The intensification of upstream bioprocessing sounds daring, as though it were equivalent to flooring a car’s accelerator pedal and pushing the tachometer into the red zone. But with the right engineering and well-timed responses to performance metrics, upstream bioprocessing’s engines needn’t blow up. Instead, they can run as reliably as the endurance race cars that roar around the track at Le Mans.
These engines include the platforms described by the experts who share their insights in this article. For example, one expert describes a platform for industrializing allogeneic cell therapy. Another describes a machine-learning-assisted process for selecting high-performance clones. Another describes microbioreactors that faithfully model large-scale perfusion processes. Yet another describes a streamlined approach to process analytical technology that relies on Raman spectroscopy data.
With options such as these, biomanufacturers may boost production or, as one of our experts puts it, attain a “state of readiness.” In the latter case, they may recall a line from a song by James Taylor, the one that goes, “Hurts my motor to go so slow.”
An allogeneic approach
Cell and gene therapies are finally emerging as realistic options for various diseases, including diseases for which no other therapies are available. Yet access to these cell and gene therapies is severely limited. They are, for the most part, autologous, meaning that they rely on manipulating cells that come from each individual patient. And these therapies are exceedingly expensive, often upward of a million dollars per patient.
Prospects for making cell and gene therapies more accessible were discussed by Nuno Fontes, PhD, head of global biologics development, Bayer Pharmaceuticals, in his keynote address at the BioProcess International US West (BPI West) meeting, which was held last March in San Diego. In his remarks, he emphasized that cell therapies will become more accessible if technology platforms are deployed that accelerate the industrialization of allogenic cell therapies.
“Right now, cell therapies are manufactured in a very ‘researchy’ way, by the processes that were developed at universities where the science was originally discovered,” he noted. “So, the raw materials that they are using are not what we call GMP.” The methods are manual, and the equipment and technologies that are used were designed for previously established processes such as food or large-molecule production rather than cell therapy production.
Fontes insisted that cell therapy production needs its own platform. Then, while elaborating on this point, he explained that the platform should have five elements:
- Systems for monitoring processes and ensuring quality control.
- Modular, flexible, closed, and automated equipment and facilities. (These elements should be purpose-built and capable of covering the vast majority of assets on demand.)
- Digitalized processes. (Such processes should allow a digital twin to be executed at any location.)
- A chemistry, manufacturing, and controls (CMC) ecosystem that encompasses everything a manufacturing setup needs to take a product from clinical to commercial development. (In such an ecosystem, information technology, artificial intelligence, and knowledge management technology are integrated, facilitating quality control, regulatory compliance, and supply chain management.)
- Highly specialized personnel. (This element cannot be ignored because, as Fontes remarked, “we’re still not in the full age of robots.”)
According to Fontes, having these elements in place will allow for a state of readiness. “We won’t have to have to start from scratch every time,” he stressed. “We will already have something that is pretty much ready to deploy.” He added that having the industrialized platform from the start would make it easier for a biomanufacturer to scale up/scale out its processes while maintaining product quality.
Artificial intelligence in clone selection
There is push in biomanufacturing to increase the production of large molecules by any given bioreactor as a way to help lower production costs. Part of push includes transitioning from traditional fed-batch processes toward intensified processes, such as processes that accommodate high inoculation densities.
Intensified processes have resulted in significantly higher titers from many clones. “Yet there is quite a spread across different clones,” observed Brandon Downey, PhD, associate director, high-throughput and process modeling R&D, Lonza, when he spoke at BPI West. The challenge, he added, is identifying which clones will respond to the new intensified process.
During clone selection, when the number of potential clones is reduced from thousands to perhaps the top five or so, biomanufacturers usually start by using a traditional fed-batch process before advancing to a more expensive scaled-down model. “But that’s actually not a very good prediction anymore of how that clone is going to do in an intensified process,” Downey noted. He considered whether machine learning could use the data already generated to predict how clones would behave.
Lonza ran a host of clones in parallel in a 250 mL intensified fed-batch process mimic, comparing the “truth data” generated with the previously generated data, to train the artificial intelligence model and pick up a pattern of attributes and hyperparameters. In the proof-of-concept presented at BPI West, the top three potential clones according to a machine-learning-assisted process turned out to be the actual top three performing clones. The titer for the best clone was 1.65 times greater than that for the baseline fed-batch-selected clone.
“We need to think about selecting clones that actually will respond well to that process intensification, and we think machine learning is a good tool to do that,” Downey reiterated. “There are actually a lot of data sources that you get almost for free coming out of the cell line selection process that can be used to predict how clones are going to do in intensified processes.”
True to perfusion from the start
If clone selection and development is supposed to identify clones that will do well in an intensified process, starting with a traditional fed-batch process and later transitioning to an intensified process may be suboptimal. Clones that are optimized for a fed-batch process—that is, optimized to produce the highest concentration of drug product in the final harvest—may not display the optimal characteristics required for perfusion processes.
This possibility was raised in the BPI West presentation delivered by Vince Balassi, senior scientist, cell culture upstream bioprocessing R&D, MilliporeSigma. He also posed questions that biomanufacturers should ask about clones. One of these questions was, “Do they have the right combination of growth and specific productivity?” (According to Balassi, specific productivity, the amount of drug product that is made on a per cell basis, often does not correlate with volumetric titer.) Another question was, “Are they able to handle the stresses of these perfusion processes as well?”
Balassi did a head-to-head comparison of how traditionally derived fed-batch clones compared with those derived in a simulated (“semicontinuous”) perfusion process, when brought into a true perfusion process. Clones from the latter were seeded at high density in deep-well plates, with daily media exchange. Each process used process-appropriate medium. Perfusion-derived clones were found to outperform fed-batch-derived clones in key productivity metrics when brought into true perfusion bioreactor assays.
Yet, as Balassi said at BPI West, it is necessary to remember that while simulated assays are excellent high-throughput tools to help identify top candidates, they lack the scalability and process controls found in the true perfusion systems. One solution, he said, is MilliporeSigma’s Mobius Breez, a 2 mL single-use true profusion microfluidic bioreactor that allows control of key parameters including pH and dissolved oxygen.
“It’s much more representative of what you see when you scale up,” he asserted. “It’s a better predictor of how your clone will perform, of how much drug you’re going to get, and of how to balance growth and productivity.” He added that most scaled-down bioreactors require large amounts of resources and cells and media, and that in comparison, Mobius Breez offers much higher throughput and a much more reasonable option for perfusion on a larger number of clones.
According to Balassi, upstream process intensification is a holistic process. He emphasized that “cell line development is [just] the very beginning of that intensification work.”
Raman PAT for the masses
Intensified processes require greater cell densities and higher glucose feed rates to ensure that they can achieve higher levels of production. Such processes also require tight monitoring and control to ensure that critical quality attributes (CQAs) and critical process parameters (CPPs) are maintained.
An effective way to meet these requirements is to use process analytical technology (PAT), suggests Graziella Piras, PhD, senior director, strategic marketing, 908 Devices. At the Bioprocessing Summit scheduled to take place in Boston this August, Piras is scheduled to deliver a presentation titled, “PAT for mAb Production: Real-Time Upstream Monitoring.”
for bioprocess control without imposing modeling and development requirements.
According to Piras, PAT allows data to be used for the feedback control of the monitored processes. “PAT lets you look into the bioreactor,” she explains. “It shows how the cells are consuming nutrients, how they’re producing toxic metabolites, and how fast they’re growing, in real time.”
Traditional PAT approaches include at-line and off-line analyses. With such analyses, bioprocess parameters may be determined though the analysis of Raman spectroscopy data and the development of chemometric models. However, collecting enough data may require multiple bioprocess runs over weeks or months, and data analysis and model development may require a dedicated team of experts. And even if these requirements are met, the result, Piras says, may be underwhelming: a black-box model that is specific to one cell line, one process, and one scale.
At the August Bioprocessing Summit, Piras will discuss 908’s Maverick inline Raman-based PAT solution. It is a de novo approach based on media chemistries and device physics (that is, the physics of the spectrophotometer and probe), not on empirical data. She says that the approach is generalizable across subculture media for CHO and HEK293 cells. Maverick currently can measure glucose, lactate, and biomass right out of the box, without needing to be trained. The company has plans to release other parameters. “Our mission,” Piras declares, “is to democratize analytical methods and put them into hands of bioprocess scientists.
