Special Report: Cell Biology
Expanding bioprocess analytics to ensure quality
By Randall C Willis
Standing across from his NMR spectrometrist boss, a research technician walks his lab mates through his most recent efforts to purify one of the proteins involved in a signal transduction cascade. The tech shows gels, HPLC profiles and early 2D NMR spectra provided by one of the post-docs.
“So, what’s the average protein yield?” the biophysicist boss asks.
“It’s the total yield of all experiments divided by the number of experiments,” the technician replies, unflappably.
The boss was unimpressed with the answer, but the technician knew that it was the best he could really give. A numerical answer was meaningless, he knew, because the proteins were produced in E. coli, and living organisms were notoriously fickle.
Working in an academic research lab, the technician knew he had some wiggle room in being imprecise. Had he worked in a more commercial setting, however, that answer would be significantly less acceptable.
Whether the final product is produced by living organisms—microbes, yeast, mammalian cells—or is the cells themselves, the need to know what and how much comes out the other end of the bioproduction line is vital to determining whether a project is financially viable and profitable.
“I think manufacturing of these products is underappreciated, and not just at the early stage,” reflects Liz Csaszar, development manager at the Centre for the Commercialization of Regenerative Medicine (CCRM). “You look at the CAR Ts that are on the market today, and one of the biggest challenges they have is manufacturing.”
“They are dealing with releasing products that are out of spec,” she adds. “And these are the most advanced products that we have in the field.”
The complexity of the challenge highlights to her that this is not going to be solved once and be done.
“The nature of these products is that manufacturing will be the bane of your existence forever,” she rues.
CCRM colleague and VP of Commercialization Jana Machan echoes Csaszar’s thoughts.
“I heard a speech at a major conference,” she recounts. “I won’t name the person, but they said we are in the pre-Model T days.”
“It was interesting, because I was new to the field and all I had heard was that this was leading-edge and way out there,” she continues. “To hear that juxtaposed with pre-Model T stage.”
All is not lost, however, as this is very much the reason that CCRM has brought on the staff and resources it has.
“We take on fee-for-service clients to help them automate the process, close the process, scale the process, optimize each of their steps,” Machan explains.
She also points to a supportive regulatory environment, suggesting they are open and quite savvy of the current state of affairs.
“Our guideline doesn’t quite fit,” she says, “then we need to do something, so let’s talk about what that should be.”
“That’s very refreshing and very encouraging,” Machan continues, “because it means we have partnerships in those areas that should speed the ability of these things to come to market.”
Csaszar also points to the pivotal role being played by tool providers.
“There are a lot of close interactions there,” she says, “a lot of semi-customization happening that are helping these niche cell types and these processes move forward.”
For CCRM specifically, GE Healthcare has been a critical partner in supporting and supplying the organization’s Centre of Advanced Therapeutic Cell Technologies facility. And central to this facility are banks of bioreactors and a large robotics station that, among other things, are used to screen media recipes to optimize cell growth.
The organization also recently opened its Centre for Cell and Vector Production facility to facilitate GMP practices throughout the cell and gene therapy development process. The goal is to help produce product for use in early-phase clinical trials.
So, how do we move on from the Model T days?
Fifteen years ago, the FDA introduced what is colloquially known as the Process Analytical Technology (PAT) Initiative, an effort by the regulatory body to minimize risks to public health by establishing a framework for innovation in pharmaceutical development, product manufacturing and quality assurance.
“PAT promotes a process which starts with the identification of each product’s specific Critical Quality Attributes (CQAs), then proceeds with monitoring as often as possible the related Critical Process Parameters (CPPs) and the Key Performance Indicators (KPIs), in order to automatically control them within pre-defined limits,” explains the Hamilton Co. white paper Biopharma PAT.
Effectively, understand what makes your product safe and effective, what processes ensure those features, and how to measure those processes.
As Triumvira Immunologics Chief Technology Officer Donna Rill told the audience at Cell and Gene Therapy Revolution in Toronto this past spring, it is important to start with a knowledge of CQAs as early as possible in therapeutics development. And as you develop your processes and learn more about the end product and its impacts, to refine and adjust those CQAs.
Importantly, she added, every time you decide to introduce a change to your process, you need to ask yourself if and how that change impacts the CQAs.
But as the earlier anecdote suggests, the world of biology can be messier than the traditionally precise fields of biophysics and chemical synthesis.
“With chemistry, there is still a lot that we don’t know, but we have this idea of theoretical yield,” offers Kelsey Mato, market segment manager for process analytics at Hamilton Co. “We can say our turnover is 24 to 26 percent; we have a much tighter understanding of what’s happening.”
“With biology, researchers are still learning about the complex nature of cells, including the high degree of heterogeneity that exists within the same culture from the same cell line,” she contrasts. “There are so many factors that the best they can do is measure everything possible, and then one day look back and determine that a particular measured parameter was actually an indication of the scale-up working or not working. Now that they have this data, they can back-apply it.”
To monitor the CQAs of an antibody, gene therapy vector or living therapeutic, there is no end to the number of analytical methods available throughout the downstream process steps to final product fill-finish. LC-MS or immunoassays, for example, are routinely used to monitor sample purity or characterize post-translational modifications (e.g., glycosylation).
The challenge is that these assays measure CQAs after the product has been produced in the cell, when it is often too late to change anything or fix any problems.
Recognizing this challenge, there is a move to look more closely at the bioreactor environment, to better understand the CPPs during cell culturing that impact the final product as it is being produced.
Obviously, the bioreactor environment is not a black box. Regardless of format and size, most units offer a number of analytical entry points with built-in sensors.
Different cell lines, for example, have different requirements for culture media pH and dissolved oxygen. Likewise, dissolved CO2 influences media pH as well as cellular fatty acid synthesis. Temperature and pressure also influence culture viability.
Each of these CPPs is typically measured in-situ or in-line, with probes inserted directly into the bioreactor chamber so that adjustments can be made in real-time.
Particularly with fed-batch and perfusion processes, nutrient and metabolite concentrations can be vitally important for controlling feed strategies during growth.
“Glucose or glycerol are the main C-source (carbon source), while glutamine is the main N-source (nitrogen source), together with other amino acids in these bioprocesses,” the Hamilton whitepaper reports. “During the fermentation they are consumed, and secondary metabolites such as lactate, acetate and ammonium are produced. Suboptimal feeding strategies can produce excessive secondary metabolites, which hinder cell viability and product yield.”
In-line monitoring is not possible for many these components, unfortunately, so at-line and/or off-line methods are required, slowing response times and increasing contamination risks through sampling.
“However, more sophisticated methods are starting to be adopted, including non-invasive optical sensors such as infrared spectroscopy (NIR and MIR) and Raman spectroscopy,” wrote Damian Marshall and colleagues from Cell and Gene Therapy Catapult last year. “These non-destructive technologies can be used in-line to provide simultaneous real-time information about multiple components of the culture environment, including the consumption of nutrients and the production of metabolic waste products.”
In particular, their group was looking at Raman spectroscopy and univariate modeling as potential PAT approaches to cell culture.
“Typical culture media, even when chemically defined, consists of over 40 components, including inorganic salts, amino acids, sugars, vitamins, alpha-keto acids and pH indicators such as phenol red,” they noted. “Therefore, any sensors applied for functional monitoring of cell behavior must be able to make precise measurements of target nutrients and metabolites without interference from the other multiple components present.”
Looking at T cells derived from different donors, the researchers noted multiphasic growth profiles that were relatively consistent between some cultures and not others. This divergence, the authors suggested, highlights the importance of real-time monitoring.
“The ability to measure these key parameters using an in-line Raman optical sensor makes it possible to have immediate feedback on process performance,” they concluded. “This could help significantly improve cell therapy bioprocessing by allowing proactive decision-making based on real-time process data.”
Raman spectroscopy is also being used to help overcome a challenge earlier in process development, when researchers need to screen multiple cell lines and/or growth conditions to determine optimal performance.
Traditionally, these analyses were performed with shaker flasks, but there is increasing use of minibioreactors such as the ambr system from Sartorius Stedim. The small volumes of these minibioreactors, however, mean that sampling to monitor nutrients, metabolites or product must be reduced either in size or frequency, as volume loss can significantly impair growth.
To circumvent the larger sample volumes and prolonged measurement turnaround times of traditional off-line bioanalyzers, Ruth Rowland-Jones and colleagues at Newcastle University and Lonza Biologics recently examined the potential for Raman spectroscopy to permit frequent sampling and near-real-time analysis of multiple analytes.
Testing multiple cell lines, the researchers found that although Raman measurement of the analytes was less accurate than current reference methods, the small sample volumes (<50 μL) and high-throughput methodology suggested Raman would provide meaningful productivity profiles.
“Further to this,” the authors added, “it was shown for the first time that the output of a Raman glucose and [viable cell concentration] model could be used to control the glucose feed rate and main nutrient feed rate of miniature bioreactor cultures with similar feed rate outputs to those achieved using current measurement technologies.”
The repertoire of in-line probes continues to expand as researchers strive to better understand and monitor their cultures. Mato points, for example, to Hamilton’s Incyte and Dencytee products.
“Dencytee measures total cell density using a turbidity measurement at 880 nm, which correlates to the entire cell population,” she explains. “Cells that are alive or dead or stressed are all going to be represented more or less equally in those measurements.”
In contrast, Incyte is a capacitance-based measurement, so it measures viable cells.
Typically, she suggests, people tend to perform this test off-line, using any number of methods, each of which has its pros and cons.
“But in all regards, off-line measurements provide only a quick snapshot,” Mato cautions. “The measurement only indicates what a cell density is at that minute.”
Complicating this picture is the fact that even when you have a sample, quite a bit of sample prep is required to ready it for measurement, and this can be time-consuming.
According to Mato, “By the time they get a count, they’re usually already about an hour past the sampling point.”
Thus, your snapshot doesn’t reflect what is currently true.
“By moving to Incyte and Dencytee, you’re getting information immediately, which means you can act on it immediately,” she says.
By using timely, precise automation rather than a retrospective snapshot approach, Mato states, clients are seeing improvements in product yield and quality, as well as decreasing processing costs. By being able to react more quickly to events, people are able to make their processes work better.
It also reduces the risk of users missing critical moments or only learning of them belatedly.
“A lot of our mammalian customers will sample once a day, for example,” Mato explains. “If they take a sample at nine in the morning, and something happens to the process at nine at night, they don’t know until nine o’clock the next morning.
“And if the process recovers during that time, they might not know it even happened at all.”
Thus, large stretches of a process may be completely invisible to process developers.
Upstream metrics are also being increasingly used as surrogate markers for downstream outcomes, where process developers have correlated specific product CQAs with cell culture parameters.
“It gets back to what we were talking about earlier,” says Mato, “where you measure everything, so that you can use different types of analytics post-processing to say, for example, when these five parameters hit this combination, then that’s when we get our ideal.”
“We always advocate for a direct measurement, if something is possible,” she continues, but acknowledges that this isn’t always the case, sharing the example of antibody glycosylation, which cannot yet be monitored in a bioreactor.
Mato’s colleague, Casey Snodgrass, market segment leader for pharmaceutical biotechnology at Hamilton Robotics, offers his experiences.
“On the robotics side,” he says, “I work closely with our biopharma customers that do antibody analysis to look at glycosylation, for example, or are using the multi-attribute method (MAM), pioneered by Richard Rogers and Da Ren, who also hold leadership positions in the MAM Consortium.”
Snodgrass points to the fact that when R&D groups implement tools such as MAM, it’s important to plan for how the tools will be carried through the scale-up process to production levels so that all groups are aligned. Off-line analyses of protein glycosylation or titer, for example, help process developers determine which in-line parameters are most critical to monitor.
“As the market pushes more toward the PAT initiative, the use of soft sensors will increase,” Mato suggests. “For example, it may be that dielectric spectroscopy is one day used to predict things like apoptosis, glucose consumption and things like that.”
“Long term, the benefits of in-line measurement are so prominent and so well understood that the push will continue toward real-time in-situ type measurements,” she adds.
Until that is a reality—or perhaps in a drive to make that a reality—organizations are pushing to close the distance between in-line and off-line technologies in the hope that if they can’t monitor in real-time, they can at least improve the process to near real-time.
One of the main reasons why off-line measurements are off-line is the need for extensive sample preparation, such as desalting steps, sample clarification or chemical derivatization to enhance analytical signal. Shortening, minimizing and/or automating these steps would ideally produce actionable results faster, allowing process developers to adjust culture conditions accordingly.
“Hamilton collaborates with manufacturers who specialize in tip-based technologies, where their specialized chemistries are embedded in our CO-RE pipette tips and used in fully automated workflows,” says Snodgrass. “Some examples are Protein A purification and size exclusion desalting.”
He acknowledges that this is not an on-line real-time technology, but he sees other advantages, suggesting it creates a fully walk-away workflow. This makes it a lot easier for customers to adopt, he adds, because it works directly with Hamilton’s platform, and they don’t have the added expense of extra equipment. Hamilton, Snodgrass continues, actively seeks partnerships with other companies that provide chemistries and technologies and want to automate it.
Advances in microfluidics are also starting to play a role in shortening the distance between bioreactor and product profiling.
In 2017, for example, Zoran Sosic and colleagues at Biogen attempted to monitor glycosylation profiles and antibody titers directly from cell culture medium using ZipChip CE-MS.
“Protein glycosylation can be significantly impacted by the host cell line, clone, media composition, feeding strategy and downstream processing conditions,” the authors wrote. “Therefore, N-glycosylation analysis of monoclonal antibody is widely performed in the biopharmaceutical development, and in some cases is used as release assay when glycosylation is established as a critical quality attribute.”
Methods to study these attributes tended to be time-consuming and involved numerous steps. As suggested earlier, such time delays could interfere with timely interventions to optimize culture conditions and product output.
As the researchers explained, the microfluidic ZipChip allowed them to analyze protein isoforms under reducing conditions directly from cell culture supernatant without protein purification. And the addition of heavy isotope controls allowed them to monitor protein titer within the same mass spectrum by comparing signal intensities.
Performing parallel analyses with LC-MS and RPLC-MS, the researchers found that the results using ZipChip CE-MS were largely consistent but much faster than the more traditional monitoring methods. Furthermore, they were able to detect very minor glycoforms and aglycosylated antibodies, down to the lower end of accurate quantitation by any of the methods.
“The reduced mass analysis using ZipChip CE-MS for non-purified cell culture supernatant samples allows speed and simplicity required at the early stage of cell line screening, process development and batch manufacturing,” the authors concluded. “This approach can be used for detection of undesired shifts in product quality that may lead to adjustment of related manufacturing parameters to ensure process being within the expected ranges.”
“With the future implementation of automated sample loading,” they projected, “its application could be expanded as a new analytical tool for at-line and in-line cell culture performance monitoring.”
At the recent Bioprocessing Summit in Boston, ZipChip manufacturers 908 Devices extended this type of analysis further with the launch of Rebel, a miniature CE-MS platform. The unit is designed to monitor nutrients and metabolites in bioreactor media in near real-time.
Using samples as small as 10 μL, Rebel can quantify >30 amino acids, vitamins, amines and dipeptides in as little as seven minutes, facilitating rapid control of media conditions and providing insights into cell growth and performance.
With an eye to similar applications, Microsaic Systems has decentralized mass spectrometry with its miniaturized unit for point-of-need use. In a recent proof-of-concept application with colleagues at Darlington, UK’s Centre for Process Innovation, Microsaic researchers performed IgG & metabolite analysis from bioreactor cultures.
For IgG monitoring, they injected filtered samples directly onto a Protein A column and then eluted the proteins with volatile buffers directly onto MiD ProteinID, their ESI-MS platform. This method allowed them to follow antibody modifications over a 14-day time course under favorable and unfavorable growth conditions.
“A mean difference between the favorable and unfavorable, and across the time-points, of 50 Da was seen at the one percent significance level,” they noted.
“Further to the monitoring of the main IgG product, additional species in the mass spectra could also be observed from additional peaks in the mass-to-charge spectra,” they continued. “These peaks are attributable to IgG fragments and in particular, the HHL fragment at a mass of ~122 kDa.”
The team then performed similar analyses on culture metabolites, using hydrophobic interaction chromatography (HILIC) and selected ion monitoring, and found good correlation between their method and photometric analysis.
The profiles, which included metabolites such as lactate, glutamine, glucose, glutamate and several other amino acids, also showed clear differences between cultures grown under favorable and unfavorable conditions.
“Improvements in measurement speed, precision and accuracy would be expected with further automation of sample preparation,” the researchers added, recognizing there was still room to develop. “The use of bespoke Protein A and HILIC configurations would also allow smaller sample volumes and lower flow rates to be utilized.”
More recently, Cyrus Agarabi and colleagues at FDA’s Center for Drug Evaluation and Research and University of Massachusetts performed at-line amino acid monitoring of crude bioreactor media to see how amino acid supplementation might impact not only cell viability but also antibody quality and glycan profiles.
“This analytical method for crude bioreactor media was developed with future on-line real-time PAT implementation in mind, which necessitates minimal sample preparation that is fast and avoids derivatization when possible,” the authors explained.
“With our normal phase liquid chromatography mass spectrometry (LC–MS) based method, crude media can be collected from the bioreactor, processed in less than 10 min and run on a 15-min gradient for complete amino acid characterization in near real-time,” they added. “Only 10 μL is required per replicate, allowing this method to be used for cultures with small working volumes as well, such as microbioreactors.”
Depending on the feed strategy tested, the researchers found that certain amino acid cocktails could rescue diminishing cell viability, often an otherwise irreversible condition, and extend batch age performance. Likewise, supplementation also had small but detectable impacts on antibody charge and size variants, as well as glycosylation patterns in the final products.
“Collectively, our protein structural analysis illustrates the importance of understanding how process parameters and bioreactor nutrients can affect product quality, as in this case where a favorable increase in [viable cell density] performance results in a potentially less favorable glycan profile outcome (with less galactosylation and increased high mannose glycoform amounts),” they highlighted.
The scientists stressed the importance of multivariate analysis incorporating traditional PAT metrics (e.g., pH, temperature, media variables, etc.) and those derived from amino acid utilization studies like this one to develop more robust models of cell viability and productivity.
“Better knowledge and informed supplementation strategies of the upstream process could provide increased yield of consistent quality and potentially avoid batch failures,” they concluded.
Autonomy & autology
As suggested in the Cell and Gene Therapy Catapult example discussed earlier, not all bioproduction cycles are about a protein, vector or allogeneic cell line. With autologous cell replacement therapy, each prospective patient provides his or her own batch of cells.
This potentially complicates the idea of applying lessons and CPPs from previous cell cultures to the new ones.
“At least in the near future, there is always going to be some level of starting over with each new culture,” says Mato.
That said, she continues, it is not all or nothing. With experience comes a better understanding of what might work. In Mato's opinion, it establishes boundaries of where to start.
“But the best thing you can do, especially as a technology or equipment manufacturer, is to make products that measure more directly,” she suggests.
“Raman is a great tool,” she offers as an example, “but it can give too much information in this case. If a cell line changes a little bit, the data changes a lot. The more we can work toward the implementation of specific and direct measurements, I think that will get better and better.”
More important to Mato is the understanding that in autologous-type therapies, or any cell therapies really, the cells themselves are the product.
“There’s not really the benefit of the stringent purification processes, so one of the biggest things to focus on is sterility,” she says. “Of course, a sterile environment is important in all cases, but with this case, a sterile environment is even more important.
“Bringing those measurements in-line, doing as much as you can in-line without having to take an off-line sample and exposing the reactor to contamination, helps to maintain sterility. So, one is always starting over a little bit, but implementing in-line specific measurements can help with a head start on it.”
Although we are not yet at the point where we can directly monitor how well our cell cultures are producing our desired products, the desire for fully closed systems and advancing technology platforms continue to bring us closer with each bioreactor generation.
Process and product are intimately linked under the Process Analytical Technology (PAT) initiative, as each step in a production run influences the quality of the final product. Success is defined in one of three categories.
Critical Quality Attribute (CQA): A property or characteristic that helps define a product’s desired quality (often linked to safety or efficacy).
Critical Process Parameter (CPP): A process component that helps ensure CQA and therefore needs to be monitored and controlled.
Key Performance Indicator (KPI): Metrics that define the status of each production step, offering insight on how tightly a process is adhering to CPPs to ensure CQAs.
The ultimate goal of PAT is to have each metric monitored as close to the point of origin as possible. With bioreactors, for instance, the push is to measure parameters directly within the culture chamber.
In-line/in-situ: The measurement sensor resides directly within the bioreactor, permitting real-time monitoring of attribute and/or parameter fluctuations.
On-line: A sample is diverted from the bioreactor with a bypass stream to an external sensor and may be returned. Measurements are made effectively in real-time.
At-line: A sample is removed from the bioreactor to be analyzed nearby. Sampling, sample prep and analysis may lead to time delays, which can inhibit timely intervention.
Off-line: A sample is removed from the bioreactor to be analyzed elsewhere or at another time. Sampling, prep and analysis is typically delayed, preventing real-time response.
Adapted from Biopharma PAT. Hamilton whitepaper, 2018