How consistent are molecular dynamics simulations? Comparing structure and dynamics in reduced and oxidized Escherichia coli thioredoxin.

In this study we have examined several parameters that can be used for checking the consistency and accuracy of protein structures and molecular dynamics simulations. This is done by comparing: (1) three X-ray structures of oxidized Escherichia coli thioredoxin (Trx-S2); (2) 14 NMR structures of reduced E. coli thioredoxin (Trx-(SH)2); and (3) 30 different simulations, 15 of Trx-S2 and 15 of Trx-(SH)2. The energy, the agreement with NOE data, the root-mean-square deviation between structures, and the surface characteristics of all these structures are analyzed. The 30 simulations, four water simulations, 20 standard vacuum simulations and six alternative vacuum simulations, are examined with respect to mobility, temperature factors and aromatic side-chain mobility. It is shown that although vacuum simulations may reproduce some parameters, all the features of a water simulation cannot be reproduced in any of these simulations. Several of the parameters described above are shown to be good for discriminating between an accurate and an inaccurate simulation. It is also shown that 100 ps is too short a time to obtain statistically certain temperature factors and correlation functions of aromatic side-chain motions. The results also suggest that performing ten 100 ps simulations spans the conformation space better than one 1 ns simulation.