A Shift in Perspective

Anton Simulations Upend Picture of How Proteins Work 

The Anton simulations suggest that PKA has nine distinct communities (A through H). The size of the circles shows how big each community is physically, the width of the lines connecting communities show how strongly each one’s movements affect another.

We’ve all seen the optical illusion: At first, it’s clearly a picture of a white vase against a black background. But then your perspective changes and it becomes two black faces against a white background. In an instant, your understanding of the image is upended. 

Using simulations on the Anton supercomputer at PSC, researchers at the University of California, San Diego (UCSD), have had a similar experience that may upend how scientists understand protein structure. The research may open new opportunities for correcting faulty protein function in diseases as diverse as cancer, diabetes, neurological disease, inflammatory disease and more. 

Researchers led by Susan Taylor at UCSD have simulated the motion of protein kinase A (PKA). Their simulations, reported in the Proceedings of the National Academy of Sciences USA, revealed how the seemingly obvious sub-structures of the protein— “domains”—did not play the central role in function and movement that researchers had expected. 

Acid components that do move and act together have their own boundaries that don’t seem to relate to the ones researchers’ eyes had defined. This finding questions whether domains are actually relevant to the protein’s function and suggests that, to understand how proteins work, movement is as important as structure. 




The UCSD researchers chose PKA in part because previous research had provided a great amount of information about the protein, including its function and its three-dimensional structure. PKA is also medically important, acting in blood sugar regulation and diabetes, and has also served as a model for a family of signaling proteins called protein kinases that touch virtually every biological function in health and disease. 

Protein structure, researchers once assumed, dictated function in a very straightforward way. Using X-ray crystallography and other methods, researchers had produced static “snapshots” of proteins that show how a chain of amino acids folds to make obvious smaller-scale structures—domains—that then form the larger protein. The assumption was that these domains would carry out parts of the protein’s functions. 

In addition to understanding how domains worked, researchers wanted to understand a related phenomenon called allostery. When researchers change some specific amino acid in the chain of amino acids that makes up PKA, it could turn off an action carried out on the opposite side of the protein. 

Somehow the change was transmitted across the protein, and researchers suspected that the domains somehow communicated with each other. The real puzzle was how these changes often made no visible difference in the protein’s structure. 

“I mean, the kinase activity is broken in these mutants,” says Alexandr Kornev, a project scientist at UCSD and collaborator in the work. “You have the structure, it looks the same, but it behaves completely differently.” 

The difference had to be in how these very similar structures moved differently. The Anton supercomputer at PSC, the UCSD group reasoned, might help shed light on the mystery. 



Anton is a special-purpose supercomputer hard-wired to simulate the movements of large molecules using molecular dynamics. Unlike general-purpose supercomputers, Anton simulates large molecules at rates of microseconds per day. The Anton machine hosted at PSC was developed and made available without cost by D. E. Shaw Research. 

“On Anton, our simulation took about a day of full-time use of the machine,” says Christopher McClendon, a postdoctoral fellow in Taylor’s laboratory and first author in the study. “On another resource it would take about five months to a year.” 

In the case of PKA, Anton allowed the researchers to push their simulations to 5 microseconds—a timeframe that revealed a surprising set of behaviors. 



The group’s Anton simulations showed that groups of amino acids in PKA tended to move in concert as the protein changed its shape. The shocker, though, was that the communities identified by the simulations and the domains identified from the static structures were not one and the same. The communities had their own boundaries, which didn’t seem to honor the “obvious” boundaries of the domains. 

“We discovered eight communities of amino acids,” says Taylor, the group’s principal investigator. “What was so striking was that each community didn’t segregate by the structural elements we had previously understood but by functional units.” 

More interesting, these communities interacted with each other to varying extents. Some affect each other strongly; others, weakly or not at all. Here, it seemed, was a potential explanation for allostery. An amino-acid substitution that makes no obvious change in the protein’s static shape nevertheless transforms how that amino acid’s community moves. Indirectly, 

changing that movement in turn modifies the movements of other communities, sometimes on the other side of the protein. 

“This is really important in drug discovery and provides an alternative route to identifying novel therapeutics,” McClendon says. 

For example, a drug that affects an amino acid in a certain community far from the active site might be able to shut down an unwanted function carried out by that community without interfering with desired functions. 

“If you can make more subtle changes distant from the active site, it might not have as many side effects as if you just totally gum up the works,” he adds. 

“This finding opens the door to a mechanistic understanding of many interactions and mutations that were completely obscure,” says Kornev. “The communities … look very clear and logical, knowing how the kinase works. That gives us confidence that we are onto something very promising.”