Directed evolution—a process in which a protein library is subjected to repeated rounds of mutation and selection to achieve desired properties—can improve the catalytic performance of computationally designed enzymes. From a structural point of view, the evolved enzymes are often similar to the designed enzymes, so it has been unclear why the evolved enzymes are so much better.
Dorothee Kern of Brandeis University, Donald Hilvert of ETH Zurich, and coworkers may now have an explanation for the activity differences. The directed evolution process selects for small protein changes that increase the probability that the enzyme will adopt highly active conformations in its active site.
The team focused on a Kemp eliminase that Hilvert’s group had evolved from an enzyme designed by Stephen L. Mayo’s group at the California Institute of Technology. The evolved enzyme, reported in 2013, is more than 1 billion times as catalytically active as the original designed enzyme.
To figure out what was going on, the researchers used X-ray crystallography, nuclear magnetic resonance, and kinetic methods to characterize the original designed enzyme, the enzyme after seven rounds of directed evolution, and the optimized enzyme, which had gone through 17 rounds (Science 2020, DOI: 10.1126/science.abd3623). They found two folded conformations of the optimized enzyme—one active and the other inactive—that interconvert on the hour timescale.
The researchers found the same inactive form in the original designed enzyme, but a larger fraction of the population of individual molecules was in that conformation compared with the evolved enzyme. Directed evolution increased the population of an already-present but less-populated active conformation. The inactive conformation turned out to be the same as the xylanase that served as the scaffold for the enzyme design. The difference between the active and inactive conformations, Kern says, is the flipping of several β strands.
The major increase in catalytic power via directed evolution is achieved by lowering the probability of unproductive conformations in the transition state ensemble, Kern says. The original designed enzyme works, but dynamic fluctuations as the protein wiggles between many conformations mean that key catalytic amino acids have “a low probability of actually talking to the transition state,” Kern says. “It’s all about increasing probabilities.” In the optimized evolved enzyme, the number of possible conformations is much smaller than that of the designed enzyme, making the active one more likely. The researchers were able to design a highly efficient enzyme with only two of the 17 mutations introduced by directed evolution, highlighting predictive power for future computational design.
“These experimental findings suggest that improvements in de novo enzyme design may likely be achieved through consideration of both static and dynamic computational approaches,” William F. DeGrado and Samuel Schneider, protein designers at the University of California, San Francisco, write in an email.
“It’s clear that changes in dynamics on many scales were absolutely crucial,” Hilvert says. “The real lesson of this—I mean, we already knew that proteins are dynamic; that’s not a big surprise—is that we need to really take into account and model the conformational dynamics very early in the design process.”