Progress in the Prediction of Protein Aggregation

In silico analysis and the evaluation of formulability, aided by new analytical tools such as hydrogen deuterium exchange mass spectrometry (HDX MS), are enabling the improved design, production, and formulation of protein-based drugs. As the development of biologic drugs based on antibodies and other protein/peptide molecules continues apace, the control and prevention of aggregation during manufacturing, downstream processing, formulation, storage, shipping, and administration remains an important issue. As a result, significant efforts have been invested in understanding the mechanisms behind aggregation and the establishment of predictive models and more advanced analytical techniques, particularly high-throughput methods, to enable the design and formulation of stable, protein-based therapeutics.

The importance of aggregation
Because aggregation can occur at any point during manufacturing, processing, formulation, and use of protein-based drugs, the consequences of this phenomenon can be significant. “Cells may stop secreting a protein if it takes on a new, aggregated form, leading to low titers and reduced yields and productivity. In addition, the formation of larger precipitates can cause the physical clogging of equipment and handling of the product during downstream processing,” says Jesus Zurdo, head of innovation with Lonza Custom Manufacturing’s development services team.

If aggregation occurs during formulation, which is of particular concern for protein-based products that require high concentration, the performance and safety of the product can be affected. “There is evidence to suggest that the presence of aggregates increases the immunogenicity of biologic drugs. How aggregates trigger immune responses, and which aggregates are responsible, isn’t known at this point, but it is thought that aggregates are more easily recognized than the parent protein,” Zurdo explains. As a result, there is increasingly regulatory pressure for biopharmaceutical manufacturers to reduce the risk of immunogenicity by not only controlling, but also preventing, aggregate formation.

Understanding aggregation
Aggregation involves the undesired unfolding or misfolding of proteins, which then enables protein-protein interactions and the formation of a possible range of larger molecules, from oligomers to very high molecular weight species. The most well understood aggregates are amyloid fibrils, which are composed of beta sheets stabilized by hydrogen bonds.

Typically, short polypeptide segments of 5-15 amino acids in length within folded protein structures that are affected by changes in the protein environment are responsible for nucleating the aggregation of the entire protein. Aggregates are classified as soluble or insoluble, as well as covalent (involves the formation of a covalent bond, often a disulfide linkage) or hydrogen-bonded (weaker interactions). Self-association of therapeutic proteins via covalent bonds is typically irreversible while aggregates formed via weaker interactions may be reversible upon changes in protein concentration, temperature, and pH, for example. As a result, aggregates can range in size from minute, invisible, non-filterable particles to large precipitates that are visible with the naked eye. In addition, some aggregates may be static while others may be dynamic.

External factors do not alone determine when protein aggregation will occur. The physicochemical characteristics of the short polypeptide segments that comprise protein therapeutics also determine the propensity of each segment to undergo aggregation. While much remains unknown about the actual mechanism(s) of aggregation, a great deal has been learned about the influence of these physicochemical factors on the likelihood of an amino-acid sequence to serve as an aggregation nucleating site, according to Zurdo.

Predicting protein aggregation
This knowledge about the relationship between the structural characteristics of peptide segments and the likelihood of aggregate formation has been used to develop predictive models. Other factors that have been used to develop these models include the unfolding kinetics and the thermal and colloidal stability of native proteins. The properties of the peptide segments of interest generally include the polarity, charge, dipole moment, hydrophilicity and hydrophobicity patterns, accessible surface area, charge, and number and location of aromatic residues.

“The idea is to match specific amino acid sequences (and structural features) within proteins to the probability of aggregation by comparing their physicochemical and structural attributes to those of sequences with known aggregation behavior using predictive algorithms generated based on actual experimental data,” notes Zurdo. “It is very difficult to predict aggregation de novo,” he adds. Thus, efforts have been invested in synthesizing analogs and studying their behavior to expand the database of information on known proteins and antibodies. Lonza’s database currently includes information on the aggregation behavior and properties of more than 1000 individual antibody molecules.

Potential protein-based drug candidates are screened against the database to identify the potential for aggregation. If the results indicate that any amino acid sequences within the molecule might have the propensity for aggregation, then the protein can be modified, or if necessary, redesigned. “We have been successful to date in using such in silico analyses to address the issue of protein aggregation much earlier in the drug development process, which helps avoid potential problems throughout the scale-up, manufacturing, and formulation stages,” asserts Zurdo.

New analytical tools for aggregate prediction
Because the success of these predictive models depends in large part on the accuracy of the physicochemical property data of the peptide segments, effort has also been directed at developing more advanced analytical methods for obtaining more detailed structural protein information. One such method is hydrogen deuterium exchange mass spectrometry (HDX MS), which enables the detection of potential sites of aggregation because they are systematically protected from solvent exposure (HD exchange) when in the aggregate state, but not protected when in the free protein (monomer) state. “By knowing the specific site or sites prone to aggregation, targeted formulations or mutagenesis studies can be devised,” says Asish Chakraborty, business development manager with Waters Corporation.

Peptide-level HDX MS
In a peptide-level HDX MS study, HD exchange for labeled soluble protein and aggregated protein samples is evaluated. The regions of a peptide involved in aggregate formation will show a different level of deuterium uptake from those regions in the free protein. Targeted electron-transfer dissociation (ETD) fragmentation of the peptides that exhibit deuterium uptake provides further detailed information. “Additional information on the dynamics of these interaction sites can be obtained by looking at uptake over a series of time points, where more tightly bound regions (less dynamic) will have a lower deuterium uptake profile than more transiently interacting sites that have more average solvent exposure,” observes Chakraborty.

HDX MS complements other technologies (e.g., nuclear magnetic resonance and X-ray crystallography) and has the advantages of requiring a minimal sample size, tolerance for many formulation components, and simplified sample preparation. It can also be used for larger proteins (hundreds of kilodaltons) and protein mixtures, and the presence of certain protein impurities is tolerated. The most important recent advances in this technology, however, have been in the development of informatics tools, such as Waters’ DynamX software, that enable much more rapid screening of the data produced during HDX MS analyses to determine the key regions of interaction. In addition, Waters has developed an integrated HDX MS system that has made HDX more routinely accessible and enabled automated data acquisition.

“With these two advances, it is now possible to rapidly reduce a very large quantity of analytical data very quickly to interpretable results,” Chakraborty states. “What used to take weeks can now be accomplished in a couple of days.” He also points to the use of the liquid chromatography/mass spectrometry technique LC/MSE for data-independent fragmentation and identification of hundreds of peptic peptides evaluated in sub-10-min HDX-MS runs and ion mobility LC/MSE (LC/HDMSE) for gas-phase separation of ions to achieve cleaner ion detection and better quantitation of deuterium uptake for individual peptides as being additional important developments. He would, however, like to see ETD fragmentation more widely used. To achieve that goal, Waters is working on peptide-specific optimizations, looking to improve HDX MS structural data generated for larger and more complex protein systems to develop new informatics tools to facilitate more efficient acquisition and interpretation of ETD fragmentation data for HDX MS experiments.

Evaluating the formulability of protein-based drugs
There have also been significant advances in analytical methods for the determination of protein aggregation, including high throughput capillary electrophoresis, immunoassays, and the optical imaging of aggregates for both quantitative and qualitative evaluations, according to Zurdo. Many of these methods, however, do not meet the requirements for very high-throughput analysis of potential drug formulations.

“The formulation readiness of protein-based drugs--or their formulability--can be an issue, particularly for biotherapeutics that must be administered in high concentrations. Because there are too many variables, there as yet have been no models developed to predict protein aggregation in a product formulation. Therefore, manufacturers must rely on physical analyses. As a consequence there is a significant need for inexpensive, simple, highly robust, and very high-throughput techniques for the determination of protein aggregation that give either yes/no or better/worse-type information that can be used to map the regions of experimental space where products appear to behave well,” says Zurdo.

The development of such methods is in the early stages. Some examples include modified immunoassays, self-interaction chromatography, and visual viscosity determination based on the diffusion of latex beads. “Unfortunately, there have been very few such techniques commercialized to date, and those that have been have not yet been fully validated. But progress is being made, and we expect to see some new methods being implemented in the not too distant future,” Zurdo notes.

The behavior of the formulated product once administered to patients must also be considered, because protein aggregation can occur in the human body as well (e.g., Alzheimer’s disease is caused by protein aggregation). There are ongoing studies aimed at developing sophisticated cell-based assays for the evaluation of the immune response to protein aggregates, but despite some preliminary links between in vitro data and clinical outcomes, to date there is not yet a definitive or absolute direct correlation between results obtained through this type of assay and the actual responses observed in patients.

It is a challenge to do this type of analysis because the final formulation of a therapeutic protein often does not contain the exact same composition that was used during clinical testing, and even the minor changes that are acceptable during final formulation can lead to protein modification or aggregation and thus promote immunogenic responses,” observes Zurdo.