Engineering Antibodies: A FAb-ulous Future for Therapeutics


Engineering the elements of an antibody can create potential therapeutics—ones designed to be more effective and safer than ever. But this often requires optimization of not just selectivity and affinity for the given antigen, but also stability, solubility and immunogenicity of the antibody. The familiar Y-shaped structure includes a variable domain called the fragment antigen-binding (Fab) region. The entire Fab or fragments of it can be engineered to change an antibody’s capabilities, and focused research in this area is still evolving. “If there is any general trend, in terms of platform/format, it’s that the number of solutions to the same problem keeps increasing, and there still isn’t much convergence on a single platform,” says David Pearlman, director, biologics software platform, and senior scientist at Schrödinger.

For antibody-based therapeutics, today’s scientists tend to use monoclonal symmetrical antibodies, single-chain antibodies, asymmetric bispecific antibodies, antibody-drug conjugates (ADCs) and so on. For most of these, there is no consensus on how to make them. “In the area of bispecifics, which can simultaneously recognize two different targets, there are now dozens of published platforms,” Pearlman explains. “The proliferation of different proposed solutions arises both from the fact that all have their own clinical advantages and disadvantages and from the commercial need to avoid IP overlap and patent infringement.”

Despite juggling multiple approaches to engineering antibodies, scientists are starting to agree on one thing. “Engineering is critical in the discovery process to yield antibody drugs with the best chances for survival in the clinic,” says Pearlman. Nonetheless, scientists face some significant issues in getting this engineering right.

To move ahead in developing an antibody-based drug, scientists often face challenges from aggregation, immunogenicity and viscosity.

“As with the platform, there is not, as yet, convergence on an efficient and reliable approach that satisfactorily addresses these issues,” Pearlman says. “Some of these properties, such as viscosity and aggregation, are expensive and time-consuming to evaluate.” The immunogenicity can remain uncertain until testing in humans.

Make a wish

At the University of Virginia, Thomas Barker, biomedical engineering professor, and his team used directed evolution to engineer antibodies. Specifically, Barkers shares that they used this process to engineer antibodies “from a parent library following phage screening.”

Getting the best results depends on the complementarity-determining regions (CDRs), which are the molecules in Fab that connect the antibody to a target antigen.

And non-CDRs matter, too. Barker points out that better computational models would improve the ability to predict the non-CDR regions that affect an antibody’s specificity and affinity.

In testing, Barker could also use some improvement in analytical tools. For example, he would like to see “cheaper instrumentation for high-throughput kinetic binding analysis.”

So, there are few wishes for vendors to fulfill.

Computing the connections

Barker mentions the desire for better computational tools, and others agree. “There is growing interest in in silico approaches to assessing protein liabilities,” Pearlman says. “The advantage of such approaches is that they are relatively fast and inexpensive to run, that they can be included in workflows addressing hundreds or even thousands of potential therapeutic candidates, and that they often offer not only predictions that can be used for triage but also insights into where you might engineer changes into the protein to remove or reduce the liabilities.”

Like some of the analytical approaches to engineering antibodies, the in silico side remains in development. However, advances in computation promise ongoing improvements. EpiVax, for instance, offers its “EpiMatrix High Throughput Antibody Immunogenicity Prediction Report.” This uses in silico screening and other technologies to provide what the company calls “an overall assessment of potential clinical immunogenicity for a large set of antibody candidates.” At the University of Melbourne in Australia, David Ascher, group leader for structural biology and bioinformatics, and his team also use computational tools for guided affinity maturation and construct design. Benchmarking the currently available tools has shown that there is still significant room for improvement [1]. Recently, the Ascher group released mCSM-AB, which, Ascher says, outperformed all the currently available commercial and academic tools for in silico optimization; the team made it freely available through a fast and easy-to-use tool [2].

Schrödinger also keeps creating new approaches. As an example, Pearlman says, “We have been focusing on new in silico structure-based techniques that facilitate early-stage liability assessment.” To computationally screen a collection of antibodies, scientists need an easy and automated tool. So, Schrödinger developed an antibody-structure-prediction tool that is robust, fast and reliable. Pearlman explains, “The ability of this tool to provide good predictions was demonstrated in the recent Antibody Modeling Assessment II—a blinded evaluation where our fully automated approach did quite well” [3]. In addition, Charlotte Deane, professor of structural bioinformatics at the University of Oxford in England, and her colleagues recently developed SAbPred, which is a fast and freely available tool for antibody-structure modeling and epitope prediction [4].

Furthermore, Schrödinger is developing tools that predict the potential liabilities in new structures. To build a model like this, the company needed a large experimental dataset to use for training and testing the computational tool. “A huge amount of relevant experimental data exists but is hidden behind corporate firewalls,” Pearlman says. “Recently, Schrödinger entered into several substantial collaborations that have given us access to these hidden data, and this access is making it possible to develop predictive, validated tools for liability assessment.” In brief, this technique created an automated quantitative structure-activity relationship (QSAR) model engine that computes the possible liabilities from a predicted antibody.

Increasing opportunities

Ascher’s team also works on antibodies for therapeutic possibilities. Here, he says, a lot of work in collaboration with Lisa Kaminskas, a research fellow at the University of Queensland in Australia, “has focused on the optimization of biomolecules, including antibodies and Fabs, pharmacokinetic and pharmacodynamic properties, and for this we have found PEGylation to be very useful [5].” However, much remains unknown about the optimal modifications that are needed to obtain the biological properties required.

Beyond antibodies, scientists can also study antigens. By optimally identifying the antigen being targeted by the antibody, Ascher explains, you can “avoid off-target effects and avoid or minimize the development of escape mutations.”

Despite the challenges of engineering antibodies for therapeutics, the promising potential they hold will result in better medicine. “Antibody therapies have enormous benefits and represent a larger proportion of the approved drugs,” Ascher explains. “In fact, if you look at the top-selling therapies approved in the last 10 years, many of them are biological therapies/antibodies, as opposed to the traditional small molecules.”