Atomistic Research Articles

Coupling of an atomistic model and bond-based peridynamic model using an extended Arlequin framework

A general nonlocal coupling technique between an atomistic (AM) model and the bond-based peridynamic (PD) model is proposed, based on the Arlequin framework. This technique applies the complementary weight function and constraint conditions to transmit energies through the overlapping region between the AM and PD regions. We extend the original Arlequin framework to discrete cases by redefining constraint conditions by the peridynamic differential operator, which enables the interpolation and corresponding derivative of scattered data. Besides, the preconditioning of calibration for the PD effective micromodulus is implemented to guarantee the equilibrium of energy. One-dimensional benchmark tests investigate the coupling effects influenced by several key factors, including the coupling length, weight function, grid size and horizon in the PD model, and constraint conditions. Two- and three-dimensional numerical examples are provided to verify the applicability and effectiveness of this coupling model. Results illustrate this AM–PD coupling model takes the mutual advantages of the computational efficiency of PD model and the accuracy of AM model, which provides a flexible extension of the Arlequin framework to couple particle methods.

Membrane-binding mechanism of the EEA1 FYVE domain revealed by multi-scale molecular dynamics simulations.

Early Endosomal Antigen 1 (EEA1) is a key protein in endosomal trafficking and is implicated in both autoimmune and neurological diseases. The C-terminal FYVE domain of EEA1 binds endosomal membranes, which contain phosphatidylinositol-3-phosphate (PI(3)P). Although it is known that FYVE binds PI(3)P specifically, it has not previously been described of how FYVE attaches and binds to endosomal membranes. In this study, we employed both coarse-grained (CG) and atomistic (AT) molecular dynamics (MD) simulations to determine how FYVE binds to PI(3)P-containing membranes. CG-MD showed that the dominant membrane binding mode resembles the crystal structure of EEA1 FYVE domain in complex with inositol-1,3-diphospate (PDB ID 1JOC). FYVE, which is a homodimer, binds the membrane via a hinge mechanism, where the C-terminus of one monomer first attaches to the membrane, followed by the C-terminus of the other monomer. The estimated total binding energy is ~70 kJ/mol, of which 50-60 kJ/mol stems from specific PI(3)P-interactions. By AT-MD, we could partition the binding mode into two types: (i) adhesion by electrostatic FYVE-PI(3)P interaction, and (ii) insertion of amphipathic loops. The AT simulations also demonstrated flexibility within the FYVE homodimer between the C-terminal heads and coiled-coil stem. This leads to a dynamic model whereby the 200 nm long coiled coil attached to the FYVE domain dimer can amplify local hinge-bending motions such that the Rab5-binding domain at the other end of the coiled coil can explore an area of 0.1 μm2 in the search for a second endosome with which to interact.

Density-functional-theory approach to the Hamiltonian adaptive resolution simulation method

In the Hamiltonian adaptive resolution simulation method (H–AdResS) it is possible to simulate coexisting atomistic (AT) and ideal gas representations of a physical system that belong to different subdomains within the simulation box. The Hamiltonian includes a field that bridges both models by smoothly switching on (off) the intermolecular potential as particles enter (leave) the AT region. In practice, external one-body forces are calculated and applied to enforce a reference density throughout the simulation box, and the resulting external potential adds up to the Hamiltonian. This procedure suggests an apparent dependence of the final Hamiltonian on the system’s thermodynamic state that challenges the method’s statistical mechanics consistency. In this paper, we explicitly include an external potential that depends on the switching function. Hence, we build a grand canonical potential for this inhomogeneous system to find the equivalence between H–AdResS and density functional theory (DFT). We thus verify that the external potential inducing a constant density profile is equal to the system’s excess chemical potential. Given DFT’s one-to-one correspondence between external potential and equilibrium density, we find that a Hamiltonian description of the system is compatible with the numerical implementation based on enforcing the reference density across the simulation box. In the second part of the manuscript, we focus on assessing our approach’s convergence and computing efficiency concerning various model parameters, including sample size and solute concentrations. To this aim, we compute the excess chemical potential of water, aqueous urea solutions and Lennard–Jones (LJ) mixtures. The results’ convergence and accuracy are convincing in all cases, thus emphasising the method’s robustness and capabilities.

Back-mapping augmented adaptive resolution simulation.

Concurrent multiscale techniques such as Adaptive Resolution Scheme (AdResS) can offer ample computational advantages over conventional atomistic (AT) molecular dynamics simulations. However, they typically rely on aphysical hybrid regions to maintain numerical stability when high-resolution degrees of freedom (DOFs) are randomly re-inserted at the resolution interface. We propose an Energy Minimized AT (DOF) Insertion (EMATI) method that uses an informed rather than random AT DOF insertion to tackle the root cause of the issue, i.e., overlapping AT potentials. EMATI enables us to directly couple AT and coarse-grained resolutions without any modifications of the interaction potentials. We exemplify AdResS-EMATI in a system of liquid butane and show that it yields improved structural and thermodynamic properties at the interface compared to competing AdResS approaches. Furthermore, our approach extends the applicability of the AdResS without a hybrid region to systems for which force capping is inadequate.

Extending the Martini Coarse-Grained Force Field to N-Glycans.

Glycans play a vital role in a large number of cellular processes. Their complex and flexible nature hampers structure-function studies using experimental techniques. Molecular dynamics (MD) simulations can help in understanding dynamic aspects of glycans if the force field parameters used can reproduce key experimentally observed properties. Here, we present optimized coarse-grained (CG) Martini force field parameters for N-glycans, calibrated against experimentally derived binding affinities for lectins. The CG bonded parameters were obtained from atomistic (ATM) simulations for different glycan topologies including high mannose and complex glycans with various branching patterns. In the CG model, additional elastic networks are shown to improve maintenance of the overall conformational distribution. Solvation free energies and octanol-water partition coefficients were also calculated for various N-glycan disaccharide combinations. When using standard Martini nonbonded parameters, we observed that glycans spontaneously aggregated in the solution and required down-scaling of their interactions for reproduction of ATM model radial distribution functions. We also optimized the nonbonded interactions for glycans interacting with seven lectin candidates and show that a relatively modest scaling down of the glycan-protein interactions can reproduce free energies obtained from experimental studies. These parameters should be of use in studying the role of glycans in various glycoproteins and carbohydrate binding proteins as well as their complexes, while benefiting from the efficiency of CG sampling.

Ligand-protein interactions in lysozyme investigated through a dual-resolution model.

A fully atomistic (AT) modeling of biological macromolecules at relevant length‐ and time‐scales is often cumbersome or not even desirable, both in terms of computational effort required and a posteriori analysis. This difficulty can be overcome with the use of multiresolution models, in which different regions of the same system are concurrently described at different levels of detail. In enzymes, computationally expensive AT detail is crucial in the modeling of the active site in order to capture, for example, the chemically subtle process of ligand binding. In contrast, important yet more collective properties of the remainder of the protein can be reproduced with a coarser description. In the present work, we demonstrate the effectiveness of this approach through the calculation of the binding free energy of hen egg white lysozyme with the inhibitor di‐N‐acetylchitotriose. Particular attention is payed to the impact of the mapping, that is, the selection of AT and coarse‐grained residues, on the binding free energy. It is shown that, in spite of small variations of the binding free energy with respect to the active site resolution, the separate contributions coming from different energetic terms (such as electrostatic and van der Waals interactions) manifest a stronger dependence on the mapping, thus pointing to the existence of an optimal level of intermediate resolution.

Multiscale Simulations of Membrane Recognition by Lipid Kinases

Peripheral membrane proteins associate with lipids in the plasma membrane in many important cell signalling processes. Type I phosphatidylinositol phosphate kinase A (PIP5K1A) is one such example, which phosphorylates the head group of the lipid phosphatidylinositol 4-phosphate (PI4P) to generate phosphatidylinositol (4,5)-bisphosphate (PIP2). The dynamic interactions of peripheral proteins with their target membranes is difficult to capture on experimentally. A recent crystal structure at 3.3 Å resolution enables the further investigation of the structure and dynamics of PIP5K1A and its interactions with lipids in the plasma membrane. Here we use coarse-grained (CG) and atomistic (AT) MD simulations of the PIP5K1A protein kinase to investigate the nature of monomer and dimer binding to the plasma membrane and their effects on lipid clustering. The AT simulations suggest that whilst dimerisation of the kinase results in an increase in the area of the binding surface, the monomers may not bind concurrently. This in turn suggests that an allosteric mechanism for stabilisation of the productive binding conformation is more likely. We have also used simulations to investigate the relationship between conformational dynamics of the activation loop and interactions with lipids in the cell membrane. Furthermore, large-scale CG simulations (1 million particles) reveal how lipid composition and membrane curvature relate to the interactions of an ensemble of 25 kinases with a model membrane. PIP kinases are implicated in many cancers as a result of their influence on cell processes. Studies in mice and human tissue samples suggest that inhibition of PIP5K1A is a potential target for treatment. Simulations of the interactions of PIP5K1A with realistic models of cell membranes will aid design of drugs targeted at these and other related lipid kinases.

Improved accuracy of hybrid atomistic/coarse-grained simulations using reparametrised interactions.

Reducing the number of degrees of freedom in molecular models-so-called coarse-graining-is a popular approach to increase the accessible time scales and system sizes in molecular dynamics simulations. It involves, however, per se a loss of information. In order to retain a high accuracy in the region of interest, hybrid methods that combine two levels of resolution in a single system are an attractive trade-off. Hybrid atomistic (AT)/coarse-grained (CG) simulations have previously been shown to preserve the secondary structure elements of AT proteins in CG water but to cause an artificial increase in intramolecular hydrogen bonds, resulting in a reduced flexibility of the proteins. Recently, it was found that the AT-CG interactions employed in these simulations were too favourable for apolar solutes and not favourable enough for polar solutes. Here, the AT-CG interactions are reparametrised to reproduce the solvation free energy of a series of AT alkanes and side-chain analogues in CG water, while retaining the good mixing behaviour of AT water with CG water. The new AT-CG parameters are tested in hybrid simulations of four proteins in CG water. Structural and dynamic properties are compared to those obtained in fully AT simulations and, if applicable, to experimental data. The results show that the artificial increase of intramolecular hydrogen bonds is drastically reduced, leading to a better reproduction of the structural properties and flexibility of the proteins in atomistic water, without the need for an atomistic solvent layer.

Self-assembling study of sarcolipin and its mutants in multiple molecular dynamic simulations.

The Sarcolipin (SLN) is a single trans-membrane protein that can self-assembly to dimer and oligomer for playing importantphysiological function. In this work, we addressed the dimerization of wild type SLN (wSLN) and its mutants (mSLNs) - I17A and I20A, using both coarse-grained (CG) and atomistic (AT) molecular dynamics (MD) simulations. Our results demonstrated that wSLN homodimer assembled as a left-handed helical complex, while mSLNs heterodimers assembled as right-handed complexes. Analysis of residue-residue contacts map indicated that isoleucine (Ile)-leucione (Leu) zipper domain played an important role in dimerization. The potential of mean force (PMF) demonstrated that wSLN homodimer was more stable than mSLNs heterodimers. Meanwhile, the mSLNs heterodimers preferred right-handed rather than left-handed helix. AT-MD simulations for wSLN and mSLNs were also in line with CG-MD simulations. These results provided the insights for understanding the mechanisms of SLNs self-assembling. Proteins 2017; 85:1065-1077. © 2017 Wiley Periodicals, Inc.

Adaptive Resolution Simulation of Supramolecular Water: The Concurrent Making, Breaking, and Remaking of Water Bundles.

The adaptive resolution scheme (AdResS) is a multiscale molecular dynamics simulation approach that can concurrently couple atomistic (AT) and coarse-grained (CG) resolution regions, i.e., the molecules can freely adapt their resolution according to their current position in the system. Coupling to supramolecular CG models, where several molecules are represented as a single CG bead, is challenging, but it provides higher computational gains and connection to the established MARTINI CG force field. Difficulties that arise from such coupling have been so far bypassed with bundled AT water models, where additional harmonic bonds between oxygen atoms within a given supramolecular water bundle are introduced. While these models simplify the supramolecular coupling, they also cause in certain situations spurious artifacts, such as partial unfolding of biomolecules. In this work, we present a new clustering algorithm SWINGER that can concurrently make, break, and remake water bundles and in conjunction with the AdResS permits the use of original AT water models. We apply our approach to simulate a hybrid SPC/MARTINI water system and show that the essential properties of water are correctly reproduced with respect to the standard monoscale simulations. The developed hybrid water model can be used in biomolecular simulations, where a significant speed up can be obtained without compromising the accuracy of the AT water model.

Coarse-grained bond and angle distributions from atomistic simulations: On the systematic parameterisation of lipid models

Coarse-grained (CG) models are popular alternatives to atomistic (AT) force fields as they enable simulations of larger systems at longer timescales. The bottom-up approach is a systematic parameterisation strategy whereby data from AT simulations are used to determine the CG parameters. This is particular straightforward with the bond and angle parameters as a direct Boltzmann inversion can be used. Still, a reference AT force field has to be chosen. In this study, I compare three common AT force fields (Stockholm lipids, Berger and Gromos) and investigate the sampling of bond and angle distributions in two CG models (Martini and Elba). As a test case, I choose a bilayer of POPC lipids. The AT simulations give distributions that agree to a large extent, especially in the fatty acid tails. However, the AT simulations sample distributions that differ from the distributions observed in CG simulations with respect to both location and width. The bond and angle distributions from the AT simulations are then used to re-parameterise the CG force fields. For the Martini model, this significantly alters the physical behaviour of the membrane, which likely is an effect of the mapping. However, for the Elba model the re-parameterised force field gives a membrane that is in some respects closer to the experimental membrane. Implications for CG parameterisation are discussed.

Signal Transduction by a Cytokine Receptor: Multi-Scale Computational Studies of the Membrane Associated Gp130 Receptor Complex

Signal transduction is involved in the control of various essential biological processes such as cell growth, regeneration and apoptosis. Cytokines function as regulators of acute phase response during injury of infection, but are also involved in haematopoiesis, liver and neuronal regeneration. Two gp130 cytokine receptor forms a hexameric complex with two alpha-receptors (IL-6Rα) and two interleukin 6 (IL-6). This receptor complex formation initiates signal transduction via the JAK/ STAT pathway, i.e. tyrosine kinases of the Janus family activate signal transducers and activators of transcription. Some structural information is known about the ectodomain of the gp130 receptor complex, while the structures of the transmembrane and juxtamembrane regions are currently unknown.Here we apply multi-scale molecular dynamics simulations to shed light on some of the steps occurring within or in proximity to the membrane during signal transduction via the gp130 receptor complex. Coarse-grained (CG) simulations are performed to capture events occurring on long time-scales such as formations of protein-protein and protein-lipid interactions. The information obtained from the CG simulations is then further refined by conversion to atomistic (AT) simulations. The computational studies highlight the effect of lipid composition on protein-protein assembly within the lipid bilayer. Additionally the consequences of the juxtamembrane regions on protein-protein and protein-lipid interactions patterns have been explored and the results reveal a central role of intracellular basic residues not only on the interactions between proteins but also with the membrane. The protein-protein association has additionally been probed in physiological relevant membranes, which are complex in composition and asymmetric between the upper and lower leaflet.

From the Micelle to the Membrane: Molecular Dynamics Simulations of Solution NMR Structures of Membrane Proteins

The physiological environment of membrane proteins is a complex membrane yet structural studies by necessity are largely based on solubilisation in detergent micelles or model lipid bilayers and bicelles. The choice of solubilising environment has been shown to have some impact on the conformation of membrane proteins; understanding the basis of these changes is critical in linking experimental structures to the conformations expected in native environments.Here we use a combination of coarse-grained (CG) and atomistic (AT) molecular dynamics (MD) simulations to probe the effects of experimental conditions on structure, with particular emphasis on the solution NMR structures of alpha helical membrane proteins including the influenza M2 proton channel and a mitochondrial carrier protein.We show that this multiscale simulation approach may be used to improve the packing of solution NMR structures in a lipid bilayer environment. This technique is also used to probe the roles of specific native lipids (e.g. cardiolipin) on membrane protein structure and dynamics.

Combining coarse‐grained nonbonded and atomistic bonded interactions for protein modeling

A hybrid coarse-grained (CG) and atomistic (AT) model for protein simulations and rapid searching and refinement of peptide-protein complexes has been developed. In contrast to other hybrid models that typically represent spatially separate parts of a protein by either a CG or an AT force field model, the present approach simultaneously represents the protein by an AT (united atom) and a CG model. The interactions of the protein main chain are described based on the united atom force field allowing a realistic representation of protein secondary structures. In addition, the AT description of all other bonded interactions keeps the protein compatible with a realistic bonded geometry. Nonbonded interactions between side chains and side chains and main chain are calculated at the level of a CG model using a knowledge-based potential. Unrestrained molecular dynamics simulations on several test proteins resulted in trajectories in reasonable agreement with the corresponding experimental structures. Applications to the refinement of docked peptide-protein complexes resulted in improved complex structures. Application to the rapid refinement of docked protein-protein complex is also possible but requires further optimization of force field parameters.

Adaptive Resolution Simulation (AdResS): A Smooth Thermodynamic and Structural Transition from Atomistic to Coarse Grained Resolution and Vice Versa in a Grand Canonical Fashion.

The AdResS method in molecular dynamics (MD) allows, in a grand canonical (GC) fashion, to change on-the-fly the number of degrees of freedom of a system, allowing to pass from atomistic (AT) to coarse-grained (CG) resolution and vice versa as a function of the position of the molecule in the simulation box. The coupling of resolutions is made in a thermodynamically consistent way, though in the current formulation, in the region where the molecule changes resolution, neither thermodynamic nor structural properties can be preserved. Here we propose an extension of the method where basic thermodynamic and structural properties can be systematically controlled also in the transition region; this assures a very smooth change from one molecular representation to the other. Moreover, we provide a rigorous argument which shows that if in the region where the molecules change resolution the radial distribution function (RDF) is the same as in the AT and CG region, then the AT region is, from the statistical point of view, equivalent to a subsystem embedded in a larger full AT system, at least up to a second order approximation.

Multiscale molecular dynamics simulations of micelles: coarse-grain for self-assembly and atomic resolution for finer details

A positional interpolation/extrapolation method for the mapping of coarse-grained (CG) to atomistic (AT) resolution is presented and tested for single-component micelles formed by lysophospholipids of different chain length. The target CG nanoaggregates were self-assembled from random mixtures of surfactants in explicit MARTINI water and equilibrated by molecular dynamics simulations, at the microsecond time scale. The ambiguity inherent in the definition of the CG particles was explored by mapping the same CG structures to AT resolution surfactants of different size. After the conversion, the obtained AT micelles were simulated for 250 ns and the resulting trajectories were analyzed in detail. The mean lifetime of the surfactant–solvent interactions as well as the lateral diffusion coefficients of the surfactant molecules within the micellar aggregates were obtained for the first time. The results suggest that the individual molecules within the micelle behave like lipids in a fluid membrane. The employed mapping back method is efficient and versatile, as it can be applied to diverse combinations of force fields and systems with a minimum of code development. In a general context, this work illustrates the power of multiscale molecular dynamics simulations for the generation and subsequent examination of self-assembled structures, including the fine characterization of structural and dynamic properties of the resulting aggregate.

Interactions Between a Voltage Sensor and a Toxin via Multiscale Simulations

Gating-modifier toxins inhibit voltage-gated ion channels by binding the voltage sensors (VS) and altering the energetics of voltage-dependent gating. These toxins are thought to gain access to the VS via the membrane (i.e., by partitioning from water into the membrane before binding the VS). We used serial multiscale molecular-dynamics (MD) simulations, via a combination of coarse-grained (CG) and atomistic (AT) simulations, to study how the toxin VSTx1, which inhibits the archeabacterial voltage-gated potassium channel KvAP, interacts with an isolated membrane-embedded VS domain. In the CG simulations, VSTx1, which was initially located in water, partitioned into the headgroup/water interface of the lipid bilayer before binding the VS. The CG configurations were used to generate AT representations of the system, which were subjected to AT-MD to further evaluate the stability of the complex and refine the predicted VS/toxin interface. VSTx1 interacted with a binding site on the VS formed by the C-terminus of S1, the S1-S2 linker, and the N-terminus of S4. The predicted VS/toxin interactions are suggestive of toxin-mediated perturbations of the interaction between the VS and the pore domain of Kv channels, and of the membrane. Our simulations support a membrane-access mechanism of inhibition of Kv channels by VS toxins. Overall, the results show that serial multiscale MD simulations may be used to model a two-stage process of protein-bilayer and protein-protein interactions within a membrane.

Membrane/Toxin Interaction Energetics via Serial Multiscale Molecular Dynamics Simulations.

Computing free energies of complex biomolecular systems via atomistic (AT) molecular dynamics (MD) simulations remains a challenge due to the need for adequate sampling and convergence. Recent coarse-grained (CG) methodology allows simulations of significantly larger systems (∼10(6) to 10(8) atoms) over longer (μs/ms) time scales. Such CG models appear to be capable of making semiquantitative predictions. However, their ability to reproduce accurate thermodynamic quantities remains uncertain. We have recently used CG MD simulations to compute the potential of mean force (PMF) or free energy profile of a small peptide toxin interacting with a lipid bilayer along a 1D reaction coordinate. The toxin studied was VSTx1 (Voltage Sensor Toxin 1) from spider venom which inhibits the archeabacterial voltage-gated potassium (Kv) channel KvAP by binding to the voltage-sensor (VS) domains. Here, we re-estimate this PMF profile using (i) AT MD simulations with explicit membrane and solvent and (ii) an implicit membrane and solvent (generalized Born; GBIM) model where only the peptide was explicit. We used the CG MD free energy simulations to guide the setup of the corresponding AT MD simulations. The aim was to avoid local minima in the AT simulations which would be difficult over shorter AT time scales. A cross-comparison of the PMF profiles revealed a conserved topology, although there were differences in the magnitude of the free energies. The CG and AT simulations predicted a membrane/water interface free energy well of -27 and -23 kcal/mol, respectively (with respect to water). The GBIM model, however, gave a reduced interfacial free energy well (-12 kcal/mol). In addition, the CG and GBIM models predicted a free energy barrier of +61 and +96 kcal/mol, respectively, for positioning the toxin at the center of the bilayer, which was considerably smaller in the AT simulations (+26 kcal/mol). Thus, we present a framework for serially combining CG and AT simulations to estimate the free energy of peptide/membrane interactions. Such approaches for combining simulations at different levels of granularity will become increasingly important in future studies of complex membrane/protein systems.

Molecular Dynamics Simulations of the Dimerization of Transmembrane α-Helices

The lateral association of transmembrane (TM) α-helices within a lipid bilayer environment is a key stage in the folding of membrane proteins. It may also play a role in signalling across cell membranes. Dimerization of TM helices provides a simple example of such lateral association. Direct atomistic (AT) resolution MD simulation of self-assembly of a TM helix bundle remains challenging. AT-MD may be complemented by coarse-grained (CG) simulations. We demonstrate how CG-MD may be used to simulate formation of dimers of TM helices. We also show how a serial combination of CG and AT simulation provides a multi-scale approach for generating and refining models of TM helix dimers. This approach has been applied to a number of examples, including the glycophorin TM helix dimer (a paradigm for helix/helix packing) [1], and the TM domain of the syndecan-2 receptor protein, which contains a GxxxG motif comparable to that of glycophorin. The multi-scale approach has also been applied to a more complex system, the heterodimeric αIIb/β3 integrin TM helix dimer.[1] Psachoulia, E., P. J. Bond, P. W. Fowler, and M. S. P. Sansom. 2008. Helix-helix interactions in membrane proteins: coarse grained simulations of glycophorin helix dimerization. Biochem. 47:10503-105012.

Molecular Dynamics Simulations of the Dimerization of Transmembrane α-Helices

Membrane proteins account for nearly a quarter of all genes, but their structure and function remain incompletely understood. Most membrane proteins have transmembrane (TM) domains made up of bundles of hydrophobic alpha-helices. The lateral association of TM helices within the lipid bilayer is a key stage in the folding of membrane proteins. It may also play a role in signaling across cell membranes. Dimerization of TM helices is a simple example of such lateral association. Molecular dynamics (MD) simulations have been used for over a decade to study membrane proteins in a lipid bilayer environment. However, direct atomistic (AT) MD simulation of self-assembly of a TM helix bundle remains challenging. AT-MD may be complemented by coarse-grained (CG) simulations, in which small numbers of atoms are grouped together into particles. In this Account, we demonstrate how CG-MD may be used to simulate formation of dimers of TM helices. We also show how a serial combination of CG and AT simulation provides a multiscale approach for generating and refining models of TM helix dimers. The glycophorin A (GpA) TM helix dimer represents a paradigm for helix-helix packing, mediated by a GxxxG sequence motif. It is well characterized experimentally and so is a good test case for evaluating computational methods. CG-MD simulations in which two separate TM helices are inserted in a lipid bilayer result in spontaneous formation of a right-handed GpA dimer, in agreement with NMR structures. CG-MD models were evaluated via comparison with data on destabilizing mutants of GpA. Such mutants increased the conformational flexibility and the dissociation constants of helix dimers. GpA dimers have been used to evaluate a multiscale approach: A CG model is converted to an AT model, which is used as the basis of an AT-MD simulation. Comparison of three AT-MD simulations of GpA, one starting from a CG model and two starting from NMR structures, leads to convergence to a common refined structure for the dimer. CG-MD self-assembly has also been used to model dimerization of the TM domain of the syndecan-2 receptor protein. This TM helix contains a GxxxG motif, which mediates right-handed helix packing comparable to that of the GxxxG motif in GpA. The multiscale approach has been applied to a more complex system, the heterodimeric alphaIIb/beta3 integrin TM helix dimer. In CG-MD, both right-handed and left-handed structures were formed. Subsequent AT-MD simulations showed that the right-handed structure was more stable, yielding a dimer in which the GxxxG motif of the alphaIIb TM helix packed against a hydrophobic surface of the beta3 helix in a manner comparable to that observed in two recent NMR studies. This work demonstrates that the multiscale simulation approach can be used to model simple membrane proteins. The method may be applied to more complex proteins, such as the influenza M2 channel protein. Future refinements, such as extending the multiscale approach to a wider range of scales (from CG through QM/MM simulations, for example), will expand the range of applications and the accuracy of the resultant models.