Bioprocess Characterization: What’s the Fuss?

Bioprocess characterization involves identifying key process parameters such as pH, glucose levels, feed schedule, etc., and constructing a design space that describes what occurs during a process, and ultimately to the product, when parameters change. By setting limits for tolerable parameter excursions from acceptable ranges, design space provides two benefits: a high level of comfort in the absence of such excursions, and a potential means for fixing or rationalizing processes that experience unexpected deviations.

Design space became an issue for bioprocessors as a result of the U.S. FDA’s emphasis, since 2004, on process analytics, risk-based manufacturing, and designing quality in (aka “quality by design”).

Regulations aside, it generally behooves biopharmaceutical developers to know what’s going on in their process. Like all drug developers, therapeutic biotech faces severe resource constraints, ultimately deriving from the opportunity losses caused by delays.

“Developers have limited time, but many experiments to conduct, especially when employing design of experiment (DOE),” says Mark Duerkop, founder of Novasign Solutions. “You can screen for parameters defining a design space, but that space is multidimensional and grows exponentially with the number and complexity of required experiments. The question then becomes how to screen that design space for your anticipated response variables, for example titer or glycosylation pattern, as rapidly as possible.”

For these exercises Novasign offers several tools to “lean” DOE-based development projects.

The company’s Intensified Design of Experiment (iDoE) Toolbox allows rapid definition of design space—up to 50% faster than conventional methods, according to Duerkop. Hybrid Modeling Toolbox, which defines input parameters and predicts outputs, bridges the gaps existing in conventional DOE between critical process parameters (inputs) and critical quality attributes (outputs). Hybrid Modeling combines the predictive capabilities of mechanistic models with information “hidden” in the data that process engineers frequently miss. The software addresses both QbD and PAT requirements, and feeds into a second program, Digital Twin, a process simulator that predicts product characteristics for all scenarios—as if the experiments were conducted under constant conditions.

Doing as much (or more) with less

Shortcomings of conventional process characterization (like classic DoE) and modeling include their inherent assumption that parameters remain constant during runs, while addressing critical quality attributes only at the endpoint of the process. It is very much a black-box exercise in which a set of parameters is established and correlated to the characteristics of the end product.

Using the Novasign tools, developers are able to change individual parameters, like feed schedules, pH or temperature, dynamically, individually, or simultaneously. “That means it is possible to reduce the number of experiments, while screening for more of the design space in less time,” Duerkop explains.

Operating in this fashion, process inputs are those critical process that matter—the ones that change titer, host cell protein levels, or any product- or process-specific CQA. Thus, ideally, PAT becomes the input and QbD the output.

Ultimately, approaches like Novasign’s could lead to real-time bioprocess control, a result enjoyed by other (even low-tech) process industries but which has thus far been elusive in biomanufacturing. “We’d like to go in that direction, but regulatory barriers and associated risk are very high,” Duerkop says. “We hope to convince the marketplace that reducing the number of experiments during DOE process development is itself a benefit. At the end, users will have screened for all relevant parameters and learned enough from those experiments to enable real-time process control, but that is at least three to four years down the road.”

Duerkop cited a hypothetical situation where process characterization on a 100-liter tank might cost $20,000 per experiment, and where developers might obtain the same or more relevant data from six experiments vs. ten.

No article on potentially disruptive biomanufacturing technologies would be complete without asking why, 15 years after FDA’s landmark guidance on PAT, and at least a decade since QbD became “a thing,” we are still having discussions on topics that have been non-issues to manufacturers of soda, cement, and paper clips for decades.

According to Dr. Duerkop, biopharmaceutical revenues are so high that the pursuit of relatively minor efficiencies does not make financial sense. “Their needs are not high enough, and changing processes or introducing new technologies is a pain.” To derive the most benefit, and the easiest adoption path, companies must take these measures early in development. Like other industries, which exploit modern manufacturing technologies, biotech could benefit from improving process understanding. But unlike other industries they don’t need to.

Ronald Rader, senior technical researcher at BioPlan Associates, agrees that many bioprocessors remain unconvinced that they need PAT or DOE, but not for the same reasons given by Duerkop. “There are no explicit requirements in either INDs or BLAs for these activities,” he says. Moreover few biopharmaceutical-manufacturers are prepared or have the requisite mindset, despite the fifteen years since FDA promulgated its PAT guidance. “Most companies have not made deep commitments to amassing and analyzing huge collections of process-related data. To do this requires computing power, appropriate software, and highly trained technical staff—multiple millions of dollars each year, to collect data that regulators don’t demand, just to be able to further reduce risks and optimize processes beyond what has worked well for decades.”

On the topic of parameter excursion, Rader raises the additional point of what might occur, from a regulatory and product liability perspective, if the characterization exercise works too well, and identifies problems that do not affect product quality directly. The existing biotech mindset, that “the product is the process,” would suggest taking action or trashing a batch entirely. Thus, the bioprocess equivalent of “the operation was a success but the patient died.”

Biotech is the “playground” for large companies with deep pockets, Duerkop says. “Taking a new drug to market costs approximately $2B, but 2018 biopharmaceutical revenues were an astounding $240 billion. Although this already represents half of the smartphone market worldwide biotech revenues are based on a relatively small number of products.” Smaller companies, where innovation often occurs, are essentially cut off from implementing new development technologies because of the high regulatory pressure because they exist in the shadows of larger firms that have created a regulatory and commercialization climate that is unfriendly to change.

Ultimately, the question of how much process characterization is required comes down to value and opportunity.

“Just about every company today performs a good amount of process optimization by late clinical stage, or certainly before commercial manufacturing,” Rader notes. "Many are unwilling to go beyond that with PAT and DoE. Many observers and commentators seem to forget that risk assessment and reduction, and ultimately patient benefit, are the real goals with these data-driven initiatives."