Regulatory approval is only one marker of success for biotherapeutics. Biotherapeutics must also be commercially feasible. Too high a price tag can lead to limited availability to the targeted patient population and subsequently impact the payback for the developer’s investment. The manufacturer has no incentive to produce a treatment that few intended recipients can afford.

One way to address the cost issue is to streamline upstream process development and manufacturing. This possibility was discussed at Bioprocessing Summit Europe 2024, a Cambridge Healthtech Institute event that was held last March. At this event, presenters described advanced forms of modeling that can maximize the data received from a reduced number of runs, substantially reducing costs and facilitating the development of in silico digital twins. Ultimately, this tactic can improve the simulation of unit operations and reveal the impact of and the relationships between different parameters.

Another related tactic is the use of instruments that are compatible with or provide for improved monitoring and control. Instruments of the former type include state-of-the-art bioreactors; instruments of the latter type, Raman spectrophotometers. Such instruments can help improve the health and productivity of cell cultures and ramp up nondisruptive data collection during the cultivation process. The overall goal is to get more life-altering biotherapeutics to patients who need them, regardless of their socioeconomic status.

Smart experiments

“We are a workflow company,” said Mark Duerkop, PhD, the CEO of Austria-based, Novasign. “If you really want to accelerate bioprocess development, you need to think about how to plan, execute, and learn from smart experiments in a clever way. Then take these data and generate knowledge. This requires workflow changes, which is not trivial, but the potential is huge.”

The company’s hybrid modeling approach consists of two parts that work together—the machine learning portion, which directly learns from data, and the mechanistic portion, which is a “white box” that contains prior knowledge such as differential equations or mass balances. Publications demonstrate that smart experiments show value for upstream and downstream operations such as optimizing microbial fermentation, cell culture, filtration, or chromatography. The workflow reduces up to 70% of experiments needed.

“Modeling is a great way to create knowledge,” Duerkop advised. “But if the way data are created remains unchanged, the gains are minimal. To accelerate bioprocess development, you must think about the entire workflow and the goal of the modeling.” Process modeling has limited effect on whether a drug performs as expected, but it can impact commercial feasibility significantly, especially with respect to cell and gene therapy processes.

Novasign leverages design of experiments (DOE) methods and in silico modeling methods to perform smart experiments and address the data problem in bioprocess development. DOE permits an iterative approach to process optimization. A limited initial number of experiments are put into the design space in such a way as to maximize the difference between them. Then an algorithm is used to learn from the results and generate suggestions for new experiments.

For late-stage development process characterization, intensified DOE dynamically suggests changes in process parameters during the bioprocess. In silico modeling is then used instead of physical experiments to investigate the whole design space.

“In silico modeling will never eliminate all experiments, but it can reduce them to a minimum,” Duerkop emphasized. “The ideal scenario would be to have one model for the entire upstream and downstream process concatenated together. We are developing an end-to-end model that incorporates the effect on each unit operation on the following operations in a time-resolved manner.”

Leveraging artificial intelligence

Finland-based AnalysisMode has developed a new artificial intelligence (AI) approach, Neuroevolution, that can simulate cell growth in the advanced therapy upstream space. “SimCell, our flexible, web-based software, has in silico models for cells, antibodies, and viral vectors (such as lentivirus and adeno-associated virus vectors), and it will expand to other modalities,” said Belma Alispahic, co-founder and CSO of AnalysisMode. “SimCell is used to design experimental setups and to optimize and characterize processes to explore relationships between different parameters.”

AnalysisMode Alispahic3
AnalysisMode has developed SimCell, a web-based virtual bioreactor. It leverages AI to facilitate the design of experimental setups, the optimization and characterization of processes, and the exploration of relationships between different parameters. This image shows a SimCell–Google Sheet integration.

The tool allows scientists to adjust processing parameters and run tests of their experiment setups in silico. “The goal is to have our AI software connected in real time to a scientist’s platform, for example, a LIMS system, to access existing data,” Alispahic explained. “This allows the AI software to learn the molecule and the process. Then the AI software can make decisions or implement changes.”

For calibration of the AI software, the end user provides historical process and laboratory run data to create specialized custom models for the target process. The AI software automatically identifies the interaction patterns between different parameters, assesses their importance, and determines their timing in the culture process, resulting in a digital twin. The technology provides knowledge transfer between different processes, molecules, and scales by employing state-of-the-art proprietary technology to create AnalysisMode’s General Purpose Master Model (GPM²) for bioprocessing.

According to Alispahic, bioprocessing as an industry does not have standardized or large datasets. This deficiency is the largest challenge for AI applicability. AnalysisMode’s solution creates a model from just five or six laboratory runs, empowering scientists to start their AI journeys.

“Our technology is a ‘white box’ approach,” Alispahic explained. “It utilizes breakthrough technologies in neurosymbolic machine learning, AI-informed physics models, and evolutionary neural network algorithms, allowing us to capture mechanistic equations of the biology, chemistry, and physics in logical expressions that both machines and humans understand. Every AI decision is traceable and explainable for transparency. The scientist makes the final informed call.”

“AI is science and not just technology,” Alispahic continued. “Our models make scientific and biological sense. Our role is to help people move forward and accelerate drug development, shortening process development time and reducing its cost so that these therapies can reach the market at prices that are attainable for the general public.”

Next-generation bioreactors

Traditional modalities, such as monoclonal antibodies, and newer ones, such as genomic medicines, all need more advanced cell culture processes. These processes, however, may pose their own challenges. For example, they may rely on high cell densities, which result in very high demand for oxygen mass transfer. Or they may rely on shear-sensitive cells such as stem cells.

To support demanding next-generation cell culture processes, Cytiva has launched the Xcellerex X-platform bioreactors. X-platform bioreactors in 50 and 200 L sizes are currently available, and a 2,000 L X-platform bioreactor is scheduled to be released soon. All incorporate single-use technologies, and all are designed to maintain the mass transfer properties for oxygen and carbon dioxide as well as mixing times. In addition, they all incorporate the Cytiva Bioreactor Scaler, which facilitates process transfer from other bioreactor platforms, such as legacy Cytiva Xcellerex bioreactors, laboratory-scale bioreactors, or rocking bioreactors.

Cytiva's X-platform bioreactors
Cytiva offers X-platform bioreactors to simplify single-use upstream bioprocessing operations. X-platform bioreactors, initially available in 50 and 200 L sizes, are provided with Figurate automation solution software and can increase process efficiency through ergonomic improvements, production capabilities, and simplified supply chain operations. The bioreactors also work with the Cytiva Bioreactor Scaler to determine the optimal target settings for scaling without trial and error.

Additional features include modular design for easier incorporation in cleanroom or pilot facilities, application-ready bag designs for easier process optimization, and standardized bag assemblies for shorter delivery times.

“We have leveraged computational fluid dynamics (CFD) to significantly accelerate the development of the fully characterized bioreactors,” said Andreas Castan, PhD, upstream strategic technologies leader, Cytiva. “CFD modeling was used to determine the optimal placement of the drive unit, baffles, and liquid addition lines.

“The CFD models (both one- and two-phase models) were validated with experimental data, for example, of the mass transfer or mixing time. Furthermore, the models create insight into parameters that cannot easily be measured—such as the velocity vectors in the bioreactors, the shear regime, the Kolmogorov eddy length—over the whole range of operating conditions.”

The design allows the user to dial in the shear regime as well as the mass transfer properties to support low- or high-cell-density processes, as well as shear-sensitive or robust cells. “We believe in the benefit of digital tools to accelerate workflows,” Castan emphasized. “The applied in silico tools facilitated the design of these next-generation bioreactors and will accelerate and de-risk bioreactor operation, scale up, and technology transfer.”

Nondisruptive monitoring

Quality assurance processes increasingly use Raman spectroscopy for raw material release prior to GMP use. In addition, the technology offers real-time, nondestructive monitoring of bioprocesses by analyzing the spectral information of cell culture, allowing monitoring of nutrient consumption, metabolite accumulation, and desired product generation.

Integration into bioprocessing equipment enables in-line monitoring within the bioreactor to offer rapid feedback without disruption and provide real-time critical process parameter (CPP) and critical quality attribute (CQA) data for process control.

“Nutrient concentrations and by-product accumulation can be tuned on demand to optimize cell growth and product yield,” said Jérémy Peyrol, upstream innovation expert, MilliporeSigma. “In downstream applications, Raman may demonstrate an advantage in monitoring protein-related indicators or attributes.”

“While Phase I processes can be modified and improved in Phase II, it is imperative to maintain comparable or equivalent product quality,” Peyrol advised. “Implementing process intensification in upstream operations necessitates meticulous quality monitoring during optimization for maximum productivity while ensuring maintenance of desired quality attributes.” Raman monitoring capabilities aid in reducing final release time, consequently minimizing risks associated with product quality.

MilliporeSigma’s comprehensive solution consists of the ProCellics Raman Analyzer with Bio4C PAT Raman Software. (The latter includes the Bio4C PAT Chemometric Expert module.) Also, extensive training programs and expert support for Raman data analysis and modeling are available. To meet industry requirements, MilliporeSigma supports “one-batch calibration” and stray light testing in ambient light environments.

Traditional calibration methods often involve time-consuming and resource-intensive processes, resulting in limitations in efficiency, accuracy, and technology adoption. The one-batch calibration approach can be applied to any cell line, clone, media, and feed conditions. “It eliminates the need for multiple calibration batches,” Peyrol stated. “It is versatile and customizable, and it facilitates technology transfer to other sites for efficient implementation.”

To assess the impact of ambient light on the Raman signals, stray light testing is conducted with an innovative Raman probe, and advanced software is used to reduce background noise. This approach mitigates the impact of diverse lighting environments and facilitates the effective monitoring of CPPs and CQAs.

Previous articleUpstream Bioprocessing: More Intense and More Stable
Next articleCRISPR 2.0: Kinder, Gentler, More Powerful Gene Editing