Dec. 10, 2002 | FROM ALZHEIMER'S AND Lou Gehrig's disease to mad cow and Creutzfeld-Jacob disease, there is compelling evidence for the critical importance of protein folding in health and disease (see "The Protein Prophecies," June Bio·IT World, 56).
For the past 25 years or so, researchers have attempted to simulate the process of protein folding, a gargantuan task given that a typical protein assumes its preordained 3-D conformation in a matter of microseconds. Small peptides, made up of recognizable secondary structure motifs such as alpha helices and beta sheets, can assume their native state in mere nanoseconds. Simply trying to model the folding of a short peptide containing just 23 amino-acid residues — over 10 microseconds — would take a typical computer centuries.
Writing in a recent issue of Nature, Stanford University's Vijay Pande, the University of Illinois' Martin Gruebele, and colleagues have found an ingenious way past this computational bottleneck. By tapping into the unused processing power of a small fraction of the more than 300 million personal computers hooked into the Internet — distributed computing group Folding@Home — they have been able to produce more than 30,000 folding trajectories, over a 5- to 20-nanosecond timescale. In about 100 simulations, the protein successfully folded into its natural state, in excellent agreement with experimental findings.
The study represents one of the first examples — if not the first example — of distributed computing in the medical field, as validated by publication in a leading scientific journal. The kinetics of protein folding may sound like a hopelessly abstruse application of distributed computing technology, but the latest work paves the way for more ambitious and disease-relevant applications in the future.
The subject of this research is a simple model protein called BBA5, comprising just 23 amino acids. The researchers modified the sequence so they could monitor the folding experimentally by measuring fluorescence and circular dichroism spectra, which showed that the protein folded thermodynamically via a two-state mechanism.
Though experimentalists traditionally look askance at simulation studies, Pande and co-workers at Stanford University set up Folding@Home — a distributed computing scheme designed to harness the unused processing power of thousands of idle computers. Interested volunteers download the appropriate software, automatically sending back the results when the analyses are completed. (A similar program, SETI@home, is earnestly searching for alien life, though there is no word of a Nature paper as yet.)
The computational strategy is not unlike playing the lottery, Pande explains, except rather than buying tickets in series, week after week, year after year, the researchers effectively bought thousands of tickets in parallel. By performing enough simulations, the researchers expected by chance to record some successful folding events, even though the chance is slim that any given molecule would fold completely within a few nanoseconds.
Using relatively standard molecular dynamics software programs and unused CPU time on 30,000 computers, Pande's team performed some 32,500 folding simulations, accumulating a total of 700 microseconds of folding data. That's more than a million days of computer simulation time in order to model the folding of a peptide that naturally occurs in 5 microseconds.
Of the 32,500 folding simulations, about 100 runs resulted in a successfully folded molecule within 5 to 20 nanoseconds. Despite the rapid speed of these events, the simulated folding pathways are not unnatural, the authors argue. "[T]he fact that the average unfolded molecule explores conformational space for microseconds before finding the native conformation does not preclude individual molecules from folding quickly," they explain.
Although each successful simulation took a relatively different path to reach the final folded conformation, this is not unexpected — protein folding is a highly random and stochastic process. In each case, the secondary structures formed first, and then the molecule collapsed upon itself. The final product agreed well with the experimentally determined structure.
Since the work reported in the Nature article was completed, the number of computers linked by Folding@Home has doubled to more than 60,000. There is also a new corporate partnership with Google, the ubiquitous Internet search engine, in which individuals who download the Google toolbar are being invited to participate in the Folding@Home project.
The immediate scientific goal is to extend these studies to more disease-related proteins, specifically the A-beta peptides that form the characteristic, and presumably pathogenic, protein fibrils in patients with Alzheimer's disease. Understanding the folding — and misfolding — pathways of these molecules could provide crucial clues against this insidious disease.
While the Nature paper has only four authors, they make it clear who deserves the credit. The first line of the acknowledgements reads: "We thank the Folding@Home volunteers whose processor power made this work possible."