Molecular dynamics simulations based on AMBER force fields(ff96 and ff03) and generalized Born models(igb1 and igb5) have been carried out in order to study folding/unfolding of the Trp-cage mini-protein TC5b.The thermodynamic properties of TC5b were found to be sensitive to the specific version of the solvation model and force field employed.When the ff96/igb5 combination was used,the predicted melting temperature from unfolding simulations was in good agreement with the experimental value of 315 K,but the folding simulation did not converge.The most stable thermodynamic profile in both folding and unfolding simulations was obtained when the ff03/igb5 combination was employed,and the predicted melting temperature was about 345 K,showing over-stabilization of the protein.Simulations using the igb1 version in combination with ff96 or ff03 were difficult to converge within the simulation time limit(50 ns).
Folding simulations are often time-consuming or highly sensitive to the initial conformation of the si-mulation even for mini protein like the Trp-cage. Here, we present a multiscale molecular dynamics method which appears to be both efficient and insensitive to the starting conformation based on the testing results from the Trp-cage protein. In this method the simulated system is simultaneously mod-eled on atoms and coarse-grained particles with incremental coarsening levels. The dynamics of coarse-grained particles are adapted to the recent trajectories of finer-grained particles instead of fixed and parameterized energy functions as used in previous coarse-grained models. In addition, the com-positions of coarse-grained particles are allowed to be updated automatically based on the coherence during its history. Starting from the fully extended conformation and other several different conforma-tions of the Trp-cage protein, our method successfully finds out the native-like conformations of the Trp-cage protein in the largest cluster of the trajectories in all of the eight performed simulations within at most 10 ns simulation time. The results show that approaches based on multiscale modeling are promising for ab initio protein structure prediction.
XIA XueFeng ZHANG Song HUANG Bo ZHOU Yun SUN ZhiRong
