Modeling and Multimodal Resins
Multimodal chromatography has immense potential for purifying many types of therapeutic proteins. Combining hydrophobic interaction and ion exchange in one resin creates exciting possibilities, but the complexity of multimodal interactions with diverse protein topographies has hindered new resin development and thwarted attempts at arriving at general rules for what works and what does not.
Steven Cramer, Ph.D., a professor of chemical and biological engineering at Rensselaer Polytechnic Institute, is using biophysical tools (NMR, SPR, and AFM), molecular simulations, coarse-grain modeling, and “a lot of chromatography” to develop models that predict relevant interactions between resins and proteins.
What’s interesting about multimode resins is that they invalidate many preconceptions regarding what works in chromatography. Dr. Cramer has recently used Capto™ adhere resin (GE Healthcare) as the capture step for a growth hormone analog, and achieved “exquisite capture and very high purification in one step.”
Even monoclonal antibodies, which are captured by cation exchange resins (in addition to protein A), might be purified by anion exchange multimode chromatography. “Even though the net charge on the monoclonal antibody may be positive, there are still hydrophobic regions, and negative patches on the protein surface,” explains Dr. Cramer. “So, depending on the protein, you can still get capture with a multimodal anion exchanger.”
Under the right conditions, anion change-based resins could also be used in flow-through mode to remove impurities that might otherwise be removed by precipitation.
Dr. Cramer has collaborated with several large resin and biomanufacturing companies on his approach. He found that they need to appreciate that multimodal interactions are not as straightforward as protein A binding for monoclonal antibodies. For example, many companies that express interest in the multimodal approach mistakenly believe that a resin capable of acting in either HIC or ion exchange mode should automatically bind to molecules with known affinity to these modes. It’s not that simple.
“It really depends,” Dr. Cramer says. “It’s not simply the sum of hydrophobic or electrostatic groups. Rather, it’s a matter of the patches on protein surface. How big are they, and how close together?” Ideally, the interaction should involve synergistic multimodal binding, not just a souped-up version of an ion exchanger.
That is why some resins behave like a multimodal ligand for one protein and like a unimodal ligand on another. It depends on the distribution of “patches” on the protein surface, their chemical nature, and their proximity.
Using simulation and biophysical data, Dr. Cramer has devised a way to quantify a protein’s affinity behavior and predict, through a “supermodel,” the binding of any protein with any multimodal system, at any pH.
Marcel Ottens, Ph.D., an assistant professor of microbiosystems technology and process chromotography at Delft University of Technology, uses computational and modeling tools with “biothermodynamic” inputs to predict the behavior of therapeutic proteins in common purification settings. Biothermodynamics is a term borrowed from the study of energetic characteristics of living systems and is equivalent to “thermodynamics.”
Dr. Ottens applies this modeling approach to chromatography, but it is also relevant to protein phase behavior as it relates to adsorption, extraction, solubility, partitioning, precipitation, and crystallization. This kind of approach, says Dr. Ottens, “paves the way for in silico process development, providing better process understanding.”
Normally, high-throughput process development is 100% experimental, for example, through high-throughput, automated design-of-experiment. Here Dr. Ottens uses the experiments to determine the parameters of thermodynamic or mechanistic models. “Once we have these parameters, we have what we need to perform scenario analyses on the computer. We can optimize configurations or operational conditions for a particular unit operation.”
Think of this approach as human-assisted in silico modeling where the person sets reasonable constraints and the computer works through data from actual high-throughput, automated experiments. Without human input, the computational problem becomes too complex to solve practically.
Dr. Ottens describes these inputs as “rules of thumb, heuristics, and common sense that limit the combinations you need to run in silico.” Without the computer, the design of experiment possibilities become too numerous for actual experimentation. Dr. Ottens has formed a company, Marlin Biopurification, to commercialize his technology. At present, the company is operating as a service center at Delft University of Technology.