Scientists Break Protein Folding Time Barrier
The Millisecond is Attained!
Scientists have now simulated protein folding at a timescale that begins to be relevant to biology: the millisecond. Indeed, the simulation busted through the millisecond time barrier to tackle the slowest folding protein yet studied—a 1.5 millisecond fold—using a combination of computational tools that provide both the requisite computational power and the necessary analytical methods for making sense of a slow, complicated folding event.
“It’s kind of like a coming out party for a combination of technologies that have really started to mature,” says Vijay Pande, PhD, associate professor of chemistry at Stanford University and co-author of the paper. Pande expects his team’s technologies will be useful for simulating proteins important in misfolding diseases such as Alzheimer’s and Huntington’s disease. The work was published in the Journal of the American Chemical Society in January.
Protein-folding researchers have until now focused on a unique group of small, fast-folding proteins that fold in hundreds of nanoseconds or microseconds. This is great for simulating, but it is not characteristic of most protein-folding events. Pande’s group chose to simulate a 39-amino acid chain called NTL9, which, like most proteins, dilly-dallies en route to its final structure. One side of the protein may partially fold, then unfold as another part misfolds. The process takes milliseconds or more.
The computational power for the simulation came from Folding@Home, a distributed computing project that heaps together bits of donated computer time from individual systems located around the world. To fold NTL9, they relied particularly on the speedy graphical processing units (GPUs) within those computers, which sped up the simulations and made long folding trajectories possible.
To piece together the information from the different computers, Pande and his coworkers also devised a Markov State Models (MSMs) method. The approach merges myriad variations from thousands of successive protein-folding simulations and identifies a set of relatively stable conformations along the protein’s many folding pathways. By choosing how many states to identify, whether fifteen or 100,000, researchers can dial in the degree of complexity they seek. It’s like choosing the number of pixels in a photograph, Pande says. A small number of states gives a broad, coarse picture of the conformations and folding pathways of greatest frequency, while a larger number provides a more complex picture that can show specific protein movements in greater detail.
The MSM approach allowed Pande’s group to see a real richness of range in the way NTL9 folds. NTL9 follows not just one or two pathways but many different paths to get to the final folded state. Pande expects to see similar heterogeneity in the way other proteins fold, and his group has created a tool called MSMBuilder to enable other groups to conduct a similar analysis of their own simulations.
Jed Pitera, PhD, a research staff member at IBM, says Pande’s group found a way to build a statistically and physically accurate model of protein folding. “It shows off the state-of-the-art in studies of folding kinetics and reflects a maturation of the view of how protein folding happens,” he says.