Mixing antibodies in silico: is more better?

“Science is imagination in the service of the verifiable truth, and that service is indeed communal. It cannot be rigidly planned. Rather, it requires freedom and courage and the plural contributions of many different kinds of people who must maintain their individuality while giving to the group.”

Gerald Edelman – Nobel Speech, 1972

The pioneering work of Gerald Edelman and Rodney Porter elucidating the chemical structure of antibodies, published in 1959, set the tone for decades of antibody research and breakthroughs. Edelman and Rodney were duly awarded the Nobel Prize in 1972. Antibodies are Y-shaped proteins produced by specialized blood cells to recognize and neutralize foreign antigens that are part of a pathogen. Antibody-antigen recognition follows a highly specific lock and key-like mechanism. Not only are antibodies an important component of immune surveillance, they have increasingly become the basis of various clinical and commercial applications. Thanks to their strong affinity and exquisite specificity, antibodies are the fastest growing classes of therapeutic drugs, mostly in a monoclonal form (meaning a single antibody rather than a combination).

The success of monoclonal antibodies as therapeutic agents raises the question of whether antibody mixtures can synergize to produce a superior response. The recent decades have witnessed tremendous progress in the development of successful combination therapies, with combination anti-retroviral therapy against HIV or chemotherapeutic cocktails in the case of cancer treatment far surpassing the results of any single drug. Understanding the mechanisms by which viral proteins evolve is essential for the development of therapeutic antibodies that efficiently target and neutralize viruses. The Bloom Lab (Basic Sciences Division) investigates the evolution of proteins and viruses. Employing a combination of experimental and computational approaches, they study how viruses like influenza evolve.

Dr. Tal Einav is a postdoctoral fellow in the Bloom lab, and his background is in physics. Einav and his mentor Dr. Jesse Bloom recently published a paper where they describe a computational model that predicts complex interactions between antibodies and their influenza targets. Their findings appeared in a recent issue of PLOS Computational Biology.

Einav reflects on the story behind the project. As a physicist gearing up to join a viral immunology lab, he felt he had some reading to do. “As I started reading papers on antibody combination therapies, I became fascinated by the body of evidence built around single antibody efficacy,” said Einav. In one study, researchers systemically tested the ability of dozens of single antibodies to neutralize flu, each showing a different level of potency. What about combinations of these potent antibodies?

In the past, attempts to use antibody mixtures and compare their efficacy have been hampered by the sheer number of potential combinations that makes it challenging to systemically test them in the lab. Realizing the problem, Einav wanted to come up with a computational model that can help scientists determine the most promising combinations to test empirically. “When I showed up on my first day, I showed Bloom a few back-of-the-envelope calculations, where over one hundred antibody combinations can be accurately predicted using my model.” Einav and Bloom worked together to refine their framework, testing it out against a cancer-causing receptor as well as on antibodies targeting the influenza virus. “And there are plenty of other studies out there. This was intended as a fun mini-project to inspire other groups to utilize both theory and experiment in their future work,” added Einav.

The epidermal growth factor receptor (EGFR), which is aberrantly activated in cancer, has been the subject of targeted antibody therapy, with two monoclonal antibodies already approved by the Food and Drug Administration for clinical use. A previous study attempted to systemically test the effectiveness of 10 different monoclonal antibodies to inhibit EGFR alone or in combination. “With 10 antibodies, you can make over 1000 possible two-antibody combinations at varying ratios, so trying to test all of them experimentally is a heroic undertaking,” said Einav, who started thinking whether there was a more efficient way to tackle this problem. Once he developed his computational model, he was happy to see that he was able to reproduce the findings of that study with high accuracy. Indeed, the model was able to predict the activity of two- and three-antibody mixtures based solely on the combination of the behavior of single antibodies and their epitope mapping.

The power of their model lies in its ability to make a large number of predictions based on a limited amount of data. As Einav explains: “while I was reading the literature, one thing that struck me was that it is hard to get a baseline expectation for how these combinations should behave. If a single antibody is 20% effective and another antibody is 50% effective, what would we expect their combination to be? Should there be an additive or a multiplicative effect? Will we be surprised if we find that their combination is 60% or 70% or 80% effective? Without this basic intuition, it is difficult to know which combinations to choose or how to judge the resulting experiments.”

Einav is optimistic about their model. “Combination therapy is becoming an increasingly important therapeutic tool. It is nice to have a framework to accurately predict what would happen if two antibodies are combined, before doing all the experiments. Once you start thinking about mixing multiple antibodies at different ratios, there is a nearly unimaginable number of potential combinations, and you start to appreciate the utility of our model.”

Einav T, Bloom JD. (2020). When two are better than one: Modeling the mechanisms of antibody mixtures. PLOS Comput Biol. https://doi:10.1371/journal.pcbi.1007830

This work was supported by funding from the Fred Hutchinson Cancer Research Center, the Fred Hutchinson Computational Biology Mahan Fellowship, and the National Institutes of Health and the Howard Hughes Medical Institute.