Developability Predictions for Antibody Engineering and Risk Mitigation

• Boehringer Ingelheim

Introduction

Biologics, in particular monoclonal antibodies, have revolutionized medicine in the last two decades and nowadays represent the standard of care for a variety of diseases, most notably in oncology and autoimmune disorders. This fact is reflected by the market share biologics have obtained in 2015, in which seven of the top ten selling drugs belong to this class.

Biologics differ substantially from classical small molecule drugs not only by their size (the molecular weight of an antibody is about three hundred fold higher than a typical small molecule drug) but also in the way they are discovered, and even more importantly, how they are manufactured and administered. In most cases antibodies are generated through immunization of mammalian species (e.g. mouse or rabbit) with the desired antigen, produced in cell cultures and administered either i.v. or s.c. since molecules of this size are generally not bioavailable after oral application.

Secondly, the function of antibodies is tightly coupled to the integrity of their three dimensional structure, which is sensitive to heat,interfacial stress and pH conditions to which small molecule drugs are typically resistant. Hence, manufacturing of antibodies -which involves production in living cells, separation of the antibody from the remainders of the cells, purification, filling and other manufacturing steps which essentially denote stress for proteins - represent sub-stantial challenges for development. The suitability of antibodies to be manufactured in large-scale, which is commonly referred to as developability, therefore is not always a given.

Since antibodies are a product of an evolutionary process as part of the adaptive immune response their sequences vary substantially. The driving force for the in vivo selection and maturation is biological function and not suitability for large-scale manufacturing and long-term storage. Hence, antibodies derived from immunized animals do not necessarily fulfill the requirements to be developed into drugs. The same holds true for antibodies derived from in-vitro technologies like phage-display, where the selection criterion is typically based solely on binding.

The careful selection of suitable candidates and the rational engineering of antibody sequences for improved developability while maintaining biological function are therefore intensive fields of research. Central to this effort is the understanding of the molecular mechanisms of antibody degradation and the ability to establish relationships between the sequence and undesired properties of antibodies to guide protein engineering.

Antibody Degradation Pathways

Antibodies can undergo various types of degradation which can be roughly divided into chemical degradation and protein aggregation.¹ Chemical degradation denotes the chemical transformation of the antibody. The most frequent chemical degradation reactions are:

• fragmentation, which denotes the formation of antibody fragments through hydrolysis
• deamidation, the chemical transformation of an amide to a carboxylic acid
• isomerization, conversion from aspartate into isoaspartate
• oxidation, reaction of an amino acid with oxygen

Fragmentation occurs primarily in the hinge region of antibodies.² There are two mechanisms underlying fragmentation, direct hydrolysis and beta-elimination of disulfide bonds. The reaction rate of direct hydrolysis is strongly pH-dependent with showing acceleration above as well as below pH 6.³ Beta-elimination of disulfide bonds occurs at the hinge disulfide bonds and primarily occurs at basic pH.⁴ Since most fragmentation reactions occur at the hinge region, which is not variable, there are limited options for sequence engineering to influence fragmentation rates. However, formulation optimization can keep fragmentation reactions at a minimum.

Deamidation is the chemical transformation of an amino acid with an amide side chain (asparagine or glutamine) into a negatively charged carboxylic acid side chain (aspartate or glutamate).^5,6 Factors which determine the reaction rate are:

• the amino acid itself (where asparagine is usually more critical than glutamine),
• the subsequent amino acid (N+1) where glycine in particular causes fast deamidation,
• to a lesser extend the previous amino acid (N-1),
• the local three-dimensional structure and
• the local conformational flexibility.⁷

 Figure 1. Deamidation and aspartate isomerization: Asparagine and aspartate can react with the subsequent amino acid into a cyclic intermediate (succinimide). When this cyclic intermediate is hydrolyzed, the ring opening can either result in the reaction back to aspartate or into Iso-aspartate

The reaction requires a distinct geometry at which the nitrogen of the subsequent amino acid reacts with the carbonyl moiety of the amide group forming a cyclic intermediate (succinimide). When this intermediate is hydrolyzed the former amide group is transformed into a carboxylic acid group, either aspartate or iso-aspartate (Figure 1).

Isomerization is closely related to deamidation. Here the transformed residues are either aspartate or glutamate, again with aspartate being more critical due to the energetically favored five-ring intermediate. Just as for deamidation the subsequent amino reacts with the carbonyl carbon and the same cyclic intermediates are formed - consequently the products of the hydrolysis of the cyclic intermediate are the same as for deamidation (Figure 1)

The consequence of the deamidation and isomerization reactions can be manifold. If the respective amino acid is involved in antigen binding substantial potency loss may result. But even if the transformation is neutral with respect to antigen binding it might be unfavorable for protein stability and accelerate other protein degradation pathways, e.g. by reducing the conformational stability.

Oxidation primarily affects the sulfur containing amino acids cysteine and methionine, but also aromatic amino acids like tryptophan. Free cysteines should generally be avoided since cysteine is the most reactive of all proteinogenic amino acids undergoing a variety of chemical reactions. For oxidation sensitive amino acids the probability of oxidation and hence the rate at which the reaction occurs correlates with solvent accessibility. This principle can be intuitively understood as solvent accessibility increases the probability of forming contacts with dissolved oxygen and catalysts. The degradation products for methionine are the respective sulfoxide and the sulfon. Tryptophan most frequently is transformed to Kynurenine which itself is reactive and can undergo a variety of subsequent reactions.

The consequences of oxidative degradation can be the same as for deamidation and isomerization. Affinity to the antigen can be lost and protein stability compromised. In addition, all chemical transformations create new epitopes which in principle could be immunogenic.⁸ Therefore, antibody optimization efforts aim at detecting potential sites susceptible to accelerated chemical degradation and to engineer sites with an increased risk to sequences with more favorable properties.

Besides chemical degradation reactions, aggregation is the second dominant antibody degradation pathway (see Figure 2). Antibody aggregation, which denotes the formation of high molecular weight particles, is a problem very frequently encountered in antibody development.^9,10 Aggregation is not a well-defined process leading to a well-defined product. In fact, aggregation pathways can be highly diverse. One reason for accelerated aggregation is reduced conformational protein stability. Proteins exist in equilibria between folded, partially unfolded and unfolded states where the population of the individual states is determined by the folding free energy differences between the individual states. In less stable proteins the fraction of partially or completely unfolded protein is higher than in stable proteins. In these non-native states hydrophobic residues, which in the folded state are buried in the protein core, become exposed to the solvent and, when coming into contact with unfolded parts of other molecules, tend to form disordered aggregates.¹¹

 Figure 2. Interdependence of antibody degradation pathways: Chemical modifi cations can compromise conformational stability which results in a higher fraction of partially unfolded states. Partially unfolded antibodies are prone to form disordered aggregates but also fi brilic structures if sequence stretches with high propensities for beta-sheet aggregation are present. Ultimately, aggregates grow irreversibly into high-molecular weight particles which eventually precipitate.

A second aggregation pathway, which has been shown to be relevant for antibodies, involves the formation of cross-beta-sheet fibrils.¹² Certain amino acid sequences have a strong tendency to align with their counterparts from other molecules in a highly ordered fashion forming fibril structures similar to those known from protein deposition diseases like Alzheimer’s Disease, Parkinson’s and Huntington’s Disease.

Thirdly, self-association through hydrophobic or charged surface patches can lead to formation of oligomers. With increasing protein concentrations oligomer formation can promote the formation of disordered aggregates. In addition, self-association fosters high viscosities at high antibody concentrations.^13,14

Risk Mitigation Strategies

A fundamental difference between small molecule drug discovery and biologics projects is that aggregation-related problems outlined above only become evident when the first efforts are made to produce the new biologic in larger quantities and prepare solutions with high protein concentrations. At this stage drug discovery projects are already quite advanced and changes in the sequence would denote a substantial setback and delay. It is therefore critical to address developability-related issues early on through selection of sequences with a minimum of potential liabilities, through removal of potential risk factors during sequence optimization and to have research and development in close alignment at all stages of this process. Optimization strategies for biologics utilize a mixture of complementary experimental methods and computational analyses to detect such potential risk factors.

The property which is ultimately to be optimized is long-term stability, also commonly referred to as shelf-life. However, it is usually not practicable to run long-term stability experiments for multiple candidates as these experiments consume substantial amounts of material and last several months. Therefore, in-silico technologies have become a valuable resource to assess antibody developability. Computational approaches benefit from the tremendous progress that has been made over the last years in sequencing technology, hardware and software development, and the wealth of structural information from more than 1,000 antibody Fab structures deposited in the protein data bank.^15,16

The availability of a large number of antibody structures allows researchers to build very accurate 3-dimensional models of lead candidates, allowing for precise prediction of the position of every amino acid.^17-19 This is of particular importance for sites of potential chemical degradation like deamidation and oxidation sites since solvent accessibility and local protein flexibility, two important factors, can only be reliably estimated if a 3-dimensional model of a protein is available.

A second important source of information are antibody sequences of which thousands are available in public databases.²⁰ Since antibody variable domains are highly diverse in their sequences, huge sequence databases are an invaluable source of information which allow for rigorous statistical analysis of antibody sequences and to suggest somatic mutations which might be potentially detrimental for the conformational stability and thus could potentially compromise long term stability.²¹

Detecting sequence patterns which are prone to form ordered cross beta sheet aggregates and rationally engineer these sequences has also been shown to improve antibody developability. Here modern biotechnology benefits from the decades of basic research which have been invested in understanding protein aggregation on the atomistic level and algorithms which have been developed to predict aggregation tendencies of proteins and peptides.^{9, 12}

Hence, a variety of computational methods are available to identify potential risk factors for successful antibody development and to complement experimental approaches.²² The attractiveness of computational methods is that they pinpoint potential liabilities to distinct sequence patterns and therefore can directly guide protein engineering towards improved developability properties. It has been shown in a variety of studies that single or very few changes in protein sequences can have dramatic effects on expression, aggregation and thermal stability, thereby leaving biological function untouched.²³ It is important to point out, however, that computational approaches/ methods are still in the early stages of application to biologics, and more work is necessary to establish the generalizability and ultimate robustness of these approaches.

As soon as initial quantities of a biologic become available, information from experimental methods will further inform about the developability profile of a molecule. Generally, the challenge in defining predictive methods lays in identifying experimental conditions that are “representative” for a multitude of parameters that will be optimized during development. Commonly, a small set of e.g. solution conditions is chosen to explore the intrinsic antibody properties. The previously gained information from computational approaches is orthogonal and independent of experimental conditions. Thus, the combination of computational and experimental methods proves most powerful at the interface between research and development. Synergistically, all information is used to select the best candidate. In addition, flagging risk factors informs the developer about challenges that may occur during development.

As described above, aggregation propensity is linked to protein folding stability (or conformational stability) which is commonly studied in unfolding experiments. Such experiments are well established and monitor the transition from the folded to the unfolded state triggered by temperature or a denaturant. Transition midpoints (such as “melting temperatures”) are derived and provide a relative measure for the conformational stability. Due to low material consumption and a good signal-to-noise ratio spectroscopic or calorimetric means are frequently used to probe the transition.

Temperature stress is also exerted to monitor the transition of the biologic from the native to the aggregated state. This transition displays the macroscopic net-effect of consecutive microscopic steps. Thus, it also provides valuable information on aggregation resistance.

For antibodies to aggregate also their self-association tendency is crucial. Self-association (also in the folded state) is mediated by hydrophobic patches and surface charges.^{24, 25} Exposed tryptophans contribute to the surface hydrophobicity of antibodies and the degree of exposure can be estimated from the tryptophan fluorescence spectrum. The net-effect of antibody self-association can be analyzed with methods borrowed from colloid science. In contrast to the specific binding events for enzyme-substrate complexes or antibody-antigen complexes, the self-association of biologics is a weak attraction.

Therefore, the classical binding assays in biochemistry widely fail to quantify self-association events. Instead, light scattering experiments, analytical ultracentrifugation and self-interaction chromatography are used to determine interaction parameters such as the kD and the osmotic second virial coefficient.

The tendency for chemical degradation can be studied in accelerated stability studies. Here degradation is accelerated by choosing temperatures higher than the intended storage temperature and degradation products are analyzed typically after a few days. Mass spectrometry is most powerful in identifying the specific amino acids prone to chemical degradation. However, the method requires substantial effort and investment. Chromatographic methods allow for a simplified analysis that however does not provide specific information about the degrading residues. For both methods data analysis and the interpretation of results is facilitated when the residues prone to degradation have already been pre-selected by computational tools.

Conclusions

The synergistic use of computational and experimental methods working with small of amounts of material is a powerful way to support candidate selection and to provide valuable insights for formulation development. Moreover, computational methods can be used to map undesired properties to distinct sequence patterns which can be used to guide protein engineering for improved long-term stability and to identify developability risks.

The synergistic use of computational and experimental methods working with small of amounts of material is a powerful way to support candidate selection and to provide valuable insights for formulation development. Moreover, computational methods can be used to map undesired properties to distinct sequence patterns which can be used to guide protein engineering for improved long-term stability and to identify developability risks.

Acknowledgements

We would like to thank Julia Kasper for fruitful discussions and careful reading of the manuscript.

References

1. Beck, A., et al., Characterization of therapeutic antibodies and related products. Anal Chem, 2013. 85(2): p. 715-36.
2. Cordoba, A.J., et al., Non-enzymatic hinge region fragmentation of antibodies in solution. J Chromatogr B Analyt Technol Biomed Life Sci, 2005. 818(2): p. 115-21.
3. Xiang, T., et al., Structural effect of a recombinant monoclonal antibody on hinge region peptide bond hydrolysis. J Chromatogr B Analyt Technol Biomed Life Sci, 2007. 858(1-2): p. 254-62.
4. Cohen, S.L., C. Price, and J. Vlasak, β−Elimination and Peptide Bond Hydrolysis: Two Distinct Mechanisms of Human IgG1 Hinge Fragmentation upon Storage. Journal of the American Chemical Society, 2007. 129: p. 6976-6977.
5. Robinson, N.E., Protein deamidation. Proc Natl Acad Sci U S A, 2002. 99(8): p. 5283-8.
6. Stephenson, R.C. and S. Clark, Succinimide formation from aspartyl and asparaginyl peptides as a model for the spontaneous degradation of proteins. Journal of Biological Chemistry, 1989. 264(11): p. 6164-70.
7. Sydow, J.F., et al., Structure-based prediction of asparagine and aspartate degradation sites in antibody variable regions. PLoS One, 2014. 9(6): p. e100736.
8. Filipe, V., et al., Immunogenicity of different stressed IgG monoclonal antibody formulations in immune tolerant transgenic mice. MAbs, 2012. 4(6): p. 740-52.
9. Agrawal, N.J., et al., Aggregation in protein-based biotherapeutics: computational studies and tools to identify aggregation-prone regions. J Pharm Sci, 2011. 100(12): p. 5081-95.
10. Weiss, W.F.t., T.M. Young, and C.J. Roberts, Principles, approaches, and challenges for predicting protein aggregation rates and shelf life. J Pharm Sci, 2009. 98(4): p. 1246-77.
11. Costanzo, J.A., et al., Conformational stability as a design target to control protein aggregation. Protein Eng Des Sel, 2014. 27(5): p. 157-67.
12. Fernandez-Escamilla, A.M., et al., Prediction of sequence-dependent and mutational effects on the aggregation of peptides and proteins. Nat Biotechnol, 2004. 22(10): p. 1302-6.
13. Li, L., et al., Concentration dependent viscosity of monoclonal antibody solutions: explaining experimental behavior in terms of molecular properties. Pharm Res, 2014. 31(11): p. 3161-78.
14. Tomar, D.S., et al., Molecular basis of high viscosity in concentrated antibody solutions: Strategies for high concentration drug product development. MAbs, 2016. 8(2): p. 216-28.
15. Dunbar, J., et al., SAbDab: the structural antibody database. Nucleic Acids Res, 2014. 42(Database issue): p. D1140-6.
16. Dunbar, J., et al., SAbPred: a structure-based antibody prediction server. Nucleic Acids Res, 2016.
17. Almagro, J.C., et al., Antibody modeling assessment. Proteins, 2011. 79(11): p. 3050-66.
18. Teplyakov, A., et al., Antibody modeling assessment II. Structures and models. Proteins, 2014. 82(8): p. 1563-82.
19. Gilliland, G.L., et al., Leveraging SBDD in protein therapeutic development: antibody engineering. Methods Mol Biol, 2012. 841: p. 321-49.
20. Lefranc, M.P., et al., IMGT, the international ImMunoGeneTics information system. Nucleic Acids Res, 2009. 37(Database issue): p. D1006-12.
21. Seeliger, D., Development of Scoring Functions for Antibody Sequence Assessment and Optimization. Plos ONE, 2013. 8(10): p. e76909.
22. Buck, P.M., et al., Computational methods to predict therapeutic protein aggregation. Methods Mol Biol, 2012. 899: p. 425-51.
23. Seeliger, D., et al., Boosting antibody developability through rational sequence optimization. MAbs, 2015. 7(3): p. 505-15.
24. Chennamsetty, N., et al., Design of therapeutic proteins with enhanced stability. Proc Natl Acad Sci U S A, 2009. 106(29): p. 11937-42.
25. Lauer, T.M., et al., Developability index: a rapid in silico tool for the screening of antibody aggregation propensity. J Pharm Sci, 2012. 101(1): p. 102-15.