Cultivated meat has the potential to revolutionize the way we obtain protein. Producing large quantities of cultivated meat will require efficient and reliable industrial production. Today, industrial cell culture relies on an army of operators taking samples from bioreactors, testing those samples, consulting with subject-matter experts on what to do, and then manually changing bioreactor controls to optimize operations. This process is cumbersome, timely, expensive, and increases the risk of contamination since every time the bioreactor is opened there is a chance of contamination and loss of product.
Enter Process Analytical Technology (PAT) - a relatively recent framework focused on designing, analyzing, and controlling manufacturing processes through continuous monitoring and analysis of process parameters and quality attributes. PAT provides a quick, frequent, in situ and non-destructive alternative to the current measurement methodology. The primary goal of PAT is to enable real-time process monitoring, control, and optimization, resulting in more consistent product quality, increased process efficiency, and reduced manufacturing costs. PAT includes several technologies and techniques such as spectroscopy, chromatography, microscopy, and sensors, which are used to collect data in real-time. The data collected is then analyzed using various statistical tools, such as multivariate data analysis, to identify trends, patterns, and potential deviations in the process parameters.
One of the most powerful levers within PAT is Raman spectroscopy, which has gained popularity as a method for real-time, in situ measurements of quality attributes. It enables multiparameter analysis by providing comprehensive information on cellular parameters, metabolic activity, and metabolite concentration, contributing to a detailed real-time analysis of cell culture dynamics. This piece explores the value of using Raman spectroscopy in the production of cultivated meat through a case study on using Raman to determine the metabolite concentration in a cultivated meat process.
Raman spectroscopy is one of the most versatile tools for monitoring bioprocesses. Raman spectroscopy looks at how the visible or near-infrared regions of the electromagnetic spectrum impacts molecular vibrations - the movement of functional groups within a molecule. Using one Raman sensor, numerous metabolites can be quantitatively and qualitatively measured. Raman probes transmit raw signals that require non-trivial statistical data analysis, including the building of multivariate models, to provide any meaningful insight. One of the common challenges for real-world application of this technology is in isolating a single metabolite given the diversity and dynamic nature of the cell culture environment.
To demonstrate the capability of Raman, an All-In-One inline Raman instrument from MarqMetrix was implemented in an Ark 10L bioreactor using a proprietary cell line and media to monitor glucose and lactate levels. The bioreactor and in situ probes, including Raman, were sterilized using an autoclave prior to the introduction of media solution and cell inoculation. Temperature, DO levels, pH, and biovolume were also measured via the implementation of additional inline PAT sensors during the multi-day run.
The Raman instrument settings consisted of dark spectrum subtraction, an appropriate integration time for the target cell culture environment and co-add scans resulting in a spectrum being acquired every 5 minutes. To properly train the Raman model, reference concentrations of glucose and lactate were manually sampled from the bioreactor and analyzed using an offline instrument - Vi-CELL MetaFLEX Bioanalyte Analyzer from Beckman Coulter. Samples were taken one or two times each day for the duration of the run.
The Raman spectra underwent elaborate multivariate data analysis to correlate the raw spectra to the offline analyte concentrations using Solo from Eigenvector Research. The collected spectra were randomly split into a calibration set to build a multivariate model and a test set to evaluate the model. Prior to developing a model, the spectral data were pretreated by performing one or more of the following: (1) selecting wavelength regions that are specific to the analyte of interest, which also allows for the removal of unnecessary regions, (2) smoothing the spectra to reduce noise, (3) scatter correction to remove influences from physical variations in a sample, ensuring only the chemical changes are highlighted, (4) applying derivatives to improve signal-to-noise ratios and enhancing the peak(s) of interest. Several partial least square (PLS) regression models were developed for each analyte by attempting different approaches of assigning the limited offline reference data to a large number of Raman spectra collected. Each model’s performance was evaluated using standard metrics for calibration (RMSEC), cross-validation (RMSECV), prediction of test spectra (RMSEP), Hotelling’s T2, sample residuals (Q), and fit (R2). For the purposes of illustrating the quality of the raw Raman data, minimal pretreatment was performed on the spectra prior to model development for glucose and lactate. The model for glucose that performed most favorably with respect to its prediction ability is illustrated in Figure 1, and the model for lactate is illustrated in Figure 2. The glucose and lactate amounts in Ark’s 10L bioreactor over the duration of the cell culture process as predicted by Raman in comparison to the offline measurements are well aligned, especially for lactate.
The initial models are only the first step towards harnessing the full power of Raman spectroscopy. A finely tuned Raman model can improve prediction abilities in the low and high end of the range, and additional training data from varying conditions will improve model robustness.
A well-trained Raman model can provide a real-time and non-invasive method for monitoring metabolite concentration in a bioreactor; monitoring these concentrations over time enables monitoring of production, consumption, and conversion rates; together these provide insight into the cell growth and media micro-environment and can be used to maintain the health of the cells and maximize yield. With the use of multivariate models and data analysis, the Raman instrument can provide accurate predictions of glucose and lactate levels in real-time, allowing for timely adjustments to the manufacturing process and enabling closed-loop control. A more refined control strategy can increase the yield per batch and ultimately leads to more efficient and cost-effective production of cultivated meat.
It is also possible to extend the use of Raman for modeling beyond metabolites to implicitly-correlated parameters, such as cell density, viability, osmolarity, and other attributes. Models for these parameters are not based on direct electromagnetic radiation interactions; instead, predictions can be made based on statistical correlations. The addition of these parameters further expands the versatility of Raman for process monitoring, control, and optimization purposes.
