Now, measurements from the University of Illinois at Urbana-Champaign and molecular dynamics simulations from Stanford University have at last been compared and found to be in very good agreement. A paper describing the work has been accepted for publication in the journal Nature, and was posted on its Web site www.nature.com/nature.
"By crossing the microsecond barrier, we can directly compare simulated and experimental protein folding dynamics, such as folding rates and equilibrium constants," said Martin Gruebele, an Illinois professor of chemistry, physics and biophysics.
To allow experiment and theory to meet on a microsecond time scale, the researchers designed a small protein based on the work of Barbara Imperiali and her colleagues, now at the Massachusetts Institute of Technology. Consisting of only 23 amino acids, the protein contains all three basic elements of secondary structure -- helices, beta sheets and loops -- but can fold simply and rapidly.
At Illinois, Gruebele and graduate student Houbi Nguyen measured folding times using a fast temperature jump experimental procedure. To initiate the folding and unfolding dynamics, the solution was heated rapidly by a single pulse from an infrared laser. As the proteins began twisting into their characteristic shapes, a series of pulses from an ultraviolet laser caused some of the amino acids to fluoresce, revealing to the researchers a time-sequence of folding and unfolding events from which the folding rate constant was obtained.
At Stanford, physical chemist Vijay Pande and graduate student Christopher Snow accumulated more than 700 microseconds of molecular dynamics simulations by dividing the work among more than 30,000 volunteer computers distributed around the world.
Inspired by previous distributed computing initiatives, such as SETI@Home -- an immensely popular program that searches radio telescope data for evidence of extraterrestrial transmissions -- Pande developed a similar screen saver, which he called Folding@Home. The program broke the number crunching into many thousands of tiny pieces, each covering only 5-20 nanoseconds of folding time, and ran them using spare time on the volunteer computers.
"The computational predictions were in extremely good agreement with our experimentally determined folding times and equilibrium constants," Gruebele said. "For example, our group came up with an average folding time of 7.5 microseconds, while PandeÕs group came up with 8.0 microseconds."
Moreover, while distributed computing initiatives like SETI@Home have offered the promise of novel scientific advances, Folding@HomeÕs success is both an important advance in understanding protein folding and it is the first time a distributed computing project has yielded a significant advance.
The simulations also demonstrated the heterogeneous nature of the folding event and the funnel-shaped appearance of the protein's energy landscape.
"The protein can fall downhill to its native state through many different scenarios," Pande said. "In some of the simulations, the beta sheet formed first, in others the alpha helix formed first, and in still others the loop formed first. Because the protein can follow more than one pathway, a variety of folding times will result."
By comparing absolute quantities through experiment and simulation, the researchers can determine energy barriers and relative energies more accurately than before. The next step, they say, is to perform a similar comparison using a larger, more complex protein.
The National Institutes of Health, American Chemical Society and National Science Foundation funded the work.