4. Systematic comparison of empirical force fields for molecular dynamic simulation
4.2 Introduction
In the last few decades molecular dynamics (MD) simulations have emerged as a powerful tool for the characterisation of biomolecular structure and dynamics at the atomic level. The technique has helped us understand complex molecular processes associated with protein conformational changes, ranging from studies of enzyme-reaction mechanisms and ligand binding to problems of protein re-folding and denaturation (see references within Ponder and Case [286]). Fundamental to such simulations are the forces that govern the atomic motions, derived from a pairwise atom-atom interaction function referred to as an empirical potential energy function or a force field. Details on force field theory and development are presented in Chapter 2. The choice of mathematical function describing a force field is important, since it will ultimately determine the quality of the results. In order to understand possible systematic effects of force fields on the simulated dynamics of insulin a detailed comparison of five commonly used force fields was performed.
The biomolecular force fields selected for our study include the recent versions of the all-atom CHARMM27 [150], AMBER03 [165], OPLS [163] and the united-atom GROMOS 43A1 [174] and GROMOS 53A6 [153] force fields. The parameters for all five force fields were extensively optimised over a number of years with particular emphasis on the treatment of proteins. Detailed discussion on the parameterisation and functions of each force field is presented in Chapter 2. AMBER03 was selected for this study, because it showed very good agreement with experimental data in an extensive force field investigation between the readily available AMBER force fields [287], not considering some custom modified AMBER versions.
An important consideration in the use of a force field for biomolecular simulations is the accurate treatment of the condensed aqueous environment. When selecting a water model to use for a particular study, its compatibility with the biomolecular force field used is of crucial importance. The reason for this is that most force fields have been developed in conjunction with a specific water model. For example, AMBER, CHARMM and OPLS have been developed with the TIP3P water model, OPLS also with TIP4P while GROMOS with SPC and F3C models.
CHAPTER 4. Systematic comparison of empirical force fields
for molecular dynamics simulations of insulin
76
Recent simulations of three proteins performed using the AMBER94, CHARMM and OPLS force fields, concluded that these force fields behave comparably or at least that it was not possible to distinguish between the force fields in the 2 ns time scale investigated [288]. In a more recent article by Villa et al. the sensitivity of molecular dynamics simulation to different force fields was examined [289]. The three parameter sets (43A1, 53A5 and 53A6) of the GROMOS force field were used to simulate 36 structures of 31 proteins. In their investigation, no major differences were detected in a wide range of structural properties such as the root-mean-square deviation from the experimental structure, radii of gyration, solvent accessible surface, secondary structure, or hydrogen bond propensities on a 5 to 10 ns time scale, despite the differences in the force field parameters. On the contrary, Mu et al. showed that the recent version of the AMBER, CHARMM, GROMOS, and OPLS force fields used in their study of trialanine differ considerably in the description of the dynamical properties of the system. It was found that the lifetime of some conformational states differ by more than an order of magnitude, depending on which force field model was used [169]. Several empirical force fields and quantum mechanics/molecular mechanics (QM/MM) force fields were used by Hu et al. to investigate the sampled conformations of unfolded polypeptide chains in aqueous solution [170]. This group also found variation in their results which they related to the different force fields that were applied, rather than to a direct comparison with experimental data. The findings by Sorin and Pande [290] on helix-coil systems in conjunction with work by Zeman et al. on alanine dimers and trimers [291] suggested that a dependence of force field accuracy on peptide length exists even for simple models such as polyalanine-based systems. In the article by Yoda et al. [292] discrepancies in simulation results due to the choice of force field have been well documented for a significant number of systems.
In order to understand any possible systematic bias of the common force fields the protein insulin for the present investigation was selected. Insulin is a widely studied protein and there is an extensive experimental and computational literature available for comparison (summarised in Chapter 3). Molecular dynamics enables sampling of conformational states of a protein under controlled conditions and therefore delivers detailed understanding of the protein’s dynamics. Previous MD simulations using the GROMOS 37c force field, show conformational flexibility of insulin chain B, reporting a high degree of movement in aqueous solution of both the monomer and dimer [256]. Simulations of monomeric insulin in the T-
CHAPTER 4. Systematic comparison of empirical force fields
for molecular dynamics simulations of insulin
77
state conformation were performed by Zoete et al. [260] for 5-10 ns using the CHARMM22 force field where they identified the flexibility of the N- and C-terminal regions of chain B. This inherent flexibility was also observed in a previous computational study of possible thermal and chemical effects on the dynamics of chain B [225] using the NAMD simulation package and the CHARMM27 parameter set. The most recent of these studies performed by Legge and co-workers yielded information clarifying uncertainties about the structure and dynamics of insulin with respect to its biological behaviour by performing multiple MD simulations with the CHARMM27 force field in explicit solvent [230], as an alternative way to improve the conformational sampling.
In this chapter we attempt to identify and compare possible systematic effects of multiple force fields on the structure and dynamics of chain B of insulin in a course of identical molecular dynamics simulations. We also aim to investigate which force field gives the best representation of the experimentally observed conformational features and behaviour of chain B within our selected timeframe. The assessment of the force fields has been performed by comparison of results with experimental data which illustrates the 3- dimensional morphology of the protein, such as X-ray crystallographic structures [241, 242, 245] (see Figure 3.2), Nuclear Overhauser Enhancement (NOE) distance restraints from the solution NMR structure of isolated chain B of insulin [232], and a previous molecular dynamics study of chain B which reproduced known experimental structural features [230]. To the best of our knowledge this is the only information available in the literature for direct comparison of conformational features observed in the simulations with different force fields. Dynamic and structural properties such as the secondary structure evolution, solvent accessible surface area (SASA), radius of gyration, interproton distance violations, have been obtained from each force field simulation for comparison and are presented in the Results and Discussions section of this chapter.