We have built clusters of PC using up to 320 processors in parallel. The first cluster was built in 1996 from 64 computers. The parallel architectures are named RAMSES. In this article, we share our experience and discuss our choices. The architecture is composed of Pentium processors connected by a 100 Mbits/s Ethernet network. Because RAMSES are built of independent computers, the system is easily maintained, and can be modified by adding or removing machines. Two programs, which are dedicated to molecular modeling and were previously developed on single machines (OSIRIS and IMPALA), have been adapted. IMPALA models peptide and protein interaction with membranes, OSIRIS simulates an ab initio folding of globular proteins. After parallelization, each program consists of two parts: one is the human/RAMSES interface, and the second is the simulation per se. The parallel architecture is transparent for the users. Tests of the IMPALA and OSIRIS efficiencies are shown and discussed. © 2000 John Wiley & Sons, Inc. J Comput Chem 22: 172–177, 2001

A continuous membrane model (IMPALA) was previously developed to predict how hydrophobic spans of proteins insert in membranes (Mol. Mod. 2 (1996) 27). Using that membrane model, we looked for the interactions between several hydrophobic spans. We used the glycophorin A dimer as an archetype of polytopic protein to validate the approach. We find that the native complex do not dislocate when it is submitted to a 10(5) steps optimisation whereas separated spans converge back to a native-like complex in the same conditions. We also observe that IMPALA restraints are not strictly mandatory but do increase the efficiency of the procedure.

In order to better understand the driving forces that determine the alignment of amphipathic helical polypeptides with respect to the surface of phospholipid bilayers, lysine-containing peptide sequences were designed, prepared by solid-phase chemical synthesis, and reconstituted into membranes. CD spectroscopy indicates that all peptides exhibit a high degree of helicity in the presence of SDS micelles or POPC small unilamellar vesicles. Proton-decoupled 31P-NMR solid-state NMR spectroscopy demonstrates that in the presence of peptides liquid crystalline phosphatidylcholine membranes orient well along glass surfaces. The orientational distribution and dynamics of peptides labeled with 15N at selected sites were investigated by proton-decoupled 15N solid-state NMR spectroscopy. Polypeptides with a single lysine residue adopt a transmembrane orientation, thereby locating this polar amino acid within the core region of the bilayer. In contrast, peptides with ≥3 lysines reside along the surface of the membrane. With 2 lysines in the center of an otherwise hydrophobic amino acid sequence the peptides assume a broad orientational distribution. The energy of lysine discharge, hydrophobic, polar, and all other interactions are estimated to quantitatively describe the polypeptide topologies observed. Furthermore, a molecular modeling algorithm based on the hydrophobicities of atoms in a continuous hydrophilic-hydrophobic-hydrophilic potential describes the experimentally observed peptide topologies well.

Nisin is a 34-residue lantibiotic widely used as food preservative. Its mode of action on the bacterial cytoplasmic membrane is unclear. It should form ion channels but a molecular description of the interaction between nisin and phospholipids is lacking. The interactions between nisin and a membrane and the influence of phospholipids are here analysed by molecular modelling. The NMR structures of nisin in a micellar environment were previously determined (Van den Hooven et al., Eur. J. Biochem. 235 (1996) 382-393) Those structures were used to start with. They were refined by running a Monte Carlo procedure at a model lipid/water interface. It was shown that nisin is adsorbing onto the interface, with its N-terminal moiety more deeply inserted in lipids than the C-end, indicating distinct hydrophobic properties of the N- and C-domains. Therefore, we suggest that the N-terminal part is implied in the insertion of nisin in lipids, while the C-terminal moiety could be involved in the initial interaction with the membrane surface. Modelling the interaction of nisin with different neutral or anionic phospholipids shows that it disturbs the lipid organisation. The disturbance is maximal with phosphatidylglycerol. In this system, nisin curves the surface of phosphatidylglycerol layer round suggesting it could induce micelle formation. This could be a preliminary step to pore formation. It suggests that phosphatidylglycerol could have a direct action on nisin insertion and on ion channel formation. Appearance of a curvature also agrees with the 'wedge model' proposed in the literature for the nisin pore formation.

The solvent of biological media is water and water and interactions between water and solutes are major to explain membrane and protein structures. Therefore, mimicking solvation effects is a challenge to compute native structures of biological molecules. Lipophily is an experimental approach of solvation since it measures the partition between two solvents, the reference being water In the last decade, different attempts were made to extract parameters from lipophily that will enable to describe solvation of molecules. We discuss here some analysis made with the atomic transfer energy and the atomic surface parameter in the study of protein folding and protein insertion in membranes.

The lipid bilayer is crucial for the folding of integral membrane proteins. This article presents an empirical method to account for water-lipid interfaces in the insertion of molecules interacting with bilayers. The interactions between the molecule and the bilayer are described by restraint functions designed to mimic the membrane effect. These functions are calculated for each atom and are proportional to the accessible surface of the latter. The membrane is described as a continuous medium whose properties are varying along the axis perpendicular to the bilayer plane. The insertion is analyzed by a Monte Carlo procedure applied to the restraint functions. The method was successfully applied to small alpha peptides of known configurations. It provides insights of the behaviors of the peptide dynamics that cannot be obtained with statistical approaches (e.g., hydropathy analysis).

The cystic fibrosis transmembrane conductance regulator is a cAMP-regulated chloride channel. We used molecular modelling to predict 3-D models for the CFTR membrane domain. Hydropathy and residue conservation in all CFTRs as well as in other proteins suggested that the membrane domain is a 12-helix bundle. If the domain is enclosing a channel for chloride, it could be made of five helices. We propose two structural models in which both lumenal and cytoplasmic entrances to the chloride pore have a ring of positively charged residues. The inner surface of the channel is covered with neutral polar plus one or two charged residues. Helices that are not directly involved in the chloride channel could organise to form a second channel; a dimeric symmetrical structure is proposed. Analysis raised interest for helix 5: this hydrophobic fragment is conserved in all CFTRs and aligns with segments present in several different ion channels and transporters. The existence of an FFXXFFXXF motif is proposed. Helix 5 could be an important domain of CFTRs. The models agree with available data from pathological mutations but does not account for the membrane insertion of a hydrophilic fragment of NBD1.

There are two models of topology for the membrane domains of the erythrocyte/brain facilitative glucose transporter, GLUT1. The first is composed of 12 membrane-spanning alpha-helices, the second of 16 membrane-spanning beta-strands. We have used J hnig s and Eisenberg s methods to identify possible transmembrane segments (10 spanning alpha-helices and 4 beta-strands). The topology proposed is more consistent with available experimental data from FTIR, CD and mapping experiment than the previous models. We suggest that GLUT1 might form two channels, one of which is responsible for glucose transport. This agrees with the theoretical and experimental arguments. Finally, an analysis of the mutation periodicity and of the mean hydrophobicity for the GLUT family is provided in order to evaluate the packing of the protein in the membrane.

There are two models of topology for the membrane domains of the erythrocyte/brain facilitative glucose transporter, GLUT1. The first is composed of 12 membrane–spanning α–helices, the second of 16 membrane-spanning β-strands. We have used Jähnig′s and Eisenberg′s methods to identify possible transmembrane segments (10 spanning α-helices and 4 β-strands). The topology proposed is more consistent with available experimental data from FTIR, CD and mapping experiment than the previous models . We suggest that GLUT1 might form two channels, one of which is responsible for glucose transport. This agrees with the theoretical and experimental arguments. Finally, an analysis of the mutation periodicity and of the mean hydrophobicity for the GLUT family is provided in order to evaluate the packing of the protein in the membrane.