495During the past several years, there have been a number ofadvances in the computational and theoretical modeling of lipidbilayer structural and dynamical properties. Moleculardynamics (MD) simulations have increased in length and timescales by about an order of magnitude. MD simulationscontinue to be applied to more complex systems, includingmixed bilayers and bilayer self-assembly. A critical problem isbridging the gap between the still very small MD simulationsand the time and length scales of experimental observations.Several new and promising techniques, which use atomic-levelcorrelation and response functions from simulations as input tocoarse-grained modeling, are being pursued. AddressesDepartment of Biological, Chemical and Physical Sciences,Illinois Institute of Technology, Chicago, IL 60616, USA; e-mail: [email protected] Opinion in Structural Biology 2002, 12:495–5020959-440X/02/$ — see front matter© 2002 Elsevier Science Ltd. All rights reserved.AbbreviationsCBMC configurational bias MCCS cholesterol sulfateDMPC dimyristoyl phosphatidylcholineDOPC dioleoyl phosphatidylcholineDOPE dioleoyl phosphatidylethanolamineDPPC dipalmitoyl phosphatidylcholineDPPG dipalmitoyl phosphatidylglycerolDPPS dipalmitoyl phosphatidylserineGMO glycerolmonooleinHNC hypernetted chain MC Monte CarloMD molecular dynamicsNOESY nuclear Overhauser enhancement spectroscopyPOPC palmitoyl-oleoyl phosphatidylcholineSDPC 1-stearoyl-2-docosahexaenoyl-sn-glycero-3-phosphocholineIntroductionBiological membranes are enormously complex in terms ofboth structure and their dynamical properties. In order tobegin to understand membrane properties, model systemsconsisting of simple lipid bilayers with well-known com-position are generally studied. However, even thesesystems pose formidable modeling challenges due to theflexible nature of lipid molecules and the occurrence ofcomplex hydrophilic and hydrophobic interactions. Forthis reason, computer simulation has emerged as a criticaltool for modeling the lipid component of membranes. Asearlier reviews of lipid bilayer modeling have pointed out[1–6], the increasing availability of fast desktop workstationsor linux clusters, and of fast modeling software has led torapid progress in atomic-level simulations. Figure 1 showsa bilayer of 1600 DPPC molecules, which is indicative of the size scales accessible to current simulations. In addition, simulations are now beginning to provide data ofsufficiently high quality that workers can begin to usethese data as input for more coarse-grained macroscopicmodeling work. Concurrently, new experimental data areavailable (e.g. [7]) and communication between membranestructure experimentalists and simulators has increased.The purpose of this review is to summarize the mostrecent advances in lipid bilayer modeling work and todescribe promising new uses of atomistic simulations asinput to efforts to model lipid bilayers over macroscopictime and length scales. Methodological advancesFigure 2 shows atomic structures for three representativelipid molecules used in simulations. Atomic-level simula-tions require as input expressions for the potential energiesbetween all atoms in the system, including those betweenbonded pairs of atoms and between nonbonded pairs ofatoms. Harmonic expressions are generally used to modelchemical bonds between atoms. For atom pairs that are notchemically bonded, potential energy expressions includecoulombic plus ‘6-12’ interactions. The 6-12 expressionconsists of a repulsive part, which falls off as 1/r12, and anattractive part, which falls off as 1/r6, and is designed tomodel interactions due to polarization effects betweenatomic electron clouds. Torsional potentials model theinteractions between next-nearest neighbor atoms on thesame molecule. The set of functions, and the parameterscharacterizing the strengths of the various interactions, iscommonly referred to as the ‘force field’. Force fields aregenerally independently developed and tested againstexperimental data before being applied to lipid bilayers. The improvement of simulation force fields is a continuousprocess. Feller and MacKerrell have reported improve-ments to the CHARMM all-atom force field for lipidsimulations for torsion and 6-12 parameters [8], and forpolyunsaturated lipids [9••]. The latter parameterizationeffort extends simulation force fields to the important classof multiply unsaturated lipid bilayers. The force fieldreveals the extremely high flexibility of these systems. Aninteresting new methodological advance is the develop-ment of periodic boundary conditions, which allow lipidsto switch leaflets [10]. This is an important advance for thesimulation of unsymmetrical bilayer systems, as it willallow the redistribution of lipids to ease stresses inducedby asymmetries. Ongoing issues in simulation methodology include thetreatment of electrostatic interactions and the use of constant surface tension ensembles. For the electrostaticsissue, recent simulations of the lipid gel phase by Venableet al. [11] show clearly the necessity of including Ewaldsummation corrections in the simulation of this phase,which consists of tightly packed, ordered hydrocarbonchains with greatly restricted lateral and conformationalModeling the lipid component of membranesH Larry Scottmolecular mobility. For the simulation of lipid fluid phases, in which lipid molecules are more disordered andliquid-like than in the gel phase, the situation, in the opinion of this writer, is less clear. The use of Ewald summations in zwitterionic systems in disordered phasescan potentially induce unwanted periodicities, leading toerrors no less significant than those which result when theinteraction is simply ‘switched off’ for interatomic distances beyond some cutoff (spherical cutoffs). Forsmaller simulation boxes, the effect will be more severe.For the surface tension issue, discussions are ongoing concerning the nature of the lipid/water interface and theappropriateness of a constant surface tension ensemble[12,13]. The difficulty associated with dissecting the surface stresses in a bilayer is addressed in simulations byLindahl and Edholm [14]. They use local virial