Turning Therapeutic Antibodies into Better Drugs
A computer algorithm manipulates antibodies to predict which forms will bind their targets more tightly
The word “antibody” conjures images of our bodies fighting off bacteria and viruses. But because they can latch onto their targets with great precision, antibodies are also used to treat non-infectious diseases such as cancer. Researchers at Massachusetts Institute of Technology have now designed a computer algorithm that manipulates antibodies to predict which forms will bind their targets more tightly. These predictions are then confirmed in the laboratory. The group’s work could lead to significantly improved antibody- based drug design.
“Part of the effectiveness of an antibody-based drug is related to how tightly it binds its target,” explains Bruce Tidor, PhD, professor of biological engineering and computer science at MIT. He co-authored the work with K. Dane Wittrup, PhD, professor of chemical engineering and bio-engineering at MIT, and Shaun Lippow, PhD, the paper’s lead author and a joint graduate student of both Tidor and Wittrup at the time the work was done. The research was published in the October issue of Nature Biotechnology.
Like all proteins, antibodies aren’t rigid; they are more like Play-Doh than wooden building blocks. It doesn’t take much to affect an antibody’s shape. For example, substituting any of the amino acids strung together in a protein chain may alter its final folded shape markedly. Such changes, in turn, impact how strongly the antibodies bind to other molecules.
The biggest challenge in antibody-based drug design has been tweaking amino acid sequences to obtain that ‘just-so’ fit with the target. Traditional methods miss many possible amino-acid changes that might make the altered antibody bind more tightly. MIT’s approach, combining computational structure analysis and experimental lab chemistry, may provide the missing link.
The computer algorithm works by first modeling the physical interactions that make a particular antibody latch onto its target. It then rapidly identifies all possible amino-acid substitutions for that antibody and calculates which of those changes will tighten binding. Researchers can introduce mutations that improve antibody function but might never arise naturally or with conventional techniques, and they can predict the effectiveness of these mutations.
The researchers experimentally verified their model on a drug called cetuximab (trade name Erbitux®, used to treat colorectal cancer). With guidance from the computer program, they synthesized a new version that binds 10 times more strongly to its target, a molecule called epidermal growth factor receptor. They also created a revised version of an antibody (D44.1) that is useful in laboratory experiments. It has a 140-fold improvement in binding affinity.
“This represents an interactive collaboration between calculation and experiment,” says Tidor. “The ability to have tight feedback cycles between predictions and testing was essential to our success in this work.”
Janna Wehrle, PhD, program director of the biophysics branch at the National Institute of General Medical Sciences, is enthusiastic about the new model. “Dr. Tidor and his team have developed a method that will allow much of the design work to be done on the computer, saving months or years in the lab,” she says.