Engineering Fit-for-Purpose Antibodies

This article includes excerpts from a recent Bench Tips webinar organized by Biocompare on Antibody Engineering: Strategies for Design & Optimization. We recommend watching the entire webinar for a complete understanding and context of what was discussed.

Antibodies are proteins produced by plasma cells or B cells in response to pathogens. They can exert their function by blocking or neutralizing their target, as well as by triggering parts of the immune system through processes such as antibody-dependent cellular cytotoxicity, phagocytosis, and complement-dependent cytotoxicity. Due to the exquisite affinity and specificity to their targets, antibodies can be used as detection agents in many applications to look for the presence or absence of their target such as in flow cytometry, western blot, or ELISA. As an important part of the adaptive immune response, antibodies are also used as therapeutics to activate or inhibit their target, as well as to deplete cell types that are contributing to disease progression.

According to a recent article in Nature Reviews Drug Discovery, of the 55 drugs approved by the FDA in 2023, 12 were antibodies. Antibodies also constituted 6 of the top-10 selling drugs in 2023. Recent advances in antibody therapeutics have resulted in new modalities such as antibody drug conjugates (ADCs) and bispecifics. Although there are a lot of different antibody isotypes, scientists usually work with IgG antibodies, which have two heavy and two light chains, with variable regions on both chains and a constant region that contains the Fc domain. The commonly used IgG subclasses include human- IgG1, IgG2, IgG3, IgG4, and mouse- IgG1, IgG2a/c, IgG2b, IgG3.

Selecting & modifying antibodies

Antibodies are easy to work with as they can be customized for a specific application. Using the variable heavy (VH) and variable light (VL) chain sequences on the antibody, it can be converted into a number of different formats by changing the species, the isotype, and the subtype. While antibodies are generally used in their full-length IgG format, there are alternative formats like bispecifics, Fab, scFV, and VHH, which are smaller and more tractable depending on the context of use.

Some of the questions routinely asked when searching for a good antibody are:

• Where can I find the right antibody?

Literature, patents, and vendor resources are good starting points.

• How do I get my antibody to do what I want?

Antibodies can be sequenced and characterized to ensure that they interact with the target.

• If I don’t find the right antibody, how do I make my own?

Immunization, display systems, and computational design are three ways to generate custom antibodies.

There are different types of antibody modifications available to modulate the antibody function and fine-tune its properties.

Table 1. Modulating antibody properties via Fc engineering

Source: Absolute Antibody

The expression of four different chains can also result in high levels of mispaired products with the wrong heavy and light chains coming together; there are several approaches to avoid this and favor the optimal bispecific formation. Here are some of the modifications that are used:

Bispecific antibody mispairings

• Knobs-into-holes (S354C/T366W, Y349C/T366S/L368A/Y407V)
• Electrostatic steering (K409D/K293D, D399K/E356K)
• SEEDbodies: IgG-IgA crossover
• Fab arm exchange: post expression heterodimerization
• Crossmab: correcting light chain mispairing

Source: Cytiva

If you are unable to find an antibody that has the properties or cross-reactivity needed, then you have to discover your own antibody using phage display and other techniques. After getting the hits from the phage display library the antibody sequences are analyzed and specific residues are selected for affinity maturation using techniques like yeast surface display. Using in vitro display systems for de novo antibody discovery is a very reliable way to find suitable binders. The antibody is then expressed and purified in different formats and tested in different cells for experimental validation.

Affinity maturation for optimizing antibodies

Affinity maturation is a really powerful tool to improve or change the binding interactions of antibodies.

• It uses directed evolution, which allows for researcher directed tuning of affinity, and is particularly effective when modulating existing binding interactions.
• It relies on a genotype-to-phenotype linkage to assign protein function to sequence.
• Typically performed over iterative, increasingly stringent rounds of sorting or panning
• Can be used to engineering binders with higher affinities, cross-reactivity, or binding selectivity

There are many techniques and yeast surface display is an optimal technique for affinity maturation and engineering proteins. It can help determine both the expression and binding of the protein of interest using specific tags and monitoring and sorting using flow cytometry. Library generation is an important part of the affinity maturation process.

Table 2. Mutagenesis technologies for affinity maturation

Source: Sean Yamada-Hunter, Ph.D., Postdoctoral Fellow, Stanford University

Mutagenized libraries can then be sorted over multiple, increasingly stringent rounds to enrich for high-affinity binders using techniques like magnetic activated cell sorting (MACS) or fluorescence activated cell sorting (FACS). MACS is optimal for bulk sorting on large libraries, while FACS is best for fine-tuned selection of specific cell populations to find good or selective binders for proteins of interest.

Computational protein design

Computational protein design is affinity maturation done in silico. The holy grail of antibody design is to accurately, efficiently, and reliably predict the sequence of antibody that will bind a given epitope with high affinity and high specificity. Realistically, it only ends up giving a good starting point for affinity maturation that then leads to finding the best antibody. The tasks that are associated with computational modeling of antibodies typically include structure prediction, protein-protein docking, loop modeling, protein design, and modeling with experimental data.

Challenges of antibody (and protein) design include optimizing:

• Shape complementarity
• Chemical complementarity
• Solvation and solubility (interaction with water)
• Structural flexibility

Overview of methods for antibody design

Given a crystal structure or sequence of the antibody, some methods that are used for structure predictions include AlphaFold2 (and AlphaFold3 which was recently released), ESMFold, RosettaFold, OmegaFold. These models ae readily available and provide a confidence level of the predicted structure.

Table 3. Structure prediction models

Source: Gina El Nesr, Doctoral Student, Stanford University

Designing the antibody involves:

• Antibody-target docking
• Loop design
• Sequence design

There are many methods available for predicting protein-protein interactions and docking. There are classical free energy-based calculation methods that are based on Rosetta. But the field is moving toward the use of machine learning tools like AlphaFold2. Then there are neural network methods for antibody design and docking that don’t have much experimental validation yet. There is also a lot of work being done with generative models for the de novo design of antibody structures. But it’s still early to tell how well they work.

After docking the antibody with the epitope there are many methods that can predict the sequence design. They all have their strengths and limitations and work differently for different types of proteins or for different regions on the protein such as flexible loops. Hence, it’s important to perform an in silico analysis of the predicted sequence to make sure it aligns with the structure prediction models.

• ProteinMPNN
• Frame2Seq
• EvoDiff
• 3DCNN
• gLM

Generative modeling for protein design

Generative protein design is an iterative process involving sequence, structure, and function analysis followed experimental validation that leads to feedback and further changes in the model. Some of the steps involved in generative modeling based on the development of Sculptor, a prediction tool developed in Dr. Possu Huang’s Lab at Stanford University, are listed below:

Step 1: Epitope selection

Step 2: Structure optimization

Step 3: Sequence design

Step 4: Filtering designs by in silico screening with AlphaFold2/ESMFold

• pLDDT
• pAE
• AF2 Rank composite score

Step 5: Binding design

Step 6: Experimental validation