand for commercial therapeutics. Raman spectroscopy Application note MarqMetrix All-In-One Process Raman Analyzer – Chemometric model transferability across instruments Summary • The Thermo Scientific™ MarqMetrix™ All-In-One Process Raman Analyzer represents a breakthrough in process analytical technology (PAT) by providing in-line, real-time and actionable monitoring of multiple analytes in complex systems. • This paper demonstrates the successful transferability of one chemometric model across 10 different MarqMetrix All-In-One Process Raman Analyzers while maintaining an average prediction error of 0.21 g/L, 0.16 g/L and 0.3 g/L for glucose, glutamine, and lactate, respectively in a mixture. • Measurement accuracy and precision is maintained when applying a chemometric model across multiple instruments, ensuring users do not have to spend time or resources to re-build a model for a new MarqMetrix All-In-One Process Raman Analyzer or new autoclavable probe. Introduction The MarqMetrix All-In-One Process Raman Analyzer is a Raman spectroscopy instrument designed to offer rapid, robust, scalable, and reliable identification, quantification, and characterization of molecules during any phase of R&D process development. It is an “all-in-one” instrument utilizing easily exchangeable, autoclavable probes to meet the various analytical needs for, but not limited to, upstream bioprocess monitoring, or characterization of fill and finish products. Chemometric analysis allows users to develop a data analysis model to monitor the concentration of multiple analytes from their analyzer. There is a significant investment made from a time and resource perspective to build an accurate and robust chemometric model. As a result, it is imperative for a chemometric model to be used between instruments to fully leverage the value of this investment. Once a chemometric model is developed, it can be used with any MarqMetrix All-In-One Process Raman Analyzer Authors Mike Bates, Nimesh Khadka, David Kuntz, Lin Zhang, Sue Woods to monitor multiple reactors in parallel with high fidelity of model accuracy. The measurement accuracy and precision maintained when transferring chemometric models across different system hardware ensures that customers do not have to re-build a model when using a new instrument or new autoclavable probe. Experimental set up To demonstrate the transferability of a complex chemometric model, we evaluated three relevant analytes: glucose, glutamine and lactate in Gibco™ DMEM growth media. All three analytes occur in bioreactors in the g/L concentration ranges with glucose occurring in the range of 0-12 g/L, glutamine 0-2.5 g/L and lactate 0-20 g/L. A chemometric model was developed by collecting spectra from a set of calibration standards using one MarqMetrix All-In-One Process Raman Analyzer. Within the relevant concentration ranges for each analyte, 24 samples with randomized concentrations were selected using the uniform design method2 . The model was then applied to spectra from a different set of 8 validation samples, measured using 10 unique MarqMetrix All-In-One Process Raman Analyzers. The acquisition parameters were optimized to maximize the signal-to-noise ratio of the spectral features corresponding to the analyte concentrations. Optimized parameters were 15 second integration time, 450 mW laser power and 10 replicate on-board signal averaging with automatic dark correction. These acquisition parameters resulted in a sample collection period of about 5 minutes. The model was able to predict the concentration of all three analytes in the 8-sample validation set with a high degree of accuracy and precision. Each MarqMetrix All-In-One Process Raman Analyzer evaluated in this study included a unique hardware set comprised of spectrometer box, fiber optic cables and Thermo Scientific™ MarqMetrix™ Bioreactor BallProbe™ Sampling Optic tip. Average prediction error for each analyte was 0.21 g/L for glucose, 0.16 g/L for glutamine and 0.3 g/L for lactate. These results demonstrate the excellent transferability of this complex chemometric model across numerous MarqMetrix All-In-One Process Raman Analyzers. Model analysis A single PLS model was built using Eigenvector Solo software to predict all three analytes in each mixture. The PLS calibration model was built using 24 calibration samples with 3 replicates per sample. All calibration spectra were collected using one MarqMetrix All-In-One Process Raman Analyzer. The Raman fingerprint region between 870-3096 cm-1 was used to build the model. A SavitskyGolay smoothing filter was applied to remove the random noise and improve the signal to noise ratio. Next, baselines were corrected followed by scattering correction and normalization. In addition, all data were mean-centered before model building. The calibration model was built using the cross-validation strategy of leave-onesample-out. The 3 replicates for the same sample were carried together in this process. After these crucial pre-processing and cross-validation steps were performed, the resulting optimized model with 4 latent variables was selected. Results of the calibration and cross-validation of the model are shown in Table 1. Spectra for the 8-sample validation set were subsequently collected on 10 different MarqMetrix All-In-One Process Raman Analyzers. The chemometric model was applied to the validationset spectra for each MarqMetrix All-In-One Process Raman Analyzer to predict the analyte concentrations. The results discussed below show that high degree of accuracy and precision were maintained for the prediction of all three analytes across all 10 MarqMetrix All-In-One Process Raman Analyzers. Results The development of robust chemometric models requires a significant investment of time and resources. To ensure that this investment provides long term value for our customers, we have demonstrated excellent transferability of chemometric models across numerous MarqMetrix All-In-One Process Raman Analyzers. The correlation plot in Figure 1 shows the predicted vs. reference values for glucose (Fig.1a), lactate (Fig.1b) and glutamine (Fig.1c). Each plot contains an overlay of the predicted values for all 10 MarqMetrix All-In-One Process Raman Analyzers. The precision of the chemometric model across 10 MarqMetrix AllIn-One Process Raman Analyzers provides customers with consistent results. Furthermore, the accuracy of the model can be seen from the results of Table 2. Analyte Model Parameter Glucose Glutamine Lactate RMSEC (g/L) 0.079 0.075 0.146 RMSECV (g/L) 0.095 0.088 0.176 Bias (g/L) 4.79E-05 -1.94E-05 -2.50E-05 CV Bias (g/L) -2.25E-04 -6.45E-04 -4.42E-03 R2 Calibration 0.9995 0.9902 0.9994 R2 Cross-Validation 0.9993 0.9864 0.9991 Table 1. Calibration and cross-validation results for Chemometric model. Results The development of robust chemometric models requires a significant investment of time and resources. To ensure that this investment provides long term value for our customers, we have demonstrated excellent transferability of chemometric models across numerous Thermo Scientific Ramina Process Analyzers. The correlation plot in Figure 1 shows the predicted vs. reference values for glucose (Fig.1a), lactate (Fig.1b) and glutamine (Fig.1c). Each plot contains an overlay of the predicted values for all 10 Thermo Scientific Ramina Process Analyzers. The precision of the chemometric model across 10 Thermo Scientific Ramina Process Analyzers provides customers with consistent results. Furthermore, the accuracy of the model can be seen from the results of Table 2. 0 2 4 6 8 10 12 0 2 4 6 8 10 12 Predicted Glucose (g/L) Reference Glucose (g/L) Glucose Instrument 1 Instrument 2 Instrument 3 Instrument 4 Instrument 5 Instrument 6 Instrument 7 Instrument 8 Instrument 9 Instrument 10 Linear (1:1 Fit) 0 5 10 15 20 0 2 4 6 8 10 12 14 16 18 20 Predicted Lactate (g/L) Reference Lactate (g/L) Lactate Instrument 1 Instrument 2 Instrument 3 Instrument 4 Instrument 5 Instrument 6 Instrument 7 Instrument 8 Instrument 9 Instrument 10 Linear (1:1 Fit) Results The development of robust chemometric models requires a significant investment of time and resources. To ensure that this investment provides long term value for our customers, we have demonstrated excellent transferability of chemometric models across numerous Thermo Scientific Ramina Process Analyzers. The correlation plot in Figure 1 shows the predicted vs. reference values for glucose (Fig.1a), lactate (Fig.1b) and glutamine (Fig.1c). Each plot contains an overlay of the predicted values for all 10 Thermo Scientific Ramina Process Analyzers. The precision of the chemometric model across 10 Thermo Scientific Ramina Process Analyzers provides customers with consistent results. Furthermore, the accuracy of the model can be seen from the results of Table 2. 0 2 4 6 8 10 12 0 2 4 6 8 10 12 Predicted Glucose (g/L) Reference Glucose (g/L) Glucose Instrument 1 Instrument 2 Instrument 3 Instrument 4 Instrument 5 Instrument 6 Instrument 7 Instrument 8 Instrument 9 Instrument 10 Linear (1:1 Fit) 0 5 10 15 20 0 2 4 6 8 10 12 14 16 18 20 Predicted Lactate (g/L) Reference Lactate (g/L) Lactate Instrument 1 Instrument 2 Instrument 3 Instrument 4 Instrument 5 Instrument 6 Instrument 7 Instrument 8 Instrument 9 Instrument 10 Linear (1:1 Fit) Figure 1. Prediction correlation plot across 10 Thermo Scientific Ramina Process Analyzers. Table 2 shows the average prediction error for each analyte on each Thermo Scientific Ramina Process Analyzer and the average error for each analyte across all 10 Thermo Scientific Ramina Process Analyzers. All parameters demonstrate a high degree of measurement accuracy with <0.5 g/L prediction error. Furthermore, in some cases, the prediction error is <0.1 g/L. This level of accuracy is in line with other published results for chemometric modeling3 and is on par with other relevant measurement techniques for bioreactor monitoring. Table 2. Prediction Error (RMSEP) Calculated from Chemometric Modeling Hardware Average Prediction Error (g/L) Glucose Glutamine Lactate Instrument 1 0.18 0.05 0.20 Instrument 2 0.11 0.13 0.40 Instrument 3 0.12 0.15 0.21 Instrument 4 0.18 0.14 0.26 Instrument 5 0.18 0.08 0.40 Instrument 6 0.40 0.11 0.25 Instrument 7 0.13 0.38 0.32 Instrument 8 0.21 0.11 0.26 Instrument 9 0.47 0.15 0.44 Instrument 10 0.09 0.34 0.23 Average Error (across 10 systems) 0.21 0.16 0.30 With this accuracy and precision, controlling glucose concentrations within a bioreactor in the typical 2- 4 g/L range can be readily achieved. Furthermore, through continuous monitoring and feedback utilization, even tighter control is possible leading to improved process and product consistency. Chemometric models developed using Thermo Scientific Ramina Process Analyzer have demonstrated exceptional performance for multi-analyte monitoring in full-scale bioreactor studies. (Real time 0 0.5 1 1.5 2 2.5 0.00 0.50 1.00 1.50 2.00 2.50 Predicted Glutamine (g/L) Reference Glutamine (g/L) Glutamine Instrument 1 Instrument 2 Instrument 3 Instrument 4 Instrument 5 Instrument 6 Instrument 7 Instrument 8 Instrument 9 Instrument 10 Linear (1:1 Fit) Figure 1. Prediction correlation plot across 10 MarqMetrix All-In-One Process Raman Analyzers. A B C © 2023 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. TS_MCS_AN_MMAIO_TRANS_1123 Learn more at thermofisher.com/MarqMetrixAIO Table 2 shows the average prediction error for each analyte on each MarqMetrix All-In-One Process Raman Analyzer and the average error for each analyte across all 10 MarqMetrix All-InOne Process Raman Analyzers. All parameters demonstrate a high degree of measurement accuracy with <0.5 g/L prediction error. Furthermore, in some cases, the prediction error is <0.1 g/L. This level of accuracy is in line with other published results for chemometric modeling3 and is on par with other relevant measurement techniques for bioreactor monitoring. With this accuracy and precision, controlling glucose concentrations within a bioreactor in the typical 2-4 g/L range can be readily achieved. Furthermore, through continuous monitoring and feedback utilization, even tighter control is possible leading to improved process and product consistency. Chemometric models developed using MarqMetrix All-In-One Process Raman Analyzer have demonstrated exceptional performance for multianalyte monitoring in full-scale bioreactor studies. (Real time metabolite monitoring using the MarqMetrix All-In-One Process Raman Analyzer and the Thermo Scientific™ 500L HyPerforma™ Dynadrive™ SingleUse Bioreactor (S.U.B.) MarqMetrix All-In-One Process Raman Analyzer Real time App note Conclusion Customers will benefit from generating chemometric models which are transferrable between multiple MarqMetrix All-In-One Process Raman Analyzers. The ability of the MarqMetrix All-InOne Process Raman Analyzer to utilize the same chemometric model across multiple units provides users with the confidence of measurement accuracy and precision. Advanced signal processing and model optimization may be employed to further increase the level of prediction performance. This example simply provides a benchmarking reference for the development of chemometric models using MarqMetrix All-In-One Process Raman Analyzers. References 1. FDA. Guidance for Industry: PAT—a Framework for Innovative Pharmaceutical Development, Manufacturing, and Quality Assurance, via FDA Website. 2004.. https://www.fda.gov/ media/71012/download. Accessed October 17th 2022. 2. Zhang, Lin, et al. “Uniform design applied to nonlinear multivariate calibration by ANN.” Analytica Chimica Acta 370.1 (1998): 65-77. 3. Buckley, Kevin, and Alan G. Ryder. “Applications of Raman spectroscopy in biopharmaceutical manufacturing: a short review.” Applied spectroscopy 71.6 (2017): 1085-1116 Average prediction error (g/L) Hardware Glucose Glutamine Lactate Instrument 1 0.18 0.05 0.20 Instrument 2 0.11 0.13 0.40 Instrument 3 0.12 0.15 0.21 Instrument 4 0.18 0.14 0.26 Instrument 5 0.18 0.08 0.40 Instrument 6 0.40 0.11 0.25 Instrument 7 0.13 0.38 0.32 Instrument 8 0.21 0.11 0.26 Instrument 9 0.47 0.15 0.44 Instrument 10 0.09 0.34 0.23 Average error (across 10 systems) 0.21 0.16 0.30 Table 2. Prediction error (RMSEP) calculated from Chemometric modeling. MarqMetrix All-In-One Process Raman Analyzer Raman spectroscopy Introduction Manufacturing processes can be extremely complex, especially biopharmaceutical manufacturing. With processes that rely heavily on living organisms to generate products of interest, such as those within the Biopharma industry, complete understanding of the process is essential to success, because the more you know about the process, the more control you can exercise. The primary problem process analytical instruments and technologies have sought to alleviate has always been a lack of visibility into the underlying reactions. What you see, and also understand, you can control and optimize. Subtle variations in the environment can have a huge impact on yield and quality. The consequences can range from failed batches and inefficient use of resources to products that don’t meet quality specifications, and limited data to help make improvements over time. Ultimately, patient safety and stakeholder interests are at risk. We are dedicated to making instruments that help our customers gain greater control of their processes that involve chemistry. One of the most powerful and advantageous analytical technologies that we use is Raman spectroscopy. It’s extremely accurate and versatile, generates very rapid results, covers a multitude of functions, and is non-destructive to whatever sample or substance is being analyzed. Application note Until recently, Raman Spectroscopy required a well-trained technician to operate, and demanded complex, bulky, and expensive equipment that also made it at times unsuitable for in-line or field use. We have introduced the Thermo Scientific™ MarqMetrix™ All-In-One Process Raman Analyzer, a compact, easy-to-use, reliable, and affordable system that leverages the power of Raman spectroscopy to measure key variables in a process so that they may be controlled. The MarqMetrix All-In-One Process Raman Analyzer also makes Raman technology more accessible to non-experts through its simplified user interface. Since it is also small, it is uniquely qualified to solve a common problem faced by those who use chemistry to make things, i.e., portability and ease of use whether in the field or in the process development space. What is RAMAN spectroscopy? The underlying technology, Raman spectroscopy, is not new; it has been around since the 1950s. What Raman instruments have lacked until now is the size, reliability, ease of use, and affordability that enables manufacturers to integrate Raman spectroscopy into their production processes where it can be instrumental in improving efficiency and quality. Before MarqMetrix All-In-One Process Raman Analyzer, integrating Raman technology into a manufacturing process was problematic. There were challenges with the hardware; it was difficult to use and required a scientist on staff to operate and maintain it. Reliability and cost were also issues. Better hardware was needed, as well as a simplified, more user-friendly interface. These features and advantages are now realized in MarqMetrix All-In-One Process Raman Analyzer, but what exactly is Raman spectroscopy all about? Non-destructive analysis and process monitoring The technology begins with a laser that is directed at a substance or sample through a fiberoptic cable with a probe at the end. The energy from the laser light causes covalently bonded molecules in the substance to vibrate and the light from the laser may scatter elastically (the same laser energy is released as what caused the molecule to vibrate) or inelastically (some of the laser energy is absorbed by the molecule and a lesser amount of energy is released than what caused the molecule to vibrate). Some of this inelastically scattered light makes its way back into a detector within the MarqMetrix All-InOne Process Raman Analyzer. The detector collects and interprets the light scattered from the sample to generate a “picture” called a Raman spectrum. What makes this technology powerful is that the Raman spectrum of a molecule is unique, and for that reason it’s sometimes referred to as its molecular fingerprint. Just as fingerprints may be used to identify people, we use the Raman spectrum to identify a given substance because the fingerprint can tell us not only qualitatively what something is, but also quantitatively how much there is of it- one may determine both the identity and concentration of a given analyte of interest. The non-destructive nature of Raman spectroscopy provides a tremendous value-add. This is very important because it can be integrated directly into a production line to measure and analyze on a continuous basis, serving as a process monitor. This is what is called in-line or online measurement. Raman technology is also fast, and most substances can be measured in a matter of seconds or less. Versatile, precise, providing a wealth of information Additionally, the Raman spectrum provides us with a great deal of information about the substance being tested; every one of the peaks in the Raman spectrum tells us something unique about what we’re testing, providing us with many opportunities to identify what we’re looking for. This is extremely important when measuring in a bioreactor, for example, containing many molecules of different types. The next advantage is that there’s a direct and linear relationship between the concentration of a given substance and the intensity of the peaks in the spectrum. This means that building quantitative models is easier. With a relatively small number of samples, we can build a model that accurately predicts concentration across our range of detection. We can also measure substances in all forms, whether they’re solid, liquid, gas, powder, or slurry. A final advantage is that unlike other forms of spectroscopy, water does not distort Raman measurements; therefore, we can see clearly what’s happening in the aqueous solution like that found inside of a bioreactor. Competitively, MarqMetrix All-In-One Process Raman Analyzer is better suited for inline and online measurements in aqueous-based solutions than competing technologies, but perhaps its greatest advantage is that it is non-destructive. Process control There are Raman applications that span the entire biopharma manufacturing process. Verifying the integrity of raw materials is the first step where MarqMetrix All-In-One Process Raman Analyzer can help ensure that the process gets off to a good start. In practice, we’ve observed notable variations in the chemical composition of growth media, variations significant enough to have an impact on yields and cycle times. It’s important to be able to detect changes in the material in real time. For example, customers want to know what’s happening in the bioreactor so that they can make adjustments if necessary. This is true regardless of whether or not manufacturing is done via batch or continuous flow, and MarqMetrix All-In-One Process Raman Analyzer works in either scenario. Do the cells have the right amount of glucose? Are too many secondary metabolites building up? Are the cells beginning to produce the product of interest? How much product has been produced, and does it have the right characteristics? MarqMetrix All-In-One Process Raman Analyzer enables answers to all of these questions, and because the unit can interface with third-party control systems, adjustments can be made in near real-time to optimize conditions. Finally, when capturing finished product downstream, manufacturers need to know exactly when and how much protein is coming out of a purification column, for example. These are just a few examples of how MarqMetrix All-In-One Process Raman Analyzer helps achieve higher efficiency and quality in chemistry-dependent manufacturing. Robust yet simple operation MarqMetrix All-In-One Process Raman Analyzer is easy to use. A technician with no prior Raman experience can typically begin taking measurements within 15 minutes of removing the instrument from the box. Facilities across the country can use MarqMetrix All-In-One Process Raman Analyzer to take hundreds of measurements daily without the need for any scientific staff to maintain and calibrate the instrument. MarqMetrix AllIn-One Process Raman Analyzer is compact. With a footprint less than one square foot and three inches tall, MarqMetrix All-In-One Process Raman Analyzer is sized to be placed at or near the point of measurement and with no moving parts other than its cooling fan, MarqMetrix All-In-One Process Raman Analyzer is designed to be reliable and stable. Up time exceeds 99% and calibrations performed in the factory remain accurate for years. By being easy to use and reliable, MarqMetrix All-In-One Process Raman Analyzer is also less costly to install and operate over time, with a very low cost of ownership. Software built to simplify chemometrics Maximize the power of the MarqMetrix All-In-One Process Raman Analyzer with Thermo Scientific™ Lykos™ PAT Software, a remote control and monitoring software designed to simplify the complexities of chemometrics. Lykos PAT Software facilitates data acquisition and analysis by displaying, storing, and exporting bioprocess monitoring data in real-time, enabling remote control and monitoring of the MarqMetrix All-In-One Process Analyzer and providing an intuitive workflow to streamline processes for nonRaman spectroscopists and Raman experts alike. A wide range of easily-swappable probes are available The MarqMetrix All-In-One Process Raman Analyzer can be adapted to almost any R&D production or process system. It features a Thermo Scientific™ MarqMetrix™ Fiber Head and Thermo Scientific™ MarqMetrix™ BallProbe™ Sampling Optic that fits any need due to a wide array of easily interchangeable probes. Swapping the probe is easy; simply unscrew the fastener, remove the probe, swap in another, and tighten it down again. Our tapered fitting allows for easy indexing, requires no alignment, and there’s no need to recalibrate the system. The MarqMetrix All-In-One Process Raman Analyzer lineup of standard fiber probes allows measurement of compounds in any form, e.g., solids, liquids, gas, slurries, pastes, and gels. It can measure by contact or immersion with its performance half inch ball probe and/or standard ball probes which range in size from 1/2” down to 1/8”. For higher temperatures and harsh chemicals, there are process ball probes that are gold sealed and temperature safe Learn more at thermofisher.com/MarqMetrixAIO © 2023 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. TS_MCS_BROCHURE_MMAIO_2_1123 to 350 degrees Celsius. MarqMetrix BallProbe Sampling Optics are also available in bioreactor-specific versions with a seal and nut to allow for immersion in a reaction vessel, be it a dedicated stainless-steel bioreactor or a single-use bioreactor. These Thermo Scientific™ MarqMetrix™ Bioreactor BallProbe™ Sampling Optics are autoclavable, either while attached to the bioreactor, or on their own when cleaning and sterilizing the ball probe separately. Integration into a high-pressure flow system is simplified with patented flow cell technology operable at 2500 psi. Custom probes are also available. At Thermo Fisher, we’re realizing the full potential of Raman technology by building better instruments, and we’re all about making Raman spectroscopy more accessible to help our customers control chemistry. We’re opening up new applications for Raman and increasing the returns on existing applicationsreturns that come in the form of faster cycle times, higher yields, and better quality. For research use only. Not for use in diagnostic procedures. For current certifications, visit thermofisher.com/certifications © 2024 Thermo Fisher Scientific Inc. All rights reserved. All trademarks are the property of Thermo Fisher Scientific and its subsidiaries unless otherwise specified. MCS-CM1177-EN 11/24 Learn more at thermofisher.com/marqmetrixaio Single-use bioreactor solution The DynaDrive Bioreactor Platform is a high-performance, single-use bioreactor system designed for optimized mixing, mass transfer, and scalability in biopharmaceutical production. Process Raman analyzer The MarqMetrix All-In-One Process Raman Analyzer is a compact, powerful spectrometer for nondestructive real-time analysis of critical process parameters directly in-line with processes. It can be used for direct process monitoring and control in upstream and downstream bioprocessing. Learn more at thermofisher.com/dynadrive