Vijay Pande, Stanford University

Do proteins designed by man fold like their natural counterparts? We compared the folding of two designed zinc fingers to a human zinc finger and found that the folding pathway is similar in all three proteins: all fold via a partially structured, -helical intermediate state.

Research Objectives
We propose a new algorithm for parallel molecular dynamics (MD), which bootstraps on top of current parallel methodologies and uses relatively little communication bandwidth. This allows scaling of current codes to 10 to 1000 times more processors than currently possible. With this new method, we will be able to simulate considerably longer timescales than currently possible — instead of tens of nanoseconds, tens of microseconds. This increase is particularly important for protein folding, since the fastest proteins fold in the microsecond regime. Thus we will be able for the first time to directly fold small proteins using MD simulations in all-atom detail.

Computational Approach
Our computational approach takes advantage of the inherent kinetics in our system. Like other free energy barrier crossing problems, a protein wanders around one free energy minimum, waiting for some rare thermal fluctuation to push it over the barrier. It has been demonstrated that the time to cross the barrier is much less than the overall folding time — most of the time is spent waiting for this fluctuation. Our method parallelizes this "waiting stage": starting from some initial coordinates, we run M parallel simulations (each with different initial velocities). If a system has multiple barriers, this method can be extended to handle them as well.

We bootstrap this method on top of current parallel MD (using NAMD). Each simulation is a 16-processor NAMD simulation. Thus, we get double benefits of parallelization: from traditional MD and from our "ensemble dynamics" method.

Accomplishments
We have simulated the unfolding pathways of two small proteins (zinc fingers) which fold to the same structure but have very little sequence similarity. We have shown that in spite of little similarity in sequence, these proteins have similar unfolding pathways. This lends strong evidence that native state topology is a primary determinant of the nature of the folding pathway.

Significance
Computationally, this method should be broadly applicable to any free energy barrier crossing problem, and should be useful for anyone using MD simulations. Scientifically, the folding of a protein in all-atom detail is a holy grail of modern computational biology. Understanding how proteins fold has broad implications in many areas, including protein design, protein misfolding (believed to be related to several diseases), and the design of self-assembling protein-like nanostructures.

Publications
V. S. Pande and D. S. Rokhsar, "Molecular dynamics simulation of unfolding and refolding of a b-hairpin fragment of protein G," Proc. Natl. Acad. Sci. U.S.A. 96, 9062 (1999).

V. S. Pande and D. S. Rokhsar, "Folding pathway of a lattice model for proteins," Proc. Natl. Acad. Sci. U.S.A. 96, 1273 (1999).

V. S. Pande, A. Yu. Grosberg, and T. Tanaka, "Heteropolymer freezing and design: Towards physical models of protein folding," Rev. Mod. Phys. 72, 259 (2000).

http://www.stanford.edu/group/pandegroup