The concept of viral escape from antibody neutralization seems intimately familiar in our post-COVID world—it’s the reason that we wait in line for new iterations of vaccines while dreadingthe inevitable arrival of new viral variants which can evade those vaccines. It’s a stark reminder that as our immune systems, scientists, and governments fight this virus, the virus fights back. In a recent preprint posted to bioRxiv, Timothy Yu, a graduate student in the lab of Dr. Jesse Bloom, and colleagues report efforts to predict viral escape from complex mixtures of neutralizing antibodies. In doing so, they hope to leverage state-of-the-art experimental and computational techniques to stay ahead in the arms race between virus and man, while potentially gaining new insight into how antibody mixtures interact with viral antigens on a fundamental level.
First, some vocabulary: antibodies are small proteins produced by our immune system whose job it is to bind viral proteins called antigens (for example, the spike protein on the surface of SARS-CoV-2) and neutralize or prevent them from invading our cells. To get more specific, any given antibody only binds a specific portion of its corresponding antigen—this region is called an epitope. We would like to imagine a simple scenario, whereby a viral infection causes your body to produce a single antibody type targeting a specific epitope, which the virus will slowly mutate to disrupt antibody binding and escape neutralization. However—as is usually the case in biology—reality is more complicated. Viral infection or immunization causes your body to produce a mixture of antibodies which recognize many different epitopes. While this is thought to increase the durability of anti-viral responses, we know from experience that viruses are still able to escape from these ‘polyclonal’ mixtures of antibodies by accumulating mutations in multiple antigenic regions (multiple epitopes). Understanding how viruses manage this escape—and developing tools to predict when they will—is of prime public health and basic science importance.
Methods exist to experimentally test whether a viral variant can lead to escape from antibody mixtures, but they are relatively low-throughput and laborious, as each variant needs to be tested individually—a tall order in situations when viral adaptation is rapid, and many different variants arise in the population. Crucially, these methods also rely on prior knowledge of the mutations to produce, which leaves us constantly ‘one step behind’ the virus we are trying to fight.
This is where Tim and colleagues step in with two key innovations. First, they leverage a state-of-the-art experimental technology pioneered by the Bloom Lab known as Deep Mutational Scanning (DMS). In short, DMS allows experimenters to systematically mutate every amino acid (or small groups of amino acids) in a given viral antigen and test its subsequent neutralization by an antibody mixture. While this gives theoretically complete information about single escaping mutations, the prospect of testing every possible pair (or triplet, or quadruplet, etc.) of mutations in a viral protein with ~1000 amino acids using DMS is still far from feasible. To surmount this hurdle, Yu and colleagues developed a model of viral escape which can be ‘fit’ to an existing DMS dataset (which, remember, can only measure a small fraction of all possible combinations of viral mutations). In contrast to other recently developed machine learning-based approaches which model viral escape, Yu et al. used classical tools from biophysics to construct their model. While this difference might appear trivial to someone outside the field, Tim is quick to point out the strengths of their approach: “Instead of the kind-of ‘black box’ approach of machine learning where you get a result but don’t know how the algorithm got there, the parameters in our modeltell us something about the biology going on—which epitopes are targeted by antibodies in the mixture, how important each epitope is, and which mutations escape antibodies at each epitope.”
By fitting their model to existing DMS data, the team can computationally predict the extent to which a mixture of antibodies will neutralize a new viral variant with an arbitrary combination of mutations—importantly, a viral variant which was not tested as part of the original DMS experiment. In this way, they’ve developed a method to extend the functionality of any given DMS experiment by ‘extrapolating’ from its results—an important tool when you consider the challenge of keeping up with rapidly-evolving viruses of public health concern. While the team’s method performed well on simulated DMS data and holds great promise for continuing use and development, they were open and honest about some of the limitations. “We’re super excited that it worked on simulated data, but we’re even more excited to apply it to some real combinatorial DMS datasets, which are still a bit few and far between,” Tim notes. “Even though we make some assumptions to simplify the modeling, I’m confident it’ll be useful and teach us something about biology in the process. In a way, looking for agreement between our model and experimental outcomes serves as a test of some long-standing hypotheses in the field regarding viral neutralization and escape, and I’m sold on the idea that a biophysically grounded modeling approach holds great promise in that regard.”
Your move, viruses.
Fred Hutch/University of Washington/Seattle Children’s Cancer Consortium member Erick Matsen contributed to this study.
The spotlighted research was funded by the National Institutes of Health, the National Science Foundation, and the Howard Hughes Medical Institute (HHMI).