Skip to content
MIT Better World

By Robin Kazmier SM ’17

They build its structure, convey information, and regulate gene expression. Their ranks include molecules as diverse as antibodies, enzymes, and hormones, so when something goes wrong with proteins, it can be a serious problem.

“Almost all diseases involve some sort of a malfunction of proteins,” says Amy Keating, the Jay A. Stein Professor of Biology and head of the Department of Biology.

In cystic fibrosis patients, for example, a protein called CFTR, which moves ions into and out of cells, is either lacking or malfunctioning. In the lungs, this failure leads to the development of mucus that clogs the airways. Some pathogens also rely on protein interactions to attack their hosts. Viruses like influenza, HIV, and the SARS coronaviruses gain entry into cells by binding to cell proteins. Blocking such interactions could block infection.

Having all this power packed into tiny proteins is fascinating to Keating, whose lab designs protein-protein interactions that could potentially be harnessed for drug therapies. Yet, she didn’t start out in biology; she was simply intrigued by the three-dimensional properties of molecules and the way they interact. Following her intellectual curiosity led her from an undergraduate focus on physics to a PhD in organic chemistry, and ultimately to work on protein interactions—a field where she could explore that interest further.

“I was always interested in inter-molecular recognition and self-assembly,” she says.

In the 20 years since she joined MIT, she has focused on the biology of protein interactions in nature and, more recently, designing protein interactions not found in biology.

Taking clues from nature

There are about 20,000 different proteins in the human body, circulating in any given cell by the millions. All start with the cell’s own DNA. Individual genes code for specific proteins, and the cell’s machinery assembles them from strands of amino acids. These strands then fold in on themselves, settling into specific 3-D shapes as bonds form between different parts of the strand. This structure is key to the protein’s function, including how it recognizes and binds with other proteins in interactions that underlie myriad biological processes.

Keating’s lab examines groups of related proteins to learn more about how they do their jobs. “Every eight to ten years we pick an interesting new protein family whose interactions are important for cell decisions in biology and do a deep dive,” she says. In recent years, that has included Bcl-2 proteins, which help regulate programmed cell death, a process by which a cell whose DNA is damaged beyond repair kills itself. If Bcl-2 proteins can’t trigger cell death, the damaged cells can continue reproducing, potentially leading to cancer.

Proteins play countless roles, and Keating is particularly interested in creating proteins that can block interactions that contribute to disease. In other words, designing something that binds with Protein A to prevent it from binding with Protein B, thereby interfering with the basic processes that allow disease to progress.

For Keating, like other biological engineers, designing new protein interactions started with building on those already known from nature. “We would choose to study protein interactions where there was already information about the geometry of how they interact,” she says. With that information in hand, they could use a biochemical approach to change one or both of the proteins to alter that interaction—perhaps making the pair bind more tightly, or changing a protein that naturally interacts with several others so that it will interact with only one.

If you’re trying to make therapeutics, for example, “you might want to make a tight-binding but very selective inhibitor of just the protein that’s problematic in disease,” Keating says. “You don’t want it going off and binding a bunch of other proteins that are actually good for the cell.” Keating refers to this kind of tweaking as “redesign.”

From redesign to de novo design

The revolution in DNA sequencing has made it possible to synthesize and lab-test proteins on a larger scale, and advances in machine learning have dramatically improved researchers’ ability to predict protein structure and expand the library of known structures. So, Keating’s lab is now trying to move into de novo design of proteins, she says. Basically, “if you don’t have a good solution already from nature, can we invent one?”

This is becoming increasingly possible with computational tools such as DeepMind’s AlphaFold, an artificial intelligence program that, with a high degree of accuracy, predicts a protein’s structure on the basis of its amino acid sequence. The dream, Keating says, would be to start with nothing but the DNA sequence of a protein and design “something that would block it, or bind to it, or detect it and inhibit it.”

One approach her lab has taken is to focus on small sections of proteins. While whole protein structures are incredibly varied and complex, she says, “they’re all built out of different combinations of the same pieces.” Now, rather than looking for entire proteins that will bind, her lab is using algorithms to search known protein structures for examples of smaller surface fragments that complement each other, which they call “seeds.” They then consider whether such seeds could be stitched together into an actual protein molecule. They’ve been successful in these initial steps.

“We know they look pretty good,” Keating says. “If we stitch them together, they have a lot of the properties of natural proteins.” Lab testing has yet to prove whether this is an effective technique, but Keating is optimistic about the momentum of computational tools in the field.

“The rate of progress has just accelerated I think a hundredfold from what it was just two years ago,” she says. “Great things are going to happen.”