MOLECULAR-DYNAMICS SIMULATIONS OF A PROTEIN IN THE CANONICAL ENSEMBLE

We report the results of constant energy (microcanonical ensemble) and constant temperature (canonical ensemble) molecular dynamics simulations of bovine pancreatic trypsin inhibitor. The constant temperature simulations were carried out by using either Langevin dynamics or one of three ''extended system'' (ES) methods: a single Nose-Hoover thermostat, a single Nose-Hoover chain, and ''massive' Nose-Hoover chains (one chain coupled to each degree of freedom). We evaluate the abilities of the different methods to provide temperature control and examine their effects on kinetic energy equipartitioning rates, average structures and fluctuations, and dynamical properties. Provided the system is well-equilibrated, all of the methods considered in this study can be used to perform essentially constant temperature simulations of proteins that generate canonical velocity distributions. The methods differ in their utility as equilibration techniques. During equilibration, in microcanonical simulations, the temperature can drift, while in Nose-Hoover simulations, relatively large, undesirable oscillations in the temperature can build up. The simulations based on the Langevin or Nose-Hoover chain dynamics do not suffer from these problems. The equipartitioning rates were comparable in the microcanonical, Nose-Hoover, and Nose-Hoover chain simulations, but they were significantly faster in the ''massive'' chains and Langevin simulations, particularly at larger values of the coupling parameter(thermostat frequency and friction, respectively). The average structures from all of the simulations were similar, differing about 2 angstrom from the experimentally determined structure. The root-mean-squared fluctuations were roughly the same in the microcanonical, low-friction Langevin, Nose-Hoover, Nose-Hoover chains, and low-frequency ''massive'' simulations. The fluctuations were moderately increased at larger thermostat frequencies in the ''massive'' chain simulations but were drastically reduced with increasing friction constants in the Langevin simulations. Our analysis of various time correlation functions shows that the protein dynamics was not seriously affected by Nose-Hoover thermostats but, not surprisingly, the dynamics in the high-friction Langevin simulations were much different from the microcanonical and ES simulations. On the basis of the results of this investigation, we recommended the ''massive'' Nose-Hoover chain dynamics as an efficient technique for equilibration and any of the Nose-Hoover chain based ES methods for production dynamics where good temperature control is desired in protein simulations.