By Yahya Chaudhry
When a virus mutates, the world usually finds out the hard way. Variants of influenza and SARS‑CoV‑2 can slip past vaccines and antibody drugs before scientists fully understand what has changed.
A new study from the Department of Chemistry and Chemical Biology (CCB) suggests the hard way may not be inevitable.
In a paper published recently in the Proceedings of the National Academy of Sciences (PNAS), a team led by Eugene Shakhnovich , Roy G. Gordon Professor of Chemistry, and Vaibhav Mohanty, an M.D./Ph.D. student at the Harvard Kenneth C. Griffin Graduate School of Arts and Sciences and Harvard-MIT Division of Health Sciences and Technology, describes a way to engineer the evolutionary landscape that viruses experience, pushing them away from dangerous variants and into evolutionary dead ends.
They call the method fitness landscape design (FLD).
"It's basically the inverse of how people usually think about evolution," Mohanty said.
Biologists have long used a fitness landscape metaphor: Each viral sequence is a point on the terrain, and its "height" reflects how well it reproduces and spreads. Evolution is like a population of viruses climbing uphill toward higher fitness.
Most research takes that landscape as fixed and asks whether a given mutation will raise or lower a virus's fitness. Mohanty and Shakhnovich flipped the question: Can we shape the landscape itself by choosing antibody combinations that make most escape routes unviable?
"Usually scientists ask, 'If this mutation happens, how will it affect viral replication?'" Mohanty said. "We're asking, 'Can we directly suppress the mutations we're worried about and fence off those paths before they appear?'"
That ambition reflects recent frustrations. The World Health Organization updates flu vaccines twice a year, yet surprise strains still emerge. SARS‑CoV‑2, the virus that causes COVID-19, has produced a steady stream of subvariants that chip away at vaccine protection.
"Viruses are constantly mutating to evade our immune systems," Mohanty said. "Our vaccines and antibodies end up playing catch‑up."
The team built a biophysical model that starts with molecular details and ends with a number for viral fitness. It describes how a virus's surface protein binds to human cell receptors and antibodies, and how mutations change those bindings. Standard reaction equations then translate binding strengths into infection probabilities and growth rates.
"The fitness landscape is just a mapping from sequence to how well the virus spreads," Shakhnovich said. "What we're doing is designing that mapping through antibodies."
To test the accuracy of the model, the team first turned to murine norovirus, a common lab pathogen previously evolved in flasks with and without a neutralizing antibody. Using structural models, they estimated how each mutation affected binding energies, fed those values into the fitness equations, and compared results with the actual rise and fall of viral strains. The model's predictions closely matched the experiments.
They then applied the same framework to SARS‑CoV‑2, combining measurements of how spike variants bind to the human ACE2 receptor and antibodies with epidemiological estimates of each variant's success.
"It's fundamentally multi‑scale," Shakhnovich said. "We start from proteins, antibodies, receptors, and then look at consequences at the level of pandemics."
That blend is a hallmark of his lab.
"We fuse biophysics and evolution," Shakhnovich said. "There are not so many labs that do both molecular biophysics and population‑level evolution in one place."
With the model calibrated, the group tackled the heart of FLD: the inverse problem. Instead of asking how evolution would proceed on a given landscape, they started by specifying a desirable landscape — such as one where an entire network of related SARS‑CoV‑2 variants has suppressed fitness — and then searched for antibody ensembles that produce it.
In simulations, certain antibody cocktails could trap viral evolution, turning routes that would normally lead more fit "escape" variants into paths that end in low‑fitness valleys. In this aspect, FLD works like a chess bot, calculating several moves ahead of viral evolution and finding the best antibodies to suppress fitness gains before the mutations even appear.
"Instead of reacting to whatever mutation shows up next, we're trying to proactively corral the virus so the easy escape paths are bad for it," Mohanty said.
Beyond infectious disease, the authors see wider uses for this framework. In cancer immunotherapy, treatments such as CAR‑T cell therapy arms immune cells with engineered, antibody‑like receptors to target tumors that themselves evolve.
"We think FLD might help better design those immune defenses, so cancers have fewer escape routes," Mohanty said.
In industrial biotechnology, scientists can use a process called directed evolution to improve enzymes and other proteins for medicines and manufacturing. They make small changes to a protein, select the versions that work best, and repeat the process over several rounds.
However, this trial-and-error approach can get stuck. Scientists may find a protein that is better than the starting version but still far from the best possible one.
The researchers think fitness landscape design could help guide evolution around these dead ends and toward proteins with more useful properties. The team is also using artificial intelligence — including protein language models and generative tools — to predict how antibodies will bind to viral proteins and search through many more possible antibody designs.
"The idea of a fitness landscape has been around for a long time," Shakhnovich said. "What's new is that we can start to manually mold that landscape to fight disease."
If that vision holds experimentally, it could give drugmakers and public‑health officials a potent weapon in the viral arms race.
This work was supported by the National Institute of General Medical Sciences, R35GM139571.
