Induced Fit in Molecular Docking

Induced conformational changes upon molecular association are studied and methods to predict these changes are presented. The flexible molecular interface methodology was applied to drug repositioning, de novo inhibitor discovery and lead optimization.

Simulating the early steps of amyloid fibril formation and disassembly

More than 20 human diseases are associated with the pathological self-assembly of soluble proteins into transient cytotoxic oligomers and amyloid fibrils. Alzheimer's disease, affecting today more than 15 million people world-wide, is characterized by the aggregation of the Abeta40/Abeta42 peptides.
Because aggregation is very complex, structural characterization of the intermediate species remain to be determined. Similarly, though N-methylated Abeta16-22 peptides inhibit the fibrillogenesis of full-length Abeta and disassemble fibrils in vitro, there is little information about their mechanism of action.
Here, we review our current understanding of the dynamics and free energy surface of both amyloid-forming peptides [Abeta(16-22), beta2m(83-89, Abeta42] and non-amyloid sequences [e.g. a scrambled sequence of beta2m(83-89)] using a coarse-grained protein force field (OPEP) coupled to the activation-relaxation technique, molecular dynamics (MD) and replica exchange MD (REMD) simulations.
Then, we present coarse-grained MD and REMD simulations of Abeta(16-22) oligomers with multiple copies of an N-methylated inhibitor.
This work was done in collaboration with Normand Mousseau (University of Montreal, Canada), Guanghong Wei (Fudan University, China) and Yassmine Chebaro (IBPC, Paris).

Atomically detailed simulations of kinetics in molecular biophysics by milestoning

Milestoning is a technique to significantly extend the time scale of Molecular Dynamics simulations. It computes explicitly the time evolution of the system even if the process is not activated and diffusive. The calculations are done in two steps. First, short time trajectories are used to compute Local First Passage Time Distributions (LFPTD) between nearby “Milestones” along a reaction coordinate. In the second step a stochastic non-Markovian integral equation that uses the LFPTD computes the overall time course of the reaction.
I will describe the theory and implementation of the technique and continue to discuss a few applications that include solvated alanine dipeptide, folding of a 21 amino acid helical protein in explicit solvent, and the recovery stroke in myosin.
Myosin is a molecular machine that provides a power stroke in muscles at the cost of an ATP. Once the power stroke is completed the molecule recovers to its pre-stroke state. The relaxation step is called the recovery stroke. The recovery stroke does not require the investment of biological energy and is spontaneous and reasonably rapid in the presence of ATP.
We computed a reaction coordinate for this process using a functional optimization that provides the minimum energy path. For a molecular machine that is designed for efficiency and minimal energy loss it is expected that the process will follow a single well-focused reaction coordinate. Indeed the Milestoning calculation that uses the computed reaction coordinate (but included all degrees of freedom of an atomic and fully solvated myosin molecule) reproduces well the time scale and the expected features of the free energy surface.

Simulating biomolecules in cellular environments

The nature of the solvent environment is a critical factor in determining biomolecular structure and dynamics. Molecular simulations are commonly carried out in dilute aqueous solvent as an approximation of cellular environments. In reality, cellular environments are rather dense and highly heterogeneous. Such environments may be represented effectively based on a continuum model of its dielectric properties and the steric hindrance encountered in dense cellular environments.
Extended replica-exchange simulations of biomolecules with continuum approximations of cellular environments are presented and compared with simulations in aqueous solvent. It is found that moderately lowered dielectric environments that are likely present in cellular environments can significantly affect the conformational sampling of peptides.

Computer Simulations of Protein-Protein Association in Water and at Membranes

Protein-protein interactions are key components of most biological processes. On the one hand, about half of all cellular proteins appear to be parts of larger stable protein complexes. On the other hand, transient, pairwise protein-proteins interactions are crucial parts of bioenergetic and signal transduction pathways. Experimental work showed that complementary electrostatic interactions often accelerate the association processes by several orders of magnitude, but not always. Here, we will concentrate on such fast assembling protein pairs that are amenable to Brownian and molecular dynamics simulations, and attempt to identify the energetic principles for these binding phenomena. First, we have studied the association free energy landscape for the barnase:barstar complex by Brownian Dynamics simulations. We showed that single protein mutations can drastically alter the shape of the energy landscape and the location of the encounter complex. The same methodology was then used to study cytochrome c association to the integral membrane protein cytochrome c oxidase. We found that the diffusional approach was favored along the polar membrane surface. Finally, we will show recent results from unbiased molecular dynamics simulations for the binding process of a proline-rich peptide to an SH3 domain.

Multiscale modeling of protein and protein assemblies

Systematic sequencing of numerous genomes provides enormous library of protein sequences. Only for a small fraction of these proteins their three dimensional structures have been determined. Knowledge of protein structures is necessary for understanding and controlling of protein biological function, from enzymatic activity, through transport and signaling, to mechanisms and thermodynamics of complex macromolecular assemblies. It is also important to know protein folding mechanisms and the dynamic and thermodynamic characteristics of the denatured state. Understanding protein dynamics and folding mechanisms may be even more challenging than theoretical prediction of protein structure. Classical methods of molecular dynamics are applicable only to not too large systems and/or to a relatively narrow time frame. The time-scale of biomacromolecular processes is usually orders of magnitude wider. Therefore, simplified models could be very useful in a large scale molecular modeling. Reduced space CABS (Ca, Cb, Side chain representation) methodology proven to be one of the best performing methods for protein structure prediction. It has been demonstrated during the sixth CASP (Critical Assessment of protein Structure Prediction) community-wide experiment that the CABS based modeling technique is applicable to entire spectrum of structure prediction problems; from comparative modeling to de novo prediction of new folds. In this contribution we describe some of the newest applications of the CABS modeling technique. These applications include: prediction of proteinstructure, modeling of folding mechanisms, protein protein interactions and fully flexible docking of peptide-type ligands to protein receptors. The CABS reduced model could be easily integrated with all-atom approaches. Namely, the united atom models from CABS simulations are accurate enough for reasonable fast rebuilding of atomic details. Consequently, the proposed methodology enables multiscale simulations of large biomolecular systems.

Peptide adhesion and friction: Theoretical approaches

Single-molecule behavior combines the fields of non-equilibrium thermodynamics, elasticity theory and hydrodynamics. Theoretical approaches thus rely on molecular simulations, continuum modeling and scaling approaches.

This is demonstrated with a few examples:

• Spider silk consists of polypeptides with highly repetitive motives and readily adsorbs on hydrophobic and hydrophilic surfaces. Single molecule AFM studies yield adsorption energies and point to an extremely high mobility on hydrophobic surfaces. The dominant hydrophobic attraction can be quantitatively explained with classical MD simulations including explicit water. Both water structural effects and dispersion interactions contribute to this solvation attraction.
• The friction coefficient of bound polymers is very low on hydrophobic substrates, which is traced back to the presence of a vacuum layer between substrate and water, which forms a lubricating cushion on which a polymer can glide. Conversely, friction forces on hydrophilic substrates are large and make determining the equilibrium binding constant in computer simulations impossible.
• Shear-flow induced unfolding of proteins plays an important role in starting the coagulation cascade in small blood vessels. In the theoretical modeling the unfolding is initiated by single-chain protrusion-like excitations and leads to a hydrodynamic unfolding transition, which is well captured by a scaling nucleation argument.

Dominant Pathways in Protein Folding

The Langevin dynamics of the folding of a protein can be expressed as a path integral over all trajectories which join an unfolded state to the native state of the protein. The dominant contribution to the path integral comes from the paths which minimize the effective action between the denatured and the native state. We show how such paths can be computed numerically, and how they can be used to describe the folding, to determine transition states, and eventually to compute folding rates.

Scalable Systems for Computational Biology

Driving up processor clock speed is about to hit several limitations, a soaring energy consumption being among them. The BlueGene system from IBM is shown to provide performance in a scalable and energy efficient way. Results from NAMD and similar codes show, how the system can be used in the area of Computational Biology.

High performance computing in multiscale modeling cardiac contraction: Bridging proteins to cells to whole heart

The availability of increased computing with thousands of computational cores enables new classes of biological models that include detailed representations of proteins and protein complexes with spatial interactions. We develop such a model of the interaction of actin and myosin in the cardiac sarcomere. The model includes explicit representations of actin, myosin, and regulatory proteins. Although this is not an atomic-scale model, as would be the case for molecular dynamics simulations, the model seeks to represent spatial interactions between protein complexes that are thought to produce characteristic cardiac muscle responses at larger scales. While the model simulates the microscopic scale, when model results are extrapolated to larger structures, the model recapitulates complex, nonlinear behavior such as the steep calcium sensitivity of developed force in muscle structures. By bridging spatial scales, the model provides a plausible and quantitative explanation for several unexplained phenomena observed at the tissue level in cardiac muscles. Model execution entails Monte-Carlo-based simulations of Markov representations of calcium regulation and actin-myosin interactions. The model is computationally expensive and requires a supercomputer to simulate sub-cellular structures. While useful to understand biophysical questions, such detailed models are obviously impractical to model the billions of cells that comprise a whole human heart. We have also developed a more computational efficient model that approximated the spatial interactions at the protein level without explicit computation. The goal of this work is to bridge from cells to large organ-level anatomical structures with practical run times. We hope that the power of this approximate model to recapitulate complex force responses in cardiac tissue will foster wider use of cardiac models for research and clinical applications. The work is a case study in multiscale biological modeling where the development of a complex, detailed model was required to guide the later development of a more abstract and computationally efficient representation.

Bioinformatics, metabolomics, and systems biology

Complex biological phenomena like complex diseases, pathogen/host interactions or microbial performance in fermentation processes are reactions of the whole biological system and can only be understood as such and predicted successfully by simulation of molecular networks in the cell. The most complex of these systems is the metabolic network, which represents the main focus of experimental and bioinformatics research and development at the Schomburg group. In the context of genotype/phenotype correlation by characterisation and simulation of molecular networks the talk will address the following points:

• Metabolomics
• Experimental facts and standards as requirement for systems biology approaches. The need for nomenclature and experimental standardisation for accessibility and comparability of experimental data.
• The bioinformatics analysis of molecular networks.
• Approaches for the simulation of metabolic networks. Fluxes and pools of metabolites. Constraint-based flux analysis.
• The current developments at BRENDA, the world‘s largest enzyme information system.

Computer simulation of biomolecular systems: where do we stand?

Computation based on molecular models is playing an increasingly important role in biology, biological chemistry, and biophysics. Since only a very limited number of properties of biomolecular systems is actually accessible to measurement by experimental means, computer simulation can complement experiment by providing not only averages, but also distributions and time series of any definable - observable or non-observable - quantity, for example conformational distributions or interactions between parts of molecular systems. Present day biomolecular modelling is limited in its application by four main problems: 1) the force-field problem, 2) the search (sampling) problem, 3) the ensemble (sampling) problem, and 4) the experimental problem. These four problems will be discussed and illustrated by practical examples. Perspectives will be outlined for pushing forward the limitations of molecular modelling.

Advancing Drug and Protein Binding Affinity Predications via Generalized Ensemble Based Methods

In the past decades, computer modeling has been intimately integrated in the industrial processes for protein engineering and drug discoveries. With the guidance of computational predications, the time-costly and labor-intensive experimental trials can be reduced dramatically. Among various available computational methods, alchemical free energy simulation technique, based on molecular dynamics or Monte Carlos simulations, is unique; this is, if simulation time permits and energy potential is reasonably accurate, it can ultimately lead to quantitative predications of free energy values corresponding to the processes of interest, because this method is built upon solid statistical mechanical theories. Unfortunately, with the present computing power and in particular the state-of-the-art free energy simulation algorithms, reaching adequate simulation time for nice free energy convergence, in particular to fulfill the efficiency requirement for the screening purpose, is still quite challenging. Various factors have contributed to such challenge; these include inadequate sampling, poor potential descriptions, and even their couplings due to the fact that accurate potential description is always made possible via the sacrifice of sampling cost or vice versa when the computing resources are limited. Based on our intensive preliminary developments, the following missions have been step-wisely met in the developments of free energy simulation methods for the purpose of drug and protein binding predications:

1. the interrelated phase space overlap sampling and conformational sampling problems have been resolved to ensure fast free energy convergence;
2. the developed enhanced sampling treatments have been tailored for hybrid quantum mechanical/molecular mechanical (QM/MM) potential descriptions, where small molecule ligands can be treated quantum mechanically and protein/solvent environment can be treated classically;
3. the developed enhanced sampling treatments have been further engineered for the purpose of screening multiple drug candidates (or amino acid candidates).

Protein-ligand docking including protein flexibility – an hierarchical approach

To describe protein-ligand interactions realistically, it is necessary to account for the structural changes in both the receptor and the ligand during complex formation. However, in classical docking approaches the receptor is treated either rigid of semi-rigid. We developed new program, DynaDock, which allows full flexibility for both, the ligand and the receptor during docking. For this purpose we combined two new methods: IRECS (Iterative REduction of Conformational Space, Hartmann et al., Protein Science 16(7), 1294-1307, 2007) and OPMD (Optimized Potential Molecular Dynamics), which together allow an efficient sampling of ligand and receptor conformations.
The most commonly used methods for an efficient treatment of side chain flexibility during docking are ensemble based methods (FlexE, Claußen et al., J. Mol. Biol, 308, 377-395, 2001). Due to the large number of possible side chain conformations at the receptors binding interface, these docking approaches often face the problem of a combinatorial explosion of all possible side chain conformations. Thus it is crucial to preselect a small number of flexible side chains in the binding site, for which alternative conformations are used. IRECS was developed for this purpose and allows predicting an ensemble of the most probable conformations for each side chain of a protein. The numbers of rotamers that are assigned to each side chain correspond to the side chain’s flexibility, thus leading to a minimal, flexibility optimized set of binding site side chain conformers. The predicted side chain ensembles can be used for ensemble based docking (FlexE). To further refine the docking poses, we developed a new sampling algorithm, OPMD, which combines random/Monte Carlo sampling with molecular dynamics.
The performance of IRECS was evaluated on a set of 160 crystal structures of proteins. Predicting only a single conformation per side chain, IRECS achieved an accuracy of 84.7%. The OPMD approach was evaluated in the context of protein-peptide docking on a test set of 20 protein-ligand complexes, which represent a wide range in ligand size (2-11mers) and solvent exposure. Docked and refined poses with a ligand RMSD of less than 2.0 Å from the corresponding experimental structure could be obtained for all systems. In addition both algorithms were applied to the refinement of MHC-peptide(9mer) complexes predicted by DynaPred (Antes el al., Bioinformatics 22(14), 16-24, 2006). Using the original DynaPred algorithm we had obtained peptide conformations with an average backbone RMSD of 1.53 Å and side chain RMSDs between 1.02 and 3.84 Å, depending on the side chain’s solvent exposure (similar values were obtained by other docking methods). After IRECS/OPMD refinement all all-atom RMSDs of the peptide ligands were smaller than 2.0 Å.

Kinetics, Statistics, and Energetics of Lipid Membrane Electroporation Studied by Molecular Dynamics Simulations

Membrane electroporation is the method to directly transfer bioactive substances such as drugs and genes into living cells, as well as preceding electrofusion. Although much information on the microscopic mechanism has been obtained both from experiment and simulation, the existence and nature of possible intermediates is still unclear.
A statistical theory is developed to facilitate direct comparison of experimental (macroscopic) prepore formation kinetics with the preporation times derived from 50 atomistic electroporation simulations, which also allows to extract an effective number of lipids involved in each pore formation event [1]. A linear dependency of the activation energy for prepore formation on the applied field is seen, with quantitative agreement between experiment and simulation. The distribution of preporation times suggests a four state pore formation model. An average pore radius of approx. 0.5nm is seen, in favourable agreement with conductance measurements and electrooptical experiments of lipid vesicles.

[1] R. A. Bockmann et al. Biophys. J. (2008) in press

Probing the Prion Hydration by Molecular Dynamics Simulations: from native via misfolded to amyloid conformations

Water at the surface of proteins plays a crucial biological role. MD simulations have been employed to investigate the hydration properties of the Prion Protein (PrP), whose misfolding is associated to Transmissible Spongiform Encephalopathies. The data focus on different states along a likely PrP misfolding pathway.
For the native state, MD hydration analyses pointed out protein regions where tightly bound waters evidently add to the local structural stability (1). Among these, a particular hydration site has been suggested to disfavor PrP aggregation (2). Moreover, specific PrP surfaces were evidenced for being in contact with bulklike solvent (1).
REMD sampling of PrP conformational free energy evidenced a partially unfolded state showing peculiar structural, dynamical and hydration features of a likely aggregating-prone specie (3).
Finally, the hydration properties of “steric zipper” amyloid-like structures were addressed for the models of polyglutamine (4) and PrP(170-175) fragment (5).

References
1. De Simone et al, PNAS (2005) 102:7535-7540.
2. De Simone et al, FEBS Letters (2006) 580:2488-2494.
3. De Simone et al, Biophys J. (2007) 93:1284-1292.
4. Colombo et al, Proteins (2008) 70:863-872.
5. De Simone et al, BBRC (2008) 366:800-806.

Modeling Conformational and Molecular Weight Heterogeneity with Analytical Ultracentrifugation Experiments (AUC)

Sedimentation velocity experiments reveal information about molecular weight and shape of sedimenting solutes in noninteracting and dynamically reacting systems. The observables in such experiments are the sedimentation and diffusion coefficients and the concentration of individual solutes. We have developed novel parallel optimization algorithms that allow us to extract molecular parameters from sedimentation experiments which are parameterized in a model-independent approach. Using a mixture of deterministic and stochastic optimization, we are able to fit analytical ultracentrifugation experiments in a global fashion with excellent convergence properties. Our software uses the TIGRE grid middleware to distribute the computing effort to Teragrid and other computing resources, and offers a public web portal for the hydrodynamic analysis of AUC experiments (http://uslims.uthscsa.edu). Our solutions provide unparalleled resolution, and allow us to characterize polymerization events, modes of aggregation and provide high resolution information in structure and function studies in the solution state.

Interactions of aminoglycosidic antibiotics with the 30S subunit - Brownian dynamics study

Aminoglycosidic antibiotics are a family of antibacterial drugs. Most of them target the prokaryotic A-tRNA binding site in the 30S ribosomal subunit. Upon binding to the ribosomal RNA aminoglycosides interfere with translation by decreasing the accuracy of the decoding process.
We study the mechanism and kinetics of association of aminoglycosidic antibiotics with the ribosomal RNA. We concentrate on their diffusion toward the RNA and the formation of the encounter complex. Using Brownian dynamics methodology, we simulate the association of various aminoglycosides to the model oligonucleotide containing the ribosomal A-site and to the whole 30S ribosomal subunit. The latter case is especially challenging due to the size (approximately 95000 atoms) and high overall charge of the 30S subunit. We will present how structural and electrostatic properties of antibiotics influence their association and discuss the competition between antibiotics and magnesium ions upon binding at the A-site.

Simplified approaches to complex biological systems

Some of the emerging goals in modern medicine are to uncover the molecular origins of human diseases, and ultimately contribute to the development of new therapeutic strategies to rationally abate disease. Of immediate interests are the roles of molecular structure and dynamics in certain cellular processes leading to human diseases and the ability to rationally manipulate these processes. Despite recent revolutionary advances in experimental methodologies, we are still limited in our ability to sample and decipher the structural and dynamic aspects of single molecules that are critical for their biological function. Thus, there is a crucial need for new and unorthodox techniques to uncover the fundamentals of molecular structure and interactions. We follow a hypothesis-driven approach which is based on tailoring simplified protein models to the systems of interest. Such an approach allows significantly extending the length and time scales for studies of complex biological systems. I will describe several recent studies that signify the predictive power of simplified protein models within the hypothesis-driven modeling approach utilizing developed in our laboratory rapid Discrete Molecular Dynamics (DMD) simulations.

Insights from atomistic computer simulations of halophilic proteins

Halophilic (salt-loving) proteins, typically found in Archaea, can maintain their native structure and function in aqueous environment only at relatively high salt concentrations (>1-2M). As they are highly negatively charged at physiological conditions the competition between hydrophobic and hydrophilic solvation is strongly amplified and tuned by salt type and concentration. By performing atomistic molecular dynamics (MD) computer simulations the influence of salt on effective interactions between amino acids, protein secondary structures, and the stability of small coiled-coil proteins is investigated. Possible salt-induced specific and non-specific (de)stabilization mechanisms are identified and critically discussed.

Predicting Protein Interactions from Functional Specificity using Multi-Relief and multi-Harmony

Many protein families contain sub-families that exhibit functional specialization, which often implies differences in protein-protein interactions. We will present a two-track approach to to exploit this functional specificity for known protein interactions. The first track focuses on the detection of sub-type specific sites. For this we extended the Sequence Harmony method introduced previously [NAR 35:W495; NAR 34:6540] to handle multiple sub-groups. In addition we have developed a feature-selection method, multi-Relief [Bioinf.24:18], as an alternative specificity detection method. Protein interactions always occur at the surface, and very often involve loop regions. As a first test-case we evaluate the prevalence of surface residues in the selected sites. For an additional, more limited test-set of protein complex structures, we directly evaluate the selection of interface residues.

Short membrane proteins from viruses: channel-pore dualism?

The genomes of some viruses encode small membrane proteins which are known to alter membrane permeability by forming ion conducting pores. The alteration of the electrochemical gradient as a consequence of ‘channel activity’ has large scale consequences as it induces the fusion and budding process of the virion.
Viral channel or pore forming proteins adopt different topologies. The working hypothesis is that the proteins diffuse as monomers in the lipid membrane and finally self-assemble to form the functional unit. Self-assembly has to take place at the level of the tertiary and quaternary structures within the low dielectric medium of the lipid membrane.
Computational techniques are used to analyze the assembly process. Ion flux is simulated using steered molecular dynamics (MD) simulations and analyzed using Langevin equation of motion. Conductance measurements flank the in silico investigations. Data of the channel forming protein Vpu from HIV-1 will be shown as a test case.

Coarse-Grained Molecular Models for High-Throughput and Multi-Scale Functional Investigations

We discuss the development of a physics-based, coarse-grained molecular model and its subsequent application to in silico functional investigation. Such models overcome shortcomings of both sequence-based bioinformatics (no physics) and molecular dynamics simulations (large CPU requirements, therefore sampling issues) as an integrative approach.

We demonstrate this claim by showing

• capabilities to investigate macromolecular thermodynamics [here: assembly of the bacterial ribosome [1]]
• integrating information into the physical realm [here HIV protease [2] and drug resistance[3]]

Some Aspects of RNA Folding Studied by Lattice Simulations

We consider the RNA molecule as a self-avoiding walk on 3D lattice and use a slightly modified Wang-Landau sampling method to simulate its behavior. The basic goal is to obtain information on thermodynamics of RNA folding. The energy function includes hydrogen bond energy, stacking interactions, chain rigidity, electrostatic interactions. In some simulations we allow formation of pseudoknots and track the change of folding time. To study the folding behavior we calculate thermodynamic quantities such as free energy and specific heat as well as structural parameters like gyration radius and end-to-end distance.

Designing an automatic pipeline for protein structure prediction

Thanks to international effort in the genome sequencing projects, enormous library of protein sequences is now available. Despite extensive efforts in structural genomics, the number of experimentally determined protein structures — typically by costly X-ray crystallography or NMR spectroscopy procedures— is lagging far behind the number of protein sequences. Since proteins are involved in practically all functions performed by a cell, knowledge of protein structures is necessary for understanding and controlling molecular mechanisms of life. Current assumptions are, that for a large fraction of proteins whose structures will not be determined experimentally, computational methods can provide valuable information.
Our modeling technology is based on the CABS model [1], extensively tested, state-of-the-art approach to protein structure prediction. During the community-wide, blind prediction experiment CASP6, aimed to assess the prediction capabilities, the methodology based on the CABS model ranked as the second best among about 200 groups participating.
The modeling process is divided into two stages: fold assembly (by the CABS model in a simplified representation) followed by the model refinement/selection procedure, using an all-atom representation and more exact interaction scheme enabling high resolution structure prediction [2]. Fold assembly can be realized by a standard comparative modeling procedure, where spatial restraints are produced based on alternative sequence alignments with a template/templates. Preferentially in more difficult modeling cases, a new approach to comparative modeling can be used, which does not require the implicit alignment [3].
Selvita's goal is to provide integrated tool kit for automated protein structure predictions. However, like blind prediction experiments shows, due to high complexity of prediction tasks, fully automated approach often doesn't guarantee the highest possible performance. Therefore, human intervention is made possible at all modeling stages.

[1] A. Kolinski. Protein modeling and structure prediction with a reduced representation. Acta Biochim. Pol. 51:349-371, 2004
[2] S. Kmiecik, D. Gront, A. Kolinski. Towards high-resolution protein structure prediction. Fast refinement of reduced models with all-atom force field. BMC Structural Biology 7:43, 2007
[3] A. Kolinski, D. Gront. Comparative modeling without implicit sequence alignments. Bioinformatics 23:2522-27, 2007

Folding and aggregation of model amyloid peptides in explicit solvent and at an interface

The development of therapeutic treatments against amyloid diseases requires an understanding of the (mis)folding and aggregation of fibrillogenic species at a microscopic level. An indispensable tool to study these systems is provided by computer simulations. The most accurate treatment consists of a full atomistic description of peptides and solvent environment. We present molecular dynamics simulations of the folding and aggregation of various fibrillogenic peptides in explicit water and at a water/vapor interface. In part, elevated temperatures or replica exchange were used to enhance sampling. The systems consisted of 12- or 18-residue model amyloid peptides derived from natural sequences (LSFD and B18), the (25-35) fragment of the amyloid beta peptide related to Alzheimer’s disease, and a synthetic peptide with sequence G(VT)5. The latter was studied in a beta-rich crystalline and the others in mono- and oligomeric states.

Modelling the free energy of polypeptides in different environments

An estimate of the free energy for increasing the content of helical and elongated backbone segments in polypeptides is given.
Computer simulations of "reasonable" random walks of isolated single chain all-atom models of a polypeptide are performed and the potential energy of the obtained configurations is refined by using mean-field models for the molecular environment. The entropic contribution is computed assuming that entropy depends only on the enumeration of configurations (density of states) for each single chain.
The free energy estimate is applied on several homo polypeptides (X40 with X=G,A,V,T,K,E) and to the Abeta(1-40) peptide involved in Alzheimer disease. Also several peptides with the Abeta(1-40) sequence randomly scrambled and the mini-protein villin headpiece HP(1-36) are analyzed.

Protein-ligand Docking with a Two-scale Receptor Dynamics and a QM/MM Interaction Potential

A new docking model is presented. The docking process is correlated with global deformations (normal modes) of a receptor [1] and with reorientations of its side chains. Deformations are derived from an elastic network model [2]. Microscale motions of individual residues are generated using a stochastic potential derived from a library [3]. A basin-hopping Monte Carlo of a ligand is carried out using the SCC-DFTB energy plus protein – ligand Lennard-Jones interactions. The electrostatic contribution is computed using SCC-DFTB atomic charges in the presence of the receptor field, which accounts for the ligand polarization effects. CM3/SCC-DFTB charges [4] will also be implemented.

Acknowledgements. These studies are supported using N207 022 31/1172 funds.

References:
[1] M. Zacharias, Proteins 54, 759-767 (2004)
[2] K. Hinsen, Proteins 33, 417-429 (1998)
[3] R.P. Shetty et al., Protein Engineering, vol. 16, pp. 963-969 (2003)
[4] J.A. Kalinowski et al., J. Phys. Chem. A, 108, 2545-2549 (2004)

New Force Replica Exchange Method and Mechanical Unfolding of Proteins

We present our new force replica exchange method for efficient configurational sampling of long biomolecules subjected to an external force [1]. This method was applied to obtain the temperature-force phase diagram of the three-domain Ubiquitin.
Concerning mechanical unfolding, it is shown that the resistance of proteins to mechanical perturbation is defined by their secondary structures [2]. Helix-rich proteins are mechanically less stable compared to beta-rich proteins. The distance between the native state and the transition state depends on the helix content linearly.
The contact order, which is a measure of fraction of local contacts, was found to strongly correlate with the stability and the shape of free energy landscapes. Thus our study reveals that the nature of mechanical resistance of proteins is surprisingly simple: mechanical stability and the distance from the transition to the native state are determined either by the content of secondary structure or by the contact order. We have found simple equations to describe this relationship [2].

1. M. Kouza, C.K. Hu, and Mai Suan Li, J. Chem. Phys 128, 045103 (2008)
2. Mai Suan Li, Biophys J. 93, 2644 (2007).

Coloring the Mu transpososome

Tangle analysis has been applied successfully to study proteins which bind two segments of DNA and can knot and link circular DNA. We show how tangle analysis can be extended to model any stable protein-DNA complex.
We discuss a computational method for finding the topological conformation of DNA bound within a protein complex. We use an elementary invariant from knot theory called colorability to encode and search for possible DNA conformations. We apply this method to analyze the experimental results of Pathania, Jayaram, and Harshey (Cell 2002). We show that the only topological DNA conformation bound by Mu transposase which is biologically likely is the five crossing solution found by Pathania et al (although other possibilities are discussed).
Our algorithm can be used to analyze the results of the experimental technique described in Pathania et al in order to determine the topological conformation of DNA bound within a stable protein-DNA complex.

Computer Modeling of Small Ligands Diffusion in Drosophila Melanogaster Hemoglobin

The monomeric, intracellular hemoglobin from the fruit fly Drosophila Melanogaster (DmHb) has been discovered in 2005. It came out that the oxygen supply system in insects is more complex then previously thought. Details on diffusion of gases, ligands discrimination and hexa- to pentacoordination changes remain unclear. Here we present the results of molecular modeling and molecular dynamics (MD) simulations of gaseous ligands (O2, CO, NO) transport inside the DmHb matrix. In addition to a classical MD trajectory an approximate Locally Enhanced Sampling method (LES) and Implicit Ligand Sampling (ILS) have been employed. The structural and thermodynamics features of hexacoordinated and pentacoordinated DmHb were examined and compared to our previous results obtained for human neuroglobin and cytoglobin [1] which display similar heme coordination. Several connected cavities and diffusion pathways, based on 3D ILS free-energy maps, have been indicated. Residues that are critical for kinetics of small gaseous ligands diffusion in primitive hemoglobin were discovered. These data may help do understand the impact of evolutionary pressure on proteins’ architecture.

[1] Orlowski S., Nowak W., 2007. Locally enhanced sampling molecular dynamics study of the dioxygen transport in human cytoglobin. Journal of molecular modeling 13, 715-723.

High throughput in-silico screening against flexible protein

Virtual screening of chemical databases to targets of known three dimensional structure is developing into an increasingly reliable method for finding new lead candidates in drug development. Based on the stochastic tunneling method (STUN) we have developed FlexScreen, a novel strategy for high-throughput in-silico screening of large ligand databases. Each ligand of the database is docked against the receptor using an all-atom representation of both ligand and receptor. In the docking process both ligand and receptor can change their conformation. The ligands with the best evaluated affinity are selected as lead candidates for drug development. Using the thymidine kinase inhibitors as a prototypical example we documented the shortcomings of rigid receptor screens in a realistic system. We demonstrate a gain in both overall binding energy and overall rank of the known substrates when two screens with a rigid and flexible (up to 15 sidechain dihedral angles) receptor are compared. We note that the STUN suffers only a comparatively small loss of efficiency when an increasing number of receptor degrees of freedom is considered. FlexScreen thus offers a viable compromise between docking flexibility and computational efficiency to perform fully automated database screens on hundreds of thousands of ligands.

Mutations as Trapdoors: The Rop-dimer with two Competing Native Conformations

Protein are the molecular workhorses in biological systems. The molecular basin for their function are conformational transitions. Structure-based models, based on the funneled nature of the energy landscape, are a computationally track-able way to investigate such transitions. Here we present work on the Rop-dimer, a RNA-binding protein, which is part of the ColE1 plasmid replication process. In a series of mutational experiments, the hydrophobic interface between Rop's two monomers was mutated. This resulted in increased folding/unfolding rates of up to four orders of magnitude and changed RNA-binding ability. The symmetric nature of these mutations suggests that the native state is on the edge of structural heterogeneity between two competing native conformations, a parallel (P) and an anti-parallel one (AP). Our simulations on a dual-funneled energy landscape show, that the P conformation is kinetically more accessible than AP. We suggest, that the mutations trigger a trapdoor in Rop's energy landscape and undo the bias towards the functional AP state. The resulting competition between P and AP leads to the experimentally observed strong changes in kinetics.

A coarse-grained model for the nucleosome

In eukaryotic cells, DNA is packed with histone proteins into chromatin. Because many biological processes require a free DNA as a substrate, the chromatin structure needs to transit between tight and loose states. To investigate this long timescale dynamical process, we performed coarse-grained (CG) molecular dynamics on the nucleosome, the basic unit of chromatin.
In this coarse-grained model [1], protein residues and DNA nucleotides are represented as beads, interacting through harmonic (for neighboring) or Morse (for nonbonded) potentials. Force field parameters were estimated by Boltzmann inversion of the corresponding radial distribution functions obtained from a 5 ns all-atom molecular dynamics (MD) simulation, and were refined to produce agreement with the all-atom MD simulation. This self-consistent multiscale approach yields a coarse-grained model that is capable of reproducing equilibrium structural properties calculated from a 50 ns all-atom molecular dynamics simulation. It is also capable of describing detachment events of nucleosomal DNA extremities from the nucleosome core on the microsecond timescale.

[1] Coarse-grained force field for the nucleosome from self-consistent multiscaling, Voltz K, Trylska J, Tozzini V, Kurkal-Siebert V, Langowski J, Smith J., J. Comput. Chem., Feb. 2008

Effects of confinement on protein folding

We consider folding and unfolding of a protein contained within a sphere of radius R. We use a coarse-garined geometry based model. We find that the folding time is essentially independent of R but the possible unfolded structures do depend on R. Several proteins placed within the sphere fold mostly independent of one another unless one introduces attractive interactions between them.

Insights into the Alzheimer's Aβ peptide aggregation behavior: Molecular Dynamics Approach

Association of monomeric Aβ peptides in Alzheimer's disease leads to the formation of amyloid fibrils that are insoluble, not susceptible to proteolysis and display specific properties. Oligomers of Aβ peptide have been indicated as a possible agent of Alzheimer's disease. However, information concerning their structural properties is limited. This has motivated our conformational study of Aβ{39} monomer and its oligomers by all atom Replica exchange molecular simulations. Our results illustrate that Aβ monomer in water does not have at room temperature a unique fold but adopts one of the several low energy conformations. No stable beta-structure is observed in monomers indicating that the sheets, observed in fibrillar Aβ are formed only in the process of oligomerization and aggregation. In our further study of Aβ oligomers we restrict the Aβ chains to an imaginary sphere, mimicking the crowded cellular environment, results indicate that as chain length increase beta sheet structure is observed.

Effect of temperature on the structural and hydrational properties of human islet amyloid polypeptide in water

Structural and hydrational properties of full-length human islet amyloid polypeptide 1-37 (hIAPP) were studied in relation to the properties of hydration water in a temperature range from 250 to 450 K by computer simulations. At all temperatures studied, hIAPP does not adopt a well-defined conformation. The α-helical content and number of intrapeptide H-bonds of hIAPP decrease with temperature. The distribution of residues showing dihedral angles characteristic of β-sheets and polyproline II structures along the peptide chain is close to random, whereas a clear trend towards cooperative “condensation” is seen for residues showing dihedral angles characteristic of α-helices. This cooperativity is suppressed by heating or by introducing an intramolecular disulfide bond between residues 2 and 7. Intrinsic volumetric properties of hIAPP were estimated by taking into account the difference in the volumetric properties of hydration and bulk water. The temperature dependence of the density of hydration water indicates that the effective hydrophobicity of the hIAPP surface is close to that of carbon-like surfaces. Similarly to the case of the amyloid β (1-42) peptide [1], the thermal expansion coefficient of hIAPP is found to be negative: upon heating, it continuously decreases from ~-3·10^-4 to ~-2·10^-3 K^-1. A spanning H-bonded network of hydration water, which covers hIAPP homogeneously at low temperatures, breaks via a quasi-2D percolation transition, whose midpoint is at about 320 K. Approximately at this temperature, the experimentally measured lag time of hIAPP aggregation drops in a drastic way [2]. We discuss the possible role of the temperature-induced percolation transition of hydration water on the conformational changes and aggregation propensity of peptides.

References
[1] I. Brovchenko, R. R. Burri, A. Krukau, A. Oleinikova, and R. Winter to be published.
[2] R. Kayed et al., J. Mol. Biol., 287 (1999) 781.

Structural Basis of Protein-RNA recognition

We have dissected 81 non-redundant protein-RNA complexes derived from the Protein Data Bank (PDB). The data set was divided into four major categories depending on the properties of RNA chain: 1. complexes including tRNA, 2. complexes involving ribosomal proteins, 3. complexes with duplex RNA, and 4. complexes involving single stranded RNA. Several physicochemical and geometric properties have been investigated to characterize these interfaces, and were used to compare with protein-DNA and protein-protein interfaces with the aim to understand the specificity of a protein surface to interact with RNA. We found that interfaces formed in complexes involving tRNA are largest in size and comparable to those found in protein-DNA interfaces; that single-stranded RNA-protein interfaces are smallest and are comparable with the size of the protein-protein interfaces. In general, the protein surface that is in contact with RNA bears a strong positive electrostatic potential reflected by their amino acid composition. Interfaces are less well packed and contribute less surface area to the interface due to its concave nature compared to protein-protein interfaces. Hydrogen bonds are more abundant in protein-RNA than protein-protein interfaces (one per 125 Ų of BSA against one per 190 Ų), and also the former type is slightly more hydrated than the later (13 water molecules per 1000 Ų of BSA compare to 10). In spite of having few similarities, protein-RNA interfaces differs from the protein-DNA interfaces. In the former sugar and base contributes much more whereas phosphate is the dominant contributor to the later. Moreover the 2'OH group plays a major role in stabilizing the protein-RNA interfaces by making intensive hydrogen bonds with the polar groups in proteins. The analysis of protein-RNA interfaces was used to develop a pair-potential to evaluate protein-RNA complexes based on a coarse-grained RNA and protein model. The potentials allow rapid docking minimization and systematic searches for predicting possible protein-RNA complex structures.

References
1.Bahadur RP, Zacharias M and Janin J. (2008). Nuc. Acids Res. (doi:10.1093/nar/gkn102)
2.Bahadur RP and Zacharias M. Cell. Mol. Life Sci. (2007) (doi:10.1007/s00018-007-7451-x)
3.Nadassy K, Wodak S and Janin J. (1999). Biochemistry 38, 1999-2017

Prediction of Protein-Protein Binding Affinity

Protein-protein complexes are involved in most processes in cell and therefore an important target in pharmaceutical research. The activity of processes coupled to protein-protein complexes can be influenced by modifying the binding behaviour of the associated complexes. This can be done either by re-design of protein interaction surfaces or by identifying small drug-like molecule inhibitors. Here we developed and implemented a fast and reliable method (CC/PBSA) for the prediction of the change in binding affinity of protein-protein complexes upon mutation. The energy function of CC/PBSA is based on gas phase energies and solvation free energies. The protein flexibility is taken into account by generating random conformations based on geometrical constraints only applying the CONCOORD program. CC/PBSA was developed and validated on a test set of more than 300 mutants of various protein-protein complexes, including non-alanine, non-conservative and multiple mutations (correlation of 0.8, standard deviation of approximately 1 kcal/mol). We show results on a full mutational scan of the insulin dimerization interface.

Correlation Effects in Protein-Protein Recognition

Correlation effects in the hydrophobicity distribution in the interface of a protein-protein complex are investigated within an idealised coarse-grained model for the recognition of two rigid proteins. To this end, a two-stage approach is adopted where the biomolecules are first optimised with respect to each other and afterwards their selectivity is tested in the presence of other molecules. Correlations lead to different optimum characteristic lengths of the hydrophobic and polar patches for the design of the two biomolecules on the one hand and their selectivity in the presence of other molecules on the other hand.

Prediction of Twist of Amyloid Fibrils Using Molecular Dynamics

Many proteins and peptides form amyloid fibrils. These long, helically symmetric protein aggregates can be highly ordered but are not normally amenable to X-ray crystallography or to solution NMR. Therefore although amyloid fibrils of the same sequence can display substantial variation in gross morphological features such as twist as well as branching, persistence length and cross-sectional area the atomic-level origins of this variation remain obscure. In order to examine the origins of the diversity in fibrillar twist we use the weighted-histogram analysis method (WHAM) on atomistic molecular dynamics trajectories to probe the free energy with respect to twist of two small model polyalanine fibrils having different subunit-packing symmetries within the overall helical symmetry which defines amyloid.

Role of H3/H4 Histone N-termini in Chromatin Compaction at Mononucleosome Level

DNA assumes a highly compact structure inside a cell nucleus. At the mononucleosome level, histone N-termini are believed to play a key role in DNA packaging since they have the sites of many post-translational modifications. An explicit solvent molecular dynamics simulation is performed without H3/H4 histone tails and then compared to the full nucleosome data to investigate their specific roles. The effect of tail deletion on DNA flexibility is studied. The results of this study may help us in finding ways to fine tune the dynamical properties of nucleosome in a suitable context.

Automatic sequential NOESY assignment and NMR structure improvement by X-ray

We are developing AUREMOL (www.auremol.de), which goal is the reliable and automatic structure determination of biological macro molecules such as proteins from NMR data (1). For a fully automatic sequential NOESY assignment the tool ASSIGN has been developed. The required input consists of a homologous structure for a NOESY spectrum simulation and the experimental NOESY spectrum. ASSIGN fits the simulated NOE signals to the experimental spectrum. The fit quality given by a pseudo energy function depends on the line shapes and volumes of the signals. The assignment is varied by moving or swapping spin system assignments using a Monte Carlo approach. A threshold accepting algorithm (TA) is employed to find the minimum of the pseudo energy function.
Another newly developed tool of AUREMOL is ISIC (2) and aims at NMR structure improvement by using additional information such as a homologous X-ray structure. In the example shown we are improving the medium well defined NMR-structure of the Ras-binding domain of Byr2 (Byr2-RBD) by the equally well defined X-ray structure of Byr2-RBD. It is important to combine the information from the two sources in such a way that ensures that the newly calculated structure explains the experimental NMR data better than the original structure. We obtained an improved Byr2-RBD structure which not only looks better, but also deliver better values with AUREMOL RFAC (3) and well known structure verification tools.

(1) Gronwald, W. and Kalbitzer, H. R., Automated Structure Determination of Proteins by NMR Spectroscopy. 2004, Progr. NMR Spectr. 44, 33-96
(2) Brunner K, Gronwald W, Trenner JM, Neidig KP, Kalbitzer HR, A General Method for the Unbiased Improvement of Solution NMR Structures by the Use of Related X-Ray Data, the AUREMOL-ISIC Algorithm. BMC Struct Biol. 2006 Jun 26;6(1):14
(3) Gronwald W, Kirchhofer R, Gorler A, Kremer W, Ganslmeier B, Neidig KP, Kalbitzer HR, RFAC, a program for automated NMR R-factor estimation. J Biomol NMR. 2000 Jun;17(2):137-51

Exploring the first steps of AβL16-22 protofibril disassembly by N-methylated inhibitors

Alzheimer's disease is characterized by the self-assembly of the Aβ(1-40)/(1-42) peptides. The design of efficient inhibitors is particularly challenging because the structures of the toxic Aβ species are transient in character and the mode of action of current inhibitors on Aβ oligomers are unknown [1] .
We know that N-methylated Aβ16-22 peptides (mAβ16-22) effectively inhibit fibrillogenesis and disassemble existing fibrils in vitro [2]. In this work we report molecular dynamics (MD) [3] and replica exchange molecular dynamics (REMD) simulations using the coarse-grained OPEP [4] force field on a preformed protofibril of six Aβ16-22 peptides with either four copies of Aβ16-22 or four copies of mAβ16-22.
While MD trajectories of 100 ns do not reveal any significant differences between the two systems, REMD simulations help understand the first steps of Aβ16-22 protofibril disassembly by N-methylated inhibitors.

[1] D.M. Walsh, I. Klyubin, J. V. Fadeeva, W. K. Cullen, R. Anwyl, M. S. Wolfe, M. J. Rowan and D. J. Selkoe. Naturally secreted oligomers of amyloid beta protein potently inhibit hippocampal long-term potentiation in vivo. Nature 2002 (416): 535-539
[2] D. J. Gordon, K. L. Sciarretta and S. C. Meredith. Inhibition of beta-amyloid(40) fibrillogenesis and disassembly of beta-amyloid(40) fibrils by short beta-amyloid congeners containing N-methyl amino acids at alternate residues. Biochemistry 2001 (40): 8237-8245
[3] P. Derreumaux and N. Mousseau. Coarse-grained protein molecular dynamics simulations. J Chem Phys 2007 (126).
[4] J. Maupetit, P.Tuffery, and P. Derreumaux. A coarse-grained protein force field for folding and structure prediction. Proteins 2007 69(2):394-408.

Simulation of the outer membrane protein X in a lipid bilayer and in a micelle

The outer membrane protein OmpX from Escherichia coli has been simulated embedded in a phospholipid bilayer and as a protein-micelle aggregate. The resulting simulation trajectories have been analysed in terms of structural and dynamical properties of the membrane protein. In agreement with experimental observations it has been found that the β-barrel region, embedded in the lipophilic phase, is very stable, whereas the extracellular, protruding β-sheet, that plays an important role in cell adhesion and invasion of gram negative bacteria, shows large structural fluctuation. Additionally, it has been investigated whether water permeates the core of the β-barrel protein and it has been found that a very stable salt-bridge and hydrogen-bond network exist in that barrel and that a water flux is therefore unlikely. No great difference in protein stability and dynamics between the bilayer and the micellar systems have been observed.

OPERA: An OPtimized coarsed-grained Energy model for RnA

RNAs have many cellular functions ranging from transcription to catalysis. The gap between sequences and 3D structures is increasing, and knowledge of RNA’s dynamics and thermodynamics at an atomic level is missing. In principle, all-atom molecular dynamics (MD) and replica exchange molecular dynamics (REMD) simulations in explicit solvent can investigate these issues. However with current computer facilities, these simulations have been limited to small RNAs (1).
To move to large RNAs, we can resort to coarse-grained models. So far, two models based on the Go potential have been developed for nucleic acids (2-3). In this study we present OPERA, a generic coarse-grained model for RNA, built on structural analysis of the NDB and an implicit solvent reduced representation. In our model, the pyrimidines are represented by 6 beads, the pyrimidines by 7 beads and the analytical energy function is very similar to the one used for proteins (4-5). We report MD and REMD simulations on two RNAs of 22 nucleotides using a set of non-optimized OPERA parameters. Current results suggest that further optimization of the OPERA force field should open the door to a relevant model for studying large RNA’s such as riboswitches.

1. Villa, A., Widjajakusuma, E., Stock, G., Molecular dynamics simulation of the structure, dynamics, and thermostability of the RNA hairpins uCACGg and cUUCGg. J. Chem. Phys 112, 134-142 (2008).
2. Knotts, T. A., Rathoe, N., Schwartz, D. C., de Pablo, J. J., A coarse grain model for DNA. J. Chem. Phys. 126 (2007).
3. Hyeon, C. Thirumalai, D., Chemical theory and computation special feature: mechanical unfolding of RNA hairpins. Proc Nucl. Acad. Sc. USA. 102, 6789-6794 (2007).
4. Derreumaux, P., Mousseau, N., Coarse grained protein molecular dynamics similations. J. Chem. Phys. 126, 025101 (2005).
5. Maupetit, J., Tuffery, P., Derreumaux, P., A coarse grained protein force field for folding and structure prediciton. Proteins 69, 394-408 (2008).

Understanding of High Pathogenicity of the Avian Influenza Virus H5N1: Why H5 is Better Cleaved by Furin

To reveal the origin of high pathogenicity of an emerging avian influenza H5N1 due to the –RRRKK– insertion at the cleavage loop of the hemagglutinin H5, was studied using the molecular dynamics technique, in comparison with those of the non-inserted H5 and H3 bound to furin active site. The cleavage loop of the highly pathogenic H5 was found to bind strongly to the furin cavity, serving as a conformation suitable for the proteolytic reaction. Experimentally, the –RRRKK– insertion was also found to increase in cleavage of hemagglutinin by furin. The simulated data provide a clear answer to the question why inserted H5 is better cleaved by furin than the other subtypes, explaining the high pathogenicity of avian influenza H5N1.

Bound Water as a Tool to Differentiate Between Soluble Amyloid Oligomers, Amyloid Fibrils and Amyloid Plaques

Hydration shells of normal proteins display regions of highly structured water as well as patches of less structured, bulk-like water. Recent studies suggest that isomers with larger surface densities of patches of bulk-like water have an increased propensity to aggregate. These aggregates are toxic to the cellular environment. Hence, their early detection is of paramount medical importance. We show that various morphological states of association of such isomers can be differentiated from the normal protein background based on the characteristic distribution profiles of water structured at interfaces and magnetic resonance (MR) signals of this water. Our results indicate that single units and compact aggregates that contain no water between constituents units induce MR signals that are shifted from the value of the normal protein background towards values corresponding to bulk water, in the hyper-intensity domain (bright spots). In contrast, large plaques that cage significant amounts of water between constituent units are likely to generate MR responses in the hypo-intensity domain (dark spots), typical for strongly correlated water. The present study shed light on a current controversy, namely that amyloids can display both dark and bright spots when compared to the normal, gray background tissue on MR images. In addition, our findings suggest that the bright spots are more likely to correspond to amyloids in their early stage of development. The results help to better understand various biophysical mechanisms that set the MR signal of water surrounding amyloidogenic proteins and their model aggregates.

Stabilizing regions in membrane proteins

Around one third of a typical genome consists of membrane proteins. Misfolding of membrane proteins can often be linked to diseases, so that it is of great importance to understand, which residues and interactions are crucial for the stability. We developed a coarse-grained model to predict stabilizing regions in membrane proteins. We compare the model to experimental data from Single molecule force spectroscopy (SMFS) and literature to evaluate the effects of mutations on function and stability of five membrane proteins (bacteriorhodopsin, halorhodopsin, rhodopsin, an Na+/H+ antiporter, and aquaporin). The aim of this study is to describe all these data in an unified context, the interaction energies of amino acids in a coarse grained model to gain a deeper insight into membrane proteins.

Impact of Loop Statistics on the Thermodynamics of RNA Folding

Loops are abundant in native RNA structures and proliferate close to the unfolding transition. By including a statistical weight ~ l^-c for loops of length l in the recursion relation for the partition function, we show that the calculated heat capacity depends sensitively on the presence and value of the exponent c, even of short t-RNA. For homo-RNA we analytically calculate the critical temperature and critical exponents which exhibit a non-universal dependence on c.

Structure of Amphipathic Peptides in Vacuum, in Bulk Water, or at an Air/Water Interface Using Replica Exchange Molecular Dynamics

Biomineralization occurs in a wide variety of organisms in nature, ranging from the formation of the shell of a mollusk to that of human bone. This process is accomplished by controlling of the spatial distribution of chemical functionality, and in nature, it is often governed by bottom-up construction of supramolecular amino-acid based assemblies.
In this study, replica exchange molecular dynamics (REMD) simulation techniques have been used to design and to characterize the amphipathic peptides that assemble at an air/water interface. The behavior of these peptides was investigated in vacuum, in bulk water, and at an air/water interface in order to investigate the environment effect on the structures of peptides. It was found that these amphipathic peptides adopted different secondary structures in different environments. Furthermore, the effect of association of these monomers into higher-level structures on the secondary structure preference of monomers was also investigated. Based on a combination of the available experiment data and current simulation studies, we propose a model for the self-assembly of these peptides at air/water interface. The analysis of this behavior will provide an insight for the enlightenment of the controlling process of the nucleation of inorganic crystals.

Protein-Protein Interaction Prediction

BACKGROUND: Docking algorithms are developed to solve the three-dimensional complex structure of interacting proteins. Here we use a docking approach to investigate whether we can discriminate between a pair of proteins that is known to interact (native complex) and two non-interacting proteins (false complex). For this we are searching for scores that can support the discrimination. The aim is to identify false positives that were obtained by other methods such as yeast2hybrid assays.
METHOD: Based on intermolecular energies and amino acid based pair-potentials we calculate score distributions for native and false complexes. The difference in the distributions makes it then possible to assign a hypothetical complex.
RESULTS: For several examples we could already show that the intermolecular energies of the native complexes are considerably lower as for the false complexes. Furthermore the score distributions from pair-potentials are different for the two groups.

Statics of the ribosomal exit tunnel: implications for co-translational processes and antibiotics binding

Since the structure determination of the ribosome in atomic detail in 2000 [1] much has been learnt about the structural basis for protein synthesis. However, the functional role of the ribosomal exit tunnel in co-translational processes such as peptide elongation regulation and protein folding still remains elusive.
Here, we present an analysis of the static properties (i.e., the flexibility and rigidity) of the ribosomal exit tunnel using concepts grounded in rigidity theory [2]. For this, a new topological network representation of RNA structures had to be developed that allows analyzing RNA flexibility/rigidity based on constraint counting [3]. Applied to the large ribosomal subunit, constraint counting explains previously identified folding zones within the tunnel region and provides insights in atomic detail into the stability characteristics of the tunnel gate keeper protein L22 and known antibiotics binding crevices.

References:
[1] N. Ban, P. Nissen, J. Hansen, P.B. Morre, T.A. Steitz, A.B. Smith, C.D. Brown, Science, 289, 905 (2000).
[2] D. J. Jacobs, A. J. Rader, L. A. Kuhn, M. F. Thorpe, Proteins, 2001, 44, 150-165.
[3] S. Fulle, H. Gohlke, Biophy. J., 2008, DOI:10.1529/biophysj.107.113415.

Ab initio molecular dynamics of the Zn-binding site of the Alzheimer's amyloid beta-peptide

The aggregation of the peptide amyloid-beta (Abeta) into fibrils is considered to be a key event in Alzheimer disease. Zn(II) ion binds the N-terminal segment of Abeta peptide (region 1-16) and influences its aggregation behaviour. Many experimental evidences revealed that Zn-binding involves the three histidine residues (6, 13 and 14). As for the fourth ligand, instead, different candidates have been proposed: Asp 1, Arg 5, Ser 8, Tyr 10 and Glu 11.
We performed ab initio molecular dynamics simulations (Car-Parrinello method) at 300 K of Zn(II)-Abeta(1-16) system and we found that:

• The three His are always bonded to Zn.
• The binding of Zn by carboxyl groups (Asp 1 and Glu 11) is stable.
• Hydroxyl groups (Ser 8 and Tyr 10) bind Zn when assisted by hydrogen bonds; as far as these hydrogen bond networks break, hydroxyl groups are replaced by nearby peptide carbonyl groups.

Folding and aggregation features of proteins

In addition to "normal," native protein structure, some proteins can also form alternative, misfolded structures. During the past years, it has been shown that some diseases are connected with protein misfolding and the formation of insoluble aggregates called amyloid plaques. For some proteins which are capable to form amyloid structures, those regions which are important for amyloid formation are already experimentally outlined (they are called amyloidogenic regions). In these proteins, we predicted those residues that are important for "normal" folding (that is, which are involved into the folding nucleus of the native structure) and compared them with amyloidogenic ones. The average of the predicted Φ-values (which reflect the degree of involvement of the amino acid residue into the folding nucleus) over 12 amyloidogenic regions (of 7 globular proteins in which amyloidogenic regions are now localized experimentally) is significantly greater than the average Φ-value averaged over residues outside amyloidogenic regions. This demonstrates that amino acid residues in amyloidogenic regions in average are more included into folding nucleus than amino acid residues from non-amyloidogenic regions. In total, 8 of 12 amyloidogenic fragments are located in the folding nuclei. On the other hand, comparison of experimentally outlined folding nuclei with predicted amyloidogenic regions also demonstrates that the average Φ-value (in this case, experimental one) is also significantly greater in the predicted amyloidogenic regions. This is an indication that amyloidogenic regions are typically incorporated into the native structure early during its formation (not later than at the rate-limiting step of a "normal" folding process).
This work was supported by the Russian Academy of Sciences (“Molecular and Cell Biology” program), by the Russian Foundation for Basic Research (08-04-00561), by the INTAS grant (7747), “Russian Science Support Foundation”.

Coarse-Grained Simulations of Protein Adsorption on Solid Surfaces

The adsorption of individual proteins on solids and soft materials plays a vital role in biotechnical and biomedical applications, for example for the biocompatibility of implant material or in dental health care. Not only the properties of the sorbent surface can be changed, but also the proteins might undergo conformational changes during adsorption. To investigate collective processes such as the formation of biofilms on tailored substrates in molecular detail, but still reaching appropriate length and time scales (microseconds), coarse-grained molecular dynamics simulations were applied here. As a model system, the adsorption of lysozyme and human serum albumin to a negatively charged solid surface were studied at various ion concentrations. Adsorption rates of the two proteins, protein diffusion on the surface, and the orientation of the proteins on the surface were analyzed.

Fast Electrostatics Computation for Molecular Systems with Constraints.

Computational cost of molecular simulations increases with the system size. A number of simplification methods have been introduced in the recent years to accelerate the simulations. The most computationally expensive part of any molecular simulation is long-range interactions. Direct computation for the system with N charged particles requires O(N²) operations, with fast algorithms this cost can be reduced to O(NlogN) for the mesh Ewald[1] and tree[2] methods or O(N) for the Fast Multipole Method (FMM)[3]. A supplementary approach to reduce the computational cost of large systems is to use its simplified representation. Coarse-grained models do not provide the full atomistic description of simulated molecular systems and therefore do not always describe correctly certain phenomena. An alternative approach is to use the fully atomistic description of the simulated system, but reduce the number of degrees of freedom by applying constraints. This approach allows to use larger time steps with Molecular Dynamics, but unfortunately does not reduce the computational cost of long-range interactions for a single step, since the total number of charged particles remains the same.
We introduce a new algorithm that uses the atomistic description of the molecular system with a number a of active degrees of freedom and unlike the previous algorithms reduces the computational cost for a single time step to O(alogN). With a small number of active degrees of freedom a with respect to the number of particles N our algorithm is advantageous to standard approaches. The molecular system is described in internal coordinates[4,5] and an FMM-like algorithm is used to accelerate computations of the long-range interactions.

1. T. Darden, D. York, and L. Pedersen. Particle mesh Ewald: An N⋅ log (N) method for Ewald sums in large systems. 1993. The Journal of Chemical Physics, 98:10089-10092.
2. J. Barnes, and P. Hut. A hierarchical O (N log N) force-calculation algorithm. 1986. Nature, 324: 446-449.
3. L. Greengard, and V. Rokhlin. A new version of the Fast Multipole Method for the Laplace equation in three dimensions. 1997. Acta Numerica.
4. S. Redon, N. Galoppo, and M.C. Lin. Adaptive dynamics of articulated bodies. 2005. Proceedings of ACM SIGGRAPH, 24:936-945.
5. R. Rossi, M. Isorce, S. Morin, J. Flocard, K. Arumugam, S. Crouzy, M. Vivaudou, and S. Redon. Adaptive torsion-angle quasi-statics: a general simulation method with applications to protein structure analysis and design. 2007. Bioinformatics 23:408-417

Towards understanding the early events in the conformational transformation of amyloid beta peptides

Recently experimental evidence has implicated the toxicity of soluble oligomers of amyloid beta peptides in Alzheimer's disease. Given the metastable nature of these oligomers, it is hard to obtain experimental data for the early events taking place during the oligomerization of amyloid beta peptides. Computational simulation methods are, hence, needed to provide atomistic details of the early events in amyloid beta oligomerization and there is already a broad literature. It is experimentally known that oligomerization is accompanied by conformational transformation of amyloid beta peptides from mainly alpha to mainly random coil or beta sheets. The aim of this study is to analyze and compare the structure and dynamics of water near the peptide surface (hydration water) during the conformational transition of amyloid-beta 42 (ab42) and amyloid-beta 40 (ab40) peptides. Therefore, MD simulations of 100ns with explicit representation of solvent were performed for individual amyloid beta monomers. Analysis was based on the radial distribution function (RDF) for the peptide surface, for individual residues and for respective secondary structure elements. In both cases, initial results suggest that, in accord with the literature, RDF reveals the presence of two solvation shells around polar residues. Variations in RDF in the first solvation shell were found to be consistent with the physiochemical properties of the amino acids and were independent of the secondary structure element. Further investigations, such as dimer formation and analysis of the orientation of water molecules near peptide surfaces, are necessary to account for the role played by surrounding water in the assembly of such unstructured peptides.

Protein Interactions with the Environement

During functioning, proteins interact with and influence their environment. The analysis of interactions of proteins with their environment is of crucial importance for an understanding of protein function. Here, we focus on two aspects of protein interactions: The first topic of interest is the analysis of the protein-protein complexation behavior. In this context, we analyze the complexation and the impact of mutations on the interaction of the bacterial Ribonuclease Barnase with its natural inhibitor Barstar. In this system, our specific interest is the formation of intermediate states along the reaction coordinate as well as the driving complexation force in the system. In agreement with experimental data, we found that the complexation process is mainly driven by electrostatic interactions, dramatically reducing the conformational space of the approaching complexation partners. This results in the formation of stable encounter complexes from which the final complex structure is promoted.
The second topic is the modelling of proteins interacting with surfaces in the framework of the ProSurf EU project. In this context our focus is the simulation of protein adsorption on gold surfaces in water. As a first step we evaluated a classical set of parameters derived by ab-initio calculations from our cooperation partners. In obtaining mean force profiles for all 20 aminoacids by constrained simulations and comparing them to experimental results, we found reasonable agreement between experimental and computational results. Additionally these simulations allow us to retrieve information about the aminoacid orientation towards the surface during different stages of complexation and the free energy difference in adsorption for each aminoacid. In our simulations a clear barrier attributable to the final water layer could be observed.

Identification of differential protein expression in response to the application of bioregulators that enhance plant productivity and quality

The enhancement of plant growth and productivity by the use of chemicals is a common practice in agriculture but rather little is known about the effect these bioregulators have on plants on a molecular level. The presented work is part of an ERA-NET project with the aim to identify genes, proteins and metabolites, differentially expressed under conditions of stress while treated with certain compounds. Arabidopsis thaliana plants are grown on soil and in hydroponic cultures and are subjected to salt, cold and drought stress. The proteome profile of the plants is analysed via 2D electrophoresis and mass spectrometry. For the identification of differentially regulated proteins in response to the bioregulator treatment under stress conditions, the DIGE (difference-in-gel-electrophoresis) system will be used. The biomarkers identified universally for the different stress conditions will be used to establish a cell based assay for the screening of potential new bioregulator compounds.

A Hamiltonian Replica Exchange Molecular Dynamics Using Soft-Core Interactions to Enhance GTP and CYP2D6 Binding Site Conformational Space Sampling

We present a novel Hamiltonian replica exchange molecular dynamics (H-REMD) scheme that uses soft-core interactions between those parts of the system that contribute most to high energy barriers [1]. The advantage of this approach over other REMD schemes is the possibility to use a relatively small number of replicas with locally larger differences between the individual Hamiltonians. Because soft-core potentials are almost the same as regular ones at longer distances, most of the interactions between atoms of perturbed parts will only be slightly changed. Rather, the strong repulsion between atoms that are close in space, which is in most cases resulting in high energy barriers, is weakened within higher replicas of our proposed scheme.
We have tested the new protocol on the GTP and 8-Br-GTP molecules, which are known to have high energy barriers between the anti and syn conformation of the base with respect to the sugar moiety. During two 25 ns MD simulations of both systems the transition from the more stable to the less stable (but still experimentally observed) conformation is not seen at all. Also T-REMD over 50 replicas for 1ns did not show any transition at room temperature. On the other hand, more than 20 of such transitions are observed in our new H-REMD scheme using 6 replicas (at 3 different Hamiltonians) during 6.8 ns per replica for GTP and 12 replicas (at 6 different Hamiltonians) during 7.7 ns per replica for 8-Br-GTP. The large increase in sampling efficiency was obtained from an optimized H-REMD scheme involving soft-core potentials, with multiple simulations using the same level of softness. The optimization of the scheme was performed by fast mimicking [2].
H-REMD also revealed the relative conformational populations of the most flexible parts within the cytochrome P450 2D6 isoenzyme in the presence of five different substrates, thus showing different extents of induced fit effects.

References:
[1] Hritz, J.; Oostenbrink, C., Hamiltonian Replica Exchange Molecular Dynamics Using Soft-Core Interactions. (accepted to J. Chem. Phys. 2008)
[2] Hritz, J.; Oostenbrink, C., Optimization of Replica Exchange Molecular Dynamics by Fast Mimicking. J. Chem. Phys. 2007, 127, 204104.

TollML: a Database of Toll-like Receptor Structural Motifs

During recent years Toll-like receptors (TLRs) have spearheaded a tremendous research interest and the amount of sequenced relevant proteins grows exponentially. A critical step towards the successful TLR structure modeling is to generate the LRR motif aided sequence alignment between the target sequence and the templates. However, because of the irregularity of LRR motifs in TLRs, most TLRs have no LRR annotaions in current databases, and in those TLRs with LRR partitions, the indicated repeat number and the boundaries of LRRs are quite different among databases. In order to provide a useful platform for structure prediction and analysis of these sequences, we developed TollML, an XML based database specialized for TLR structural motifs.
Its original TLR sequences were extracted from NCBI’s protein database. The LRR motifs as well as transmembrane and TIR motifs for all known TLR sequences are identified and annotated manually and can then be used for the prediction of protein structures via alignment and threading. The resulting database was used for the structure prediction of TLR7, 8 and 9.

Understanding the enzymatic breakdown of the starch granule

Transitory starch in chloroplasts is basically composed of amylose and amylopectin. Its biosynthesis, however, is controlled in such a way as to form a complex insoluble granule. Thus, the first reactions committed to the degradation pathway involve enzymes acting on the interface of two physically different phases. Furthermore amorphous and crystalline zones are exposed to the surface during degradation which are targets for different enzymes. A glucan water dikinase and hydrolytic enzymes act synergistically to release malto-oligosaccharides into the bulk phase. We consider kinetic effects associated with adsorption/desorption of these enzymes to the starch surface, as well as with their partitioning to different regions in the outer boundary layer. Our mathematical model aims at integrating the interdependency of interfacial phenomena and enzymatic catalysis to elucidate the role of the initial degradation steps in controlling the rate of starch breakdown.

Engrailed homeodomain folds overnight by 100 processors

It has been shown that the native structure of a small protein can be efficiently found as the global minimum of a certain all-atom forcefield. In the present study we use this approach to simulate folding of Engrailed homeodomain (1ENH) and its 16-59 fragment (2P81) containing the helix-turn-helix motif. The search procedure is based on the Monte Carlo simulated annealing combined with the evolutionary algorithm. The ways of increasing the efficiency of this method are discussed.

Single-molecule FRET study of an RNA folding: How a methyl group modification changes the energy landscape

RNA is a versatile biopolymer that is involved in a variety of key biological functions, including storage and transport of information, structural scaffolding, and gene expression and regulation. Like proteins, RNA molecules fold into compact three-dimensional structures, and their self-assembly and dynamics is described by transitions within a highly complex energy landscape containing a vast number of different conformations. RNA forms secondary structural motifs that assemble into three-dimensional structures by tertiary interactions. Nucleotide modifications in the tRNA structural core are believed to stabilize its functional conformation characterized by the cloverleaf secondary structure. Here we have used single-molecule FRET spectroscopy to detect multiple conformations in a human mitochondrial tRNALys, and to show that a post-transcriptional modification consisting of a single methyl group on N1 of adenosine 9 (m1A9) strongly affect the equilibrium between a non-functional, extended hairpin structure and the functional cloverleaf form. In order to decompose the structural contribution and stabilization due to cations for each of conformations, titration with Mg2+ ions was performed and a thermodynamic model was developed. It was found that the positive charge of a single methyl group effects mostly on the extended hairpin conformation stabilizing the Mg-free state but interfering with Mg2+ binding. For the cloverleaf conformation, the stabilizing effect of the positive charge is much smaller, suggesting that the base of nucleotide m1A9 is exposed to the solvent.

Impact of induced fit on ligand scoring and a strategy of identifying a minimal set of flexible residues

Although conformational changes in receptor upon ligand binding are a very common phenomenon, incorporating protein flexibility in a docking procedure encounters significant computational problems. A possible solution is inclusion side-chain flexibility for only limited number of residues in the binding pocket, which can improve notably accuracy of docking procedure without considerable increase of computational costs. However, investigation of this approach is often limited to specifically chosen receptors and mostly focused on the impact of receptor flexibility on docking accuracy, whereas ligand scoring, the real weakness of the present-day docking methodology, is treated only peripherically.
In the present study we investigate enrichment rates of rigid-, soft-, and flexible- receptor models using 12 diverse proteins with receptor-specific ligand libraries containing up to 13000 molecules, comprising known ligands and decoys with similar physical properties but distinct topology. We also present and test a straightforward protocol for the choice of the flexible residues, which is based on the ability of the receptor structure to accommodate the set of known ligands. This strategy is an unbiased approach to identify the most important residues likely to be relevant for induced fit effects, that allowed us to improve EF1 values by ~35% on average with respect to rigid-docking.

Contributions to the hydration free energies of amino acids

Molecular solvation is a fundamental factor in biological processes, such as protein folding, receptor binding or enzymatic reactions. Currently, estimates of the hydrophobicity of amino acids are often derived from solvation (or transfer) free energies of side chain analogs. Such an approach implicitly assumes that contributions from the backbone and the side chain to the free energy of solvation are additive. However, it is well known that, in particular for polar amino acids, the properties of side chain analogs and amino acids can deviate significantly. Based on the relative hydration free energies of the amino acid pairs Ala-Ser, Val-Thr, Phe-Tyr, Val-Ala, Thr-Ser, Phe-Ala, and Tyr-Ser determined from molecular dynamics simulations, we quantitatively trace the molecular origin of these deviations to two effects, solvent exclusion and self-solvation. Solvent exclusion accounts for the reduction in solute-solvent interactions as one part of the solute occludes other parts of the solute, e.g., the presence of the backbone lowers the degree of direct interaction possible between the side chain and surrounding water. While solvent exclusion applies to polar and apolar amino acids alike, self-solvation is specific to polar amino acids and results from strong, directed intramolecular interactions between the polar functional groups of the side chains and polar moieties in the backbone, often through hydrogen bond formation. Thus, the contribution of self solvation to the solvation free energy is strongly conformation- and environment-dependent, and, therefore, the correct treatment thereof poses a challenge to applications involving solvation processes. Implications for the utility of hydrophobicity scales and connections to implicit solvent models are briefly discussed.

Conservation analysis of functional important residues of the oxygen evolving mechanism located in the D1 subunit of Photosystem II

The oxygenic photosynthesis is of central importance for the evolution of life on earth. Oxygen is produced by water splitting in the oxygen evolving center located at a manganese cluster in photosystem II. The function and binding of the manganese cluster is mediated by several residues of the D1 subunit. Possible candidates are for example TyrD1-161 (forming the tyrosine radical during the reaction) and HisD1-190 (serving as a proton acceptor for TyrD1-161), AspD1-170, GluD1-189, HisD1-332, AspD1-342, and the C-Terminus of AlaD1-344 (all coordinating the manganese cluster).
The here made study aims to analyze possible conservation of these residues. Conservation of a residue could mean, that this residue could also have a function in oxygen production, since this function is evolutionary conserved in photosystem II. Recently, we build profile Hidden Markov Models for the type II reaction center subunits of bacteria and plants to analyze the conservation of cofactor binding and proton transfer pathways in all RC proteins. The advantage of such a profile Hidden Markov Model is, that it uses position-specific parameters for the conservation of an aminoacid. Based on the created profile Hidden Markov Model we aligned 227 D1 sequences to analyze possible conservation of residues being involved in the oxygen evolving process.
Our data reveals, that residues identified by experiments to be crucial for the function such as D1-190 are strictly conserved, whereas residues being proposed to bind the manganese ions are not conserved for to the same extent. Thus, the oxygen evolving mechanism seems to be more flexible considering residues involved in binding of the manganese cluster compared to the functional crucial residues, for which not even exchange to aminoacids with a similar character is possible.

Viral membrane proteins: Flexibility and Assembly

Membrane bound and pore forming viral proteins like M2 from Influenza A, Vpu from HIV-1 or 3a and 8a from SARS-CoV show in their monomeric form high flexibility and adaptability in different lipid bilayer environments [1]. Their conformational space under these conditions has been studied by ample molecular dynamic simulations.
The understanding of the abilities of the monomeric units is used to further explore the energy landscapes of the molecular assembly of multiple monomers. The newly developed protocol screens the full high dimensional search space leading to highly reliable pore models. The quality of the method is evaluated by comparing the model of M2 from influenza A with available NMR spectroscopic data. The quality of the model is so good that it can fill the gap that arises from the inability, due to the fact that membrane proteins are barley crystallizable, to get atomistic x-ray data easily. With this method we are able to present pore models for Vpu and the newly discovered 3a, for which no structural data is available yet, on an atomic level. After minor refinement e.g. with short molecular dynamics simulations they are suitable for the use in drug screening. Furthermore the evaluation of the energy landscapes allows drawing conclusions about the gating mechanism of the pores.

Hen egg white lysozyme adsorption on a mica surface – a fully atomistic molecular dynamics study

The interactions between proteins and solid surfaces are essential for a number of applications, such as functionalising biomaterials and for medical implants. The understanding of fundamental forces and processes involved in protein adsorption has a great importance in the construction of new, biocompatible materials. Our recent effort has been focussed on adsorption processes and protein dynamics on the surface, including protein cluster formation and cluster diffusion [1]. Hen egg white lysozyme adsorption on a mica surface is an excellent model system for these investigations. Despite recent theoretical and numerical investigations, including Brownian Dynamics (BD) and Monte Carlo (MC) simulations [2-4], many atomistic details of lysozyme adsorption, cluster formation and protein/cluster(s) diffusion on the mica surface remains unknown.
Here, for the first time, we present results of fully atomistic, Molecular Dynamics (MD) simulations of hen egg white lysozyme (1iee.pdb [5]) located in the neighbourhood of a mica surface. Protein adsorption is driven by electrostatic forces and so strongly depends on ionic strength. We have therefore examined two systems: solvated, neutral hen egg white with ionic strength equal 0.5 M and 0.02 M respectively. As a reference, a trajectory obtained for the isolated and solvated protein in ionic strength 0.5 M is also used. Careful analysis of three 10 ns trajectories provides an insight into early events during lysozyme adsorption on the mica surface, as well as influence of the surface and different ionic strength on the protein structure and stability.

References:
1. P. A. Mulheran, D. Pellenc, R.A. Bennett, R.J. Green and M. Sperrin, Mechanisms and dynamics of protein clustering on a solid surface, Phys. Rev. Lett. 100, 068102, 2008.
2. S. Ravichandran and J. Talbot, Mobility of Adsorbed Proteins: A Brownian Dynamics Study, Biophys J., 78, 110 – 120, 2000.
3. S. Ravichandran, J. D. Mandura and J. Talbot, A Brownian Dynamics Study of the Initial Stages of Hen Egg-White Lysozyme Adsorption at a solid surface, J. Phys. Chem. B. 105, 3610 – 3613, 2001.
4. F. Carlsson, E. Hyltner, T. Arnebrant, M. Malmsten and P. Linse, Lysozyme Adsorption to Charged Surfaces. A Monte Carlo Study, J. Phys Chem B, 108, 9871 – 9881, 2004.
5. C. Sauter, F. Otalora, J. A. Gavira, O. Vidal, R. Giege, J. M. Garcia-Ruiz, Structure of Tetragonal Hen Egg-White Lysozyme at 0.94 A From Crystals Grown by the Counter-Diffusion Method, Acta Crystallogr. D, 57, 1119-26, 2001.

Internal dynamics of ribosomal elongation factors G and Tu studied with all-atom and coarse-grained molecular dynamics

Translation process is modulated by the binding of various factors to the ribosome - e.g., by elongation factor Tu (EF-Tu) and elongation factor G (EF-G). EF-Tu is responsible for positioning the incoming aminoacyl-tRNA in the ribosomal A-site. After the peptide bond is formed, EF-G helps to translocate the peptidyl-tRNA from the A-site to the P-site in the preparation for the next catalytic cycle. These factors share a common binding site on the ribosome.
The structure of EF-G closely resembles that of the complex between EF-Tu and tRNA. The N-terminal region of EF-G is homologous to EF-Tu, and the C-terminal region comprises a set of protein domains that adopt the shape of a tRNA molecule. Both EF-G and EF-Tu (complexed with tRNA) undergo conformational changes upon binding to the ribosome.
Based on molecular dynamics simulations conducted with all-atom and coarse-grained force fields, we describe and compare internal dynamics of both elongation factors.

Efficient Molecular Docking of Drug Molecules into DNA and Protein Targets and their Enrichment by Cutting-Edge Technologies.

Several novel DACA analogues were designed, synthesized and crystallized in reaction with the G-Quadruplex DNA that eventually failed to diffract in all strong light sources including synchrotron. These analogues were studied for their binding interactions and stabilisation potential of a G-Quadruplex DNA (TGGGGT)4 using state-of-the-art molecular docking technologies. Out of several algorithms tested, the results of spherical polar coordinate shape complimentarity based HEX docking, and fragment growth based Glide-XP (using OPLS 2005 force field) docking technologies resulted in realistic binding poses that were comparable to the crystal structure of a similar drug molecule (PDB ID: 1O0K). Docking with rigid receptor and partially flexible receptor approaches using Autodock4 was not successfull in producing the drug binding poses similar to the crystal structure. These results were then enriched by describing the polarization of charge field in the drug – DNA interface by QM/MM single point calculations using 6-31G* basis set and B3LYP density functional methodology through Quantum Polarized Docking Workflow in Schrodinger suite. Stable drug bound complexes with minimal bound free energy show potential to evolve as anti-cancer agents.
On the other hand, molecular mechanisms underlying agonism and antagonism of Androgen Receptor Ligand Binding Domains were explored by docking well proven agonist and antagonist molecules into its active site. The results of rigid receptor or limited flexible receptor as employed in Autodock4 was poor in capturing the drug induced effects both in the interiorly buried active site and a distantly connected AF-2 active site in the outer surface. The induced fit effects of drug binding are currently being studied in more detail for obtaining better results by recursive algorithms. These algorithms dock the drug molecules initially with a reduced van der Waal’s radius followed by displacing the clashing protein parts with protein structure prediction algorithms and redocking the drug analogues through the Induced Fit Docking workflow in Schrodinger suite. This technology has been reported to produce the drug induced movements as efficiently as effective simulation techniques like molecular dynamics. The outcomes of these docking experiments will allow us to study the drug induced alterations in molecular geometries in the ligand binding site and their effects on adjuscent AF-2 active site which regulates transcription process for further cell production. In future, these drug molecules can be structurally altered based on these results for improving their binding efficiencies by covering a much larger space in the active site effecting a tighter binding.

Applications of a Novel Biasing Potential to Study DNA Base Flipping and DNA Translocation

DNA transcription, replication, and damage repair usually involve DNA-protein interactions and structural distortion of the DNA duplex by various enzymes. For example, during DNA metabolism, DNA helicases have been shown to separate duplex DNA into individual strands by translocating along single stranded DNA (ssDNA) while hydrolyzing ATP [1, 2]. Alternatively, various enzymes employ a base-flipping mechanism to tackle DNA repair [3]. Experimental studies have demonstrated that DNA translocation and flipping of DNA base pairs typically occurs on the millisecond or longer timescale [4, 5]. However, current computational methods are limited to the nanosecond timescale. Thus, external restraints are often employed to enhance sampling in these low probability regions of phase space [6]. While flipping of individual bases using different restraints has been well established [3], computational studies related to DNA translocation with respect to proteins is, to the best of our knowledge, slowly emerging [7, 8]. In this study, umbrella sampling with a novel center-of-mass projection onto a predefined path reaction coordinate was utilized to study base flipping of a central (underlined) cytosine base in the sequence GTCAGCGCATGG using implicit solvent. In addition to base flipping, further application of this new biasing potential was extended to examine the relative free energies involved in DNA translocation in the context of the transcription factor, E2F-DP, protein-DNA complex.

1. Caruthers, J.M. and D.B. McKay, Helicase structure and mechanism. Current Opinion in Structural Biology, 2002. 12(1): p. 123-133.
2. Saha, A., J. Wittmeyer, and B.R. Cairns, Chromatin remodelling: the industrial revolution of DNA around histones. Nature Reviews Molecular Cell Biology, 2006. 7(6): p. 437-447.
3. Priyakumar, U.D. and A.D. MacKerell, Jr., Computational approaches for investigating base flipping in oligonucleotides. Chem Rev, 2006. 106(2): p. 489-505.
4. Dillingham, M.S., D.B. Wigley, and M.R. Webb, Demonstration of unidirectional single-stranded DNA translocation by PcrA helicase: measurement of step size and translocation speed. Biochemistry, 2000. 39(1): p. 205-12.
5. Gueron, M. and J.L. Leroy, Studies of base pair kinetics by NMR measurement of proton exchange. Methods Enzymol, 1995. 261: p. 383-413.
6. Torrie, G.M. and J.P. Valleau, Nonphysical Sampling Distributions in Monte Carlo Free-Energy Estimation: Umbrella Sampling. J Comp Phys, 1977. 23: p. 187-199.
7. Yu, J., T. Ha, and K. Schulten, Structure-based model of the stepping motor of PcrA helicase. Biophys J, 2006. 91(6): p. 2097-114.
8. Yu, J., T. Ha, and K. Schulten, How directional translocation is regulated in a DNA helicase motor. Biophys J, 2007. 93(11): p. 3783-97.

Fast, automated structure-based classification of kinases

We have been developing phylogenetic methods which are based on protein structure, but are fast enough to be applied to large families. Here we consider the example of kinases.
In many protein families, we believe there are evolutionary relationships for functional reasons, but the family may have diverged so far, that sequence alignments can become quite unreliable. The protein kinases and phosphorylases may be a good example. They are present in most forms of life and have evolved roles ranging from metabolic to regulatory. There is almost no significant sequence similarity between the more distant members, so it is difficult to build reliable sequence alignments. This family, however, has been popular amongst crystallographers, so there is a wealth of solved structures. This makes it an ideal candidate for a structure-based phylogeny.
The only similar project in the literature was based on 31 kinase structures and required human intervention to construct a phylogenetic tree. Here, we show how one can use many hundreds of structures to build a phylogeny completely automatically. Furthermore, the resulting tree maps rather well to annotated functions and known biochemical features.

Molecular modelling techniques and in-vitro mutagenesis for the characterization of the quinolone-gyrase-interaction

The functional target of the fluoroquinolone-(FQ) antibacterials has been identified as the bacterial gyrase enzyme complex with DNA by mapping resistance mutations. However, the exact molecular interaction is largely unknown. DNA-gyrase is able to alter the topology of DNA by a transient cleavage of the dsDNA. This is performed by an esterification of gyrA-Tyr122-OH of the enyzme to a 5‘-phosphate of the DNA. FQs inhibit the religation of the DNA by stabilizing the cleaved form, ultimately resulting in bacterial cell-death. Understanding the drug-enzyme-DNA interaction in molecular terms could be the basis for the development of new FQ-derivatives refractory to resistance. Since structural information is only partially available, this work aims to build a molecular complex by using molecular modelling techniques. After building an appropriate dsDNA, a protein-DNA-docking approach was used initially to search for binding modes. This yielded hundreds of different complex geometries that had to be assessed manually to identify the plausible complexes. We claim that, in order for the ester bond to form, the DNA must approach the protein in roughly the right configuration. Therefore, starting from the docking step, we performed molecular dynamics simulations of the promising non-covalent protein-DNA complexes. These simulation results were used to introduce the covalent bond between protein and DNA, to again perform molecular dynamics calculations, and to apply molecular docking for eleven different FQs. This theoretical data enables us to figure out amino acids which can be used for in-vitro mutagenesis to test the validity of the model. Our data might be used for in-vitro selection experiments for the identification of amino acids mutations which lead to resistance.

Looking for Inhibitors of RIO Kinases

RIO1, RIO2 and RIO3, atypical serine protein kinases, are conserved among archaea and eukaryotes. They are involved in the ribosome synthesis, the process fundamental to cell growth and proliferation. RIO1 and RIO2 kinases differ with their binding sites [1]. This allows to design inhibitors which can target distinct pathways. Based on the RIO2 structure (1ZAO, PDB) possible binding sites were analyzed. The structures were also titrated using the Poisson-Boltzmann model [2]. Novel, potential inhibitors were designed. Selected inhibitors interacting with the largest binding pocket will be presented.

Acknowledgements. These studies are supported by PBZ-MIN-014/P05/2005 funds and CoE BioExploratorium.

[1] N.LaRonde-LeBlanc, & A. Wlodawer, The RIO kinases: An atypical protein kinase family required for ribosome biogenesis and cell progression (Review), BBA, 1754, 14-24 (2005)
[2] M. Dlugosz, M. Geller and L. Walewski, Entry to modelling - protonation states of ionizable groups of RIO-1 kinases, Acta Biochim. Polon., 54, Suppl. 3, 61-61 (2007)

Modelling of Possible Binding Modes of Caffeic Acid Derivatives to JAK3 Kinase

Janus kinases (JAKs) belong to receptor-associated protein tyrosine kinases. JAK/STAT signaling pathways are responsible for transduction of growth factors and cytokine-mediated signals. Abnormal activation of the JAK2/STAT3 pathway was observed in many types of highly resistant brain and pancreatic cancers, and abnormal activation of JAK3/STAT5 is responsible for immunodeficiency disorders [1,2].
Caffeic acid derivatives, including AG490, are most likely competitive inhibitors of protein substrates [2,3]. Based on a sequence and structure similarity analyses of JAKs and IRK we indentified binding sites in JAK2 and JAK3. Novel JAK3 inhibitors were designed and are being synthesized. Their biological activity will be studied experimentally.

Acknowledgements. These studies are supported using PBZ-MIN-014/P05/2005 funds.

[1] H. Yu, R. Jove, Nature Reviews, 4, 97-106 (2004)
[2] L. Gu, H. et al., Bioorg. Med. Chem., 13, 4269-4278 (2005)
[3] P. Setny, B. Lesyng, W. Priebe, Acta Biochim. Polon., 54, Suppl. 3, 64-65 (2007)

1.4-DHP-lipid forms a tubular micellae

1,1’-{[3,5-bis(dodecyloxycarbonyl)-4-phenyl-1,4-dihydropyridin-2,6-diyl]dimethylene} bispyridinium dibromide (1,4-DHP lipid) is a gene transfection agent. 1,4-DHP lipid structure was calculated with ab initio quantum mechanics to obtain the charges for molecular dynamics (MD), AMBER 8.0 force field. 1,4-DHP-lipid molecules were subjected to MD from the initial structure of a periodic lipid bilayer-water box, with a small amount of excessive water on the lipid edges to ensure the mobility of lipid molecules. After 35 ns few lipid molecules turned with their charged heads to the side of the lipid bilayer and after 100 ns a profound tubular micelle structure began to form. The tubular micelle structure becomes more perfect during the course of simulation of 300 ns.
Conclusion is that one of the gene transfection agent 1,4-DHP lipid structures is a tubular micellae, and we could expect that such the micellaes are capable to form a lipoplex for the DNA transfection.

Protein Structure Prediction using Coarse Grain Force Fields

Protein structure prediction is one of the classic problems from computational chemistry or molecular structural biology. In ab initio or de novo protein modelling, one tries to build 3D protein models from scratch rather than modelling them on to known structures.
There are two aspects to this problem: 1) the score or quasi-energy function and 2) the search method. Our interest has been the development of quasi-energy functions. These could be seen as low-resolution special purpose (coarse-grain / mesoscopic) force fields, but they are rather different to most approaches. There is no strict physical model. They are statistical, but there is no assumption of Boltzmann statistics. Instead, there is a mixture of Bayesian probabilities based on normal and discrete distributions. This has an interesting consequence. If one works with a method such as Monte Carlo, one can base the acceptance criterion directly on the calculated probabilities.
Although we have not performed proper benchmark, the scoring function works reasonably well to predict 3D models of smaller proteins from their sequences. To improve prediction results further particularly of bigger proteins, our current work is focused on the improvement of solvation effects and hydrogen bonds representation.

Molecular Interaction Fingerprint (MIF): a novel approach for drug design

We have developed a novel method called Molecular Interaction Fingerprint (MIF) which is able to predict effect and side effect profiles of drug-like molecules. MIF is a binding energy pattern of a drug molecule determined by docking processes to a series of 89 different protein surfaces. We compared binding energy patterns of 969 FDA approved drug molecules to each other and found that the similarities of MIFs have high correlations to the effect and side effect profiles of the compared molecules irrespective of the structural similarities and the individual binding energies. Effect prediction reliability is more than 90% if the first twenty closest molecules are taken based on their MIF patterns. MIF database is a useful approach for finding unknown effects of existing drugs and predicting effect and side effect profiles of new drug-like molecules. MIF database (www.mifdb.com) will be available online from 15 May 2008.

Multiple alignment and classification of protein structures

We have developed a novel tool for fast multiple protein structure alignment. It is a progressive multiple alignment method which builds on a "bottom-up" pairwise structure alignment approach matching locally similar structure fragments. The progressive alignment method allows the use of average representations of groups of structures which provides similar benefits as sequence profiles. It produces high quality alignments of secondary structure elements as well as exotic loop structures. It is shown that the tool easily scales to hundreds of structures. We demonstrate the application of this method to the classification of protein structures and identification of catalytic residues.

Molecular Dynamics Simulations of Product Dissociation from the Anthrax Edema Factor: A Two-Metal Ion Binding Site Greatly Impairs Pyrophosphate Release

The Anthrax Edema Factor is a Calmodulin-dependent adenyl cyclase responsible for the overproduction of cyclic mono-phosphate (cAMP) from ATP in the host cells. For optimal catalytic activity, it should efficiently release the reaction products, CMP and pyrophosphate (PPi). Here we study the mechanisms cAMP and PPi dissociation using Locally Enhanced Sampling and Steered Molecular Dynamics simulations. Since there is no clear consensus on the number of metal ions in the catalytic site, simulations are performed in the presence of one or two cations. The simulations suggest that the presence of the second metal ion greatly impairs pyrophosphate dissociation. Analysis of previous experimental data supports this observation and thus the hypothesis of an optimal one-metal ion catalytic binding site. Further details of the dissociation mechanisms for both cAMP and PPi are discussed.

A software library for Monte Carlo based rigid body modelling based on Small Angle Scattering data

Small angle scattering (SAS) techniques probe the structure of molecules in solution. A one dimensional scattering curve as obtained by SAS directly provides some physical parameters. Contemporary modelling approaches reach beyond and SAS experiences a renaissance in biology, yielding information about 3D oligomeric organisation. The technique mostly finds its niche where conformational changes are to be traced or experimental access to protein flexibility of big complexes is required.
In general two different approaches are followed to obtain a structural model from SAS data: ab initio modelling and rigid body modelling approaches. Ab initio modelling is frequently used to get a first impression of the shape of a particle. Rigid body modelling, on the other hand, is able to provide more details, but requires initial molecular models. Several ways exist putting rigid body modelling into work with SAS data. The most precise is computationally also the most costly approach: It is possible to retrace a conformational change, by simply applying an (enormous amount of ) possible moves onto the subunits. Then a check is made which theoretical scattering curve of those conformations fits best with the experimental data. While this ‘grid search‘ provides very accurate estimations of the parameters being applied – along with uncertainties of those estimations – calculating the theoretical scattering curve many thousands of times is a severe speed bottleneck.
The presented software library divides the task into two steps: First a Monte Carlo based search in the parameter space, then a conventional grid search. In the first step a region of possibilities within the parameter space is found. The final systematic search can then be performed faster, knowing reasonable limits for the number of parameters.
The software is intentionally written as a library in the Python programming language (with some parts in C to gain speed) and not as a stand alone program. This enables easily writing more flexible scripts. Care was taken to have a simple & flexible user interface, while preventing the user to take nonsensical steps. Additional advantages of the software are:

• proteins are allowed to have arbitrary symmetry
• any domain within a protein can be considered a ‘rigid body‘
• the user can give arbitrary limits for movements, while the program assists in finding sensible ones
• example scripts are provided
• interfaces to some third-party software tools exist

Significant forces in protein-protein docking rated by an all atom free-energy forcefield - Chemokines as system analysis

Chemokines are small proteins that are recognized by corresponding receptors on leukocytes to direct them to their site of destination. We model the binding of the receptor to different types of chemokines and then calculate the binding energy to find out where points of interaction exist between the receptor and the ligand. Furthermore, we investigate which residues of the N-terminal of the receptor are crucial for binding by site-directed mutagenesis.
The simulations are performed with the POEM (Protein Optimization by Energy Minimisation) approach using forcefield PFF02 (an all-atom free-energy forcefield, which identifies the native structure of a protein/protein-complex as the global minimum of the free-energy landscape).
The results verify our understanding of the dominant contributions to the free energy of protein–protein interactions, and can guide experiments aimed at the design of protein interaction inhibitors.

A non-native helix extension channels folding in simulations

Unbiased parallel tempering Monte Carlo simulations of a 49 residue protein starting from random conformations, reveal a non-trivial path followed by the molecule to the native state. The molecule (PDB id: 2GJH) consists of an α-helix and a 3 stranded β-sheet, in which two of the adjacent strands stradle the other secondary structure elements along the sequence. In the course of folding, one of the strands making sequence non-local contacts is seen to be "cached" as a non-native extension of the native α-helix. After the other secondary structure elements have formed and assembled in their proper tertiary arrangement, the cached segment is released and it changes its secondary structure to a strand as it attaches to a β-hairpin to complete the native structure. The study is based on a physics based implicit water all-atom interaction potential called the Lund force field.

Modulation of aggregate size and shape distributions of amyloid-β peptide solutions by a designed β-sheet breaker

A peptide with 42 amino acid residues (Aβ(1-42)) plays a key role in the pathogenesis of the Alzheimer´s disease. It is highly prone to self aggregation leading to the formation of fibrils which are deposited in so-called amyloid plaques in the brain of affected individuals (1,2).
In our study we established a method to analyze the aggregation behavior of the amyloid-β peptide with a combination of sedimentation velocity centrifugation and enhanced data evaluation software as implemented in the software package Ultrascan (3).
Important information which becomes accessible by this methodology is the s-value distribution and concomitantly also the shape-distribution of the peptide aggregates generated in the process of self-association. These informations get especially valuable upon evaluating the properties of potential aggregation inhibitors.
With this method we characterized the aggregation modifying effect of a small organic molecule, designed as a β-sheet breaker.
This compound is built from three head-to-tail connected aminopyrazole moieties and represents a derivative of the already described Tripyrazole (4,5). The compound showed reduction of aggregate formation measured by FCS and decreased amyloid formation as measured by Thioflavin T measurements. By addition of this compound to a solution of the Aβ(1-42) peptide the maximum of the s-value distribution calculated for the formed amyloid-β aggregates experienced a clear shift to smaller s-values as compared to solutions where only the vehicle DMSO was added. This shift to smaller s-values was stable for at least 5 days. It could be shown that the strength of the shift was related to the amount of the added compound. The results will be discussed in terms of their significance regarding the mechanism by which the compound interferes with the fibril formation of the Aβ peptide.

(1) Hardy, J., Higgins, G. 1992. Alzheimer´s disease: The amyloid cascade hypothesis. Science 256, 184-185.
(2) Hardy, J. 2006. Has the amyloid cascade hypothesis for Alzheimer´s disease been proved? Curr. Alzheimer Res. 3, 71-73.
(3) Brookes, E., Boppana, R.V., and Demeler, B. 2006. Computing large sparse multivariate optimization problems with application in biophysics. Supercomputing. (4) Rzepecki, P. et al. 2004. Prevention of Alzheimer´s associated Aβ-aggregation by rationally designed nonpeptidic β-sheet ligands. J. Biol. Chem., 279, 47497-47505.
(5) Nagel-Steger, L. et al. 2008. A designed β-sheet breaker disassembles amyloid-β fibrils and inhibits fibril growth. Submitted.

Flexible Peptide-Protein Docking employing PSO@Autodock

Communication in biological systems occurs via specific molecular interactions. To shed light on the underlying principles these processes are investigated with the help of experimental techniques but also with biomolecular computational simulations. In this context computational molecular docking methods have proven to be viable tools. Besides their application in drug design, molecular docking techniques are employed to investigate binding interaction of natural ligands to their receptors.
The vast majority of endogenous ligands are peptides. However, their conformational flexibility renders a systematic search for the ‘correct’ binding pose impossible. Thus, flexible docking of peptides to proteins is computational demanding. Recently, we presented the novel docking program PSO@Autodock [1] for fast flexible molecular docking. It employs swarm intelligence implemented in various particle swarm optimisation (PSO) methods. PSO@Autodock efficiently screens high-dimensional search spaces.
Here, we present the application of PSO@Autodock for flexible peptide-protein docking studies and compare the results with well-established docking programs like AutoDock3 [2] or GOLD [3]. In an initial study we screened a set of 10 complexes with highly-flexible peptidic ligands (15 to 24 torsion angles). A comparison with the experimentally obtained protein-ligand complexes revealed that PSO@Autodock clearly outperforms other docking programs. The average RMSD value of all the docked complexes is with 1.6 Å significantly lower than that obtained with AutoDock3 or GOLD, which is above 3.0 Å in both cases. For instance, the co-crystallized ligand of 1lyb.pdb with 24 rotatable bonds was successfully docked by PSO@Autodock within as few as 100,000 computing steps with an RMSD 1.1 Å, while AutoDock3 and GOLD gives 3.0 Å and 5.7 Å respectively.

References:

1. V. Namasivayam, R. Günther, Chem Biol. & Drug Des. 2007 (70), 475.
2. G. M. Morris, D. S. Goodsell, R. S. Halliday, W. E. Hart, R. Belew, A. J. Olson, J. Comput. Chem. 1998, 19, 1639.
3. M.L. Verdonk, J.C. Cole, M.J. Hartshorn, C.W. Murray, R.D. Taylor, 2003, Proteins, (52), 609.

Exploration of the Energy Landscape of Protein-Protein and Antibody-Antigen Interactions

A well choreographed, dynamic interplay of protein-protein interactions is crucial for the function of a cell. To understand these interactions, knowledge of the underlying energy landscape is essential. We analyzed the energy landscape of a protein protein and an antibody antigen complex using steered molecular dynamics simulations.
First, we examined the influence of velocity and geometry of the force probing on the choice of the forced unbinding pathway of the Barnase Barstar complex. We demonstrated that in our constant velocity probing experiments, a change in the probing velocity may switch the unfolding pathway. Further more, we showed, that changing the geometry of the force probing can be used to choose between different unbinding pathways. These tools may be used for a pre-chosen sampling of the protein complex energy landscapes.
The second part of our work focussed on the examination of the dependence of the unbinding pathway on the force attachment point. The truncated leucine zipper GCN4 peptide was separated from the anti-GCN4-antibody fragment H6 in SMD simulations. Three different attachment points were examined: the C and N terminal Cα of the 12 amino acid long peptide as well as a Cα in the middle of the peptide. We identified one common barrier on the unbinding pathway formed by a common, central unbinding interaction.
Additionally, we classified the correlation between MD simulations and AFM as well as SPR measurements. We could show, that, in the examined system, the AFM probes the first barrier found in our MD simulations. Further more, our MD trajectories showed the existence of two main unbinding barriers. This supports the theory, that AFM and SPR may test different barriers. The second barrier, which is rate determining under equilibrium conditions, is tested via SPR, while the inner barrier is probed via AFM, due to the forced tilting of the energy landscape inherent to force spectroscopy measurements.

From Isotope Labeling Patterns to Metabolic Flux Rates

Background: Intracellular metabolic flux rates are the manifestation of the metabolic activities within organisms. The fluxome determines the physiological phenotype of the cell and, thus, is linked to its metabolic capability. These intracellular reaction rates, however, cannot be accessed directly from metabolome concentration data. Isotope labeling experiments offer the perspective to bridge this gap. Based on a mathematical model, several computational steps have to be performed to precisely quantify the metabolic fluxes. In this context, Metabolic Flux Analysis (MFA) based on labeling experiments has become a key technology in Systems Biology.

Current practice, recent developments and the general computational procedure: Over the past decade two types of isotope-based MFA emerged: the classical metabolically and isotopically stationary MFA and its promising isotopically non-stationary generalization. Altogether, experimental progress and new analytical techniques result in an increased demand for more efficient and novel computational evaluation methods.
Based on a biochemical reaction network and its associated atom transitions, a mathematical model describing the emerging labeling distributions constitutes the centerpiece of all computational MFA routines. However, its solution is also the limiting operation for all computational steps downstream the data evaluation - i.e. parameter fitting, statistical analysis, optimal experimental design, or even investigating model variants. The requirement for computational evaluation of the model is in each case high: the classical MFA method involves the solution of an algebraic equation system (typical dimension >> 5000), while for the non-stationary approach a high-dimensional system of ordinary differential equations (typical dimension >> 500), which is nonlinear and possibly stiff, has to be solved. Although the algorithms underlying this solution step are different, basically all of them rely on the structure of the isotope labeling network graph associated with the metabolic and atomic network.

Results: Exploiting the deep nature of the underlying (algebraic and differential) equation systems significantly improves the efficiency of the solution methods. The contribution will discuss different approaches of the new algorithms, having their roots in Graph Theory, Linear Algebra, and Compiler Theory. Particularly for the classical approach new algorithms providing analytical and fast numerical solutions will be presented which open the perspective to simulate even genome-scale metabolic models.

The inherent stability of collagen

Collagen is an extracellular protein forming cables and layers, thus strengthening our tissues. In my poster I will show how collagen preserves in itself the structure of polyglycine II. In this way not only it becomes able to form tropocollagen triple helices and so cables, but also the possibility of an accidental amyloid formation is avoided.

Molecular Dynamics Simulations of the Metaloenzyme Thiocyanate Hydrolase with Non-Corrinoid Co(III) in Active Site.

Thiocyanate Hydrolase (SCNase, pdb code 2DD5) is a novel metaloenzyme containing non-corrinoid Co(III) in the active site. Despite identical structure of the active sites, high sequence and structural similarity of SCNase and nitrile hydratases (NHases) both enzymes catalyse different reactions. The enzyme catalyses the degradation of thiocyanate to carbonyl sulfide and ammonia but related NHases the hydration of nitriles to amides. The main goal of the present work was to explain this different properties on molecular level. Extensive moleclular dynamics simulations (up to 10 ns) were performed using CHARMM27 forcefield with specially designed parametrization of the active site. Particular attention was devoted to water dynamics in the catalytic region, dynamical properties of entry channel and preferential docking sites for a substrate and products of SCNase. The theoretical modelling provides useful data for understanding this enzyme having properties useful in biotechnology.

Constraint network analysis: A computational framework for characterizing protein stability features

The molecular basis of stability relates closely to contemporary issues in protein science such as the protein folding problem, protein-protein interaction and protein-ligand binding. In addition, protein stability has industrial importance. The identification or the development of enzymes with higher stability will increase the adoption of biocatalytic syntheses in industrial production.[1] Understanding and exploiting the relationship between microscopic structure and macroscopic stability is important for developing strategies to improve protein stability in the reaction media used in industrial processes.
Due to very large contributions of stabilizing and destabilizing interactions involved in the formation of the folded state, protein structures can be described as molecular networks.[2] The analysis of protein structure networks offers a fast and efficient computational way for studying and predicting the relationship between microscopic structure and macroscopic stability.[3] Using the Floppy Inclusion and Rigid Substructure Topography (FIRST) method, the rigidity or flexibility in a protein structure network can be quantified.[4] By the dilution of non-covalent contacts in the protein structure network, FIRST has been employed to simulate thermal unfolding.[5] In going from a rigid to a flexible network, a phase transition can be observed that defines the rigidity percolation threshold and corresponds to the folded-unfolded transition in protein folding.
In the present study, thermal unfolding simulations are applied to a dataset of homologous proteins from thermophilic and mesophilic organisms. Using concepts from percolation theory and network science, the temperature of the phase transition can be identified. Our results show, as expected, that proteins from thermophilic organisms have higher phase transition temperature values than their counterparts from mesophilic organisms. We demonstrate that the approach allows for characterizing structural features in atomic detail that determine the stability of a protein structure. This information might be exploited in data-driven protein engineering by pointing to residues that should be mutated to obtain a protein with higher stability.

[1] Demirjian, D.C. et al. Curr. Opin. Chem. Biol. 2001, 5, 144-151.
[2] Jacobs, D.J., Thorpe, M.F. Phys. Rev. Lett. 1995, 75, 4051-4054.
[3] Böde, C. et al. FEBS Lett. 2007, 581, 2776-2782.
[4] Jacobs, D.J. et al. Proteins 2001, 44, 150-165.
[5] Rader, A.J. et al. Proc. Natl. Acad. Sci. U.S.A. 2002, 99, 3540-3545.

Bias detection in thermodynamic integration: getting correct ensemble averages

Thermodynamic integration is a powerful tool to calculate free energy differences between different systems/states. It requires, however, the accurate estimation of equilibrium ensemble averages for several hybrid systems on an artifical path from one system to another (alchemical path).
Correct estimates from efficient timesaving simulations can only be gained by a robust detection of the initial bias that each simulation contains.
A specific strategy for bias detection has already been suggested. Here we combine the central idea of this strategy with an algorithm for the error calculation of ensemble averages. The result is an easy to implement, robust and automatable method for bias detection.

A multiscale approach to protein structure prediction

Theoretical prediction of protein structures remains one of the most demanding tasks of computational biology. Good theoretical models could be useful in predicting of protein enzymatic function and protein interactions, in molecular replacement in crystallography, in refinement of low resolution NMR-based models and in analysis of remote evolutionary connection between proteins.
The proposed approach employs mezoscopic CABS modeling tools [1] supported by weak distance restraints derived from related structures (even very remotely related) with the help of standard bioinformatics tools and metaservers. The low resolution models obtained from the CABS simulations are then subject cluster analysis [2] followed by the all-atom reconstruction [3,4]. The top models are rank-order using an all-atom force-field after a short energy minimization [5].
The proposed procedure is tested on a representative set of protein targets, and the results are compared with earlier simulations on the same set.

1. A. Kolinski, “Protein modeling and structure prediction with a reduced representation”, Acta Biochimica Polonica 51:349-371 (2004)
2. D. Gront and A. Kolinski, “HCPM – program for hierarchical clustering of protein models”, Bioinformatics, 21:3179-3180 (2005)
3. D. Gront, S. Kmiecik & A. Kolinski, “Backbone building from quadrilaterals. A fast and accurate algorithm for protein backbone reconstruction from alpha carbon coordinates”, J. Comput. Chem. 28(9):1593-1597 (2007)
4. S. Kmiecik & A. Kolinski, “Characterization of protein folding pathways by reduced-space modeling”, Proc. Natl. Acad. Sci. USA 104(30):12330-12335 (2007)
5. S. Kmiecik, D. Gront & A. Kolinski, “Towards high-resolution structure prediction. Fast refinement of reduced models with all-atom force field”, BMC Structural Biology 7:43 (2007)

A molecular dynamics approach to study the importance of solvent in protein interactions

Water molecules are present ubiquitously in living cells. However, solvent contribution to protein-protein interactions is often ignored in protein-protein interactions studies. Previous work has suggested the importance of wet spots (residues interacting only through one water molecule) in description of protein interactions. We use a molecular dynamics approach to analyze solvent at protein interfaces to gain insights into its contribution to protein recognition and to characterize dynamic and energetic properties of wet spots.
Our results show that wet spots fluctuate less than surfacial protein residues, and that from an energetic point of view they are quantitatively comparable to other residues in protein interfaces. The residence time of water molecules in wet spots sites is found to be significantly higher than of water molecules on protein surfaces. In terms of free energy these water molecules are heterogeneous. Nevertheless, their contribution to the free energy of complex formation significantly changes the energy function of the system suggesting that water should be considered in detailed protein interface description. We find that the participation of solvent in protein interfaces allows higher sequence variability in complex counterparts and at the same time contributes to conservation of protein-protein interactions.
We believe that our results are useful for deeper understanding of the physico-chemical properties underlying protein-protein interactions. The wet spots concept might be further utilized to qualitatively improve the accuracy of docking, folding and drug design algorithms. Currently we are working on the implementation of the wet spots concept into algorithms for protein interactions prediction.

Nearly-deterministic Methods for Optimising Protein Geometry

Protein structure prediction could be seen as either a challenge or an algorithmic playground. We are certainly interested in algorithmic improvements. Selfconsistent mean field methods (SCMF) have traditionally been used in areas such as wave function optimisation or protein side-chain placement. We have been trying to apply the ideas to find the most likely conformation for a protein. The philosophy relies on precalculated distributions of structural descriptors given a set of known properties (a protein's sequence). Starting with a sequence, which is decomposed into small overlapping fragments, the conformational space is described by a fixed number of weighted multivariate Gaussians (the known distributions). As the conformational bias, introduced by the sequence fragments, is local and putatively inconsistent in overlapping parts the weights of the Gaussians for all overlapping fragments can be updated iteratively in an optimisation step. Unlike other approaches, like molecular dynamics or Monte-Carlo simulations, the optimisation is done in probability space rather than on some initial structure. Therefore, we do not need to calculate energies as in standard SCMF. When the iteration converges sample structures are generated from the weighted Gaussians. Although not all problems are solved yet, the current results show that the procedure is able to find protein-like structures. We also use this principle to predict protein sequences from structure, where our approach performs reasonably well.

Folding Channels for Coarse-Grained Heteropolymer Models

Applying multicanonical simulations [1] we investigated off-lattice heteropolymers employing the three-dimensional AB model [2,3]. The heteropolymers consist of hydrophobic (A) and hydrophilic (B) monomers. Their energy is obtained from specific Lennard-Jones potentials between nonbonded pairs of these monomers in addition to the chain's bending energy. In particular, AA contacts are favored to allow the formation of a hydrophobic core. We study for three permuted AB sequences the folding channels in the free-energy landscape by comparing the equilibrium conformations with the folded state in terms of an angular overlap parameter [4]. Although containing the same number of A (14 each) and B-monomers (6 each), the analysis of the folding channels of the exemplified sequences reveals a variety of characteristic folding behaviors known from realistic peptides.

[1] B. A. Berg and T. Neuhaus, Phys. Lett. B 267, 249 (1991).
[2] F. H. Stillinger, T. Head-Gordon, and C. L. Hirshfeld, Phys. Rev. E 48, 1469 (1993).
[3] S. Schnabel, M. Bachmann, and W. Janke Phys. Rev. Lett. 98, 048103 (2007).
[4] M. Bachmann, H. Arkin, and W. Janke, Phys. Rev. E 71, 031906 (2005).

Effect of surfaces on the aggregation of hydrophobic and hydrophilic amyloidogenic peptides

Orientational ordering of peptides enhances due to their adsorption on surfaces and can thus play an important role in the formation of ordered peptide aggregates. To explore the general effect of surface hydrophobicity/hydrophilicity on peptide adsorption and aggregation, we performed a series of computer simulation studies of oversaturated aqueous solutions of peptides in slit-like pores with smooth walls interacting via a (9-3) LJ potential with water molecules, only. Two kinds of pore walls were considered: a hydrophobic paraffin-like wall, which causes pronounced water density depletion, and a hydrophilic silica-like wall, which causes formation of two highly ordered water layers. Two kinds of amyloidogenic peptides were used: the hydrophobic peptide NFGAIL, (residues 22-27 of the human islet amyloid polypeptide), and the polar hydrophilic peptide GNNQQNY (residues 7-13 of the yeast prion Sup35). To study peptide aggregation, six peptides were placed randomly in the pore and five independent 70 ns simulations were performed for each system. Strong adsorption of peptides on the pore wall is observed only in the case of the NFGAIL peptide in a hydrophobic pore, where all peptides are adsorbed and aligned parallel to the walls already after 30 ns. In the other three cases considered, the peptides are repelled from the walls, localized near the pore center and do not show orientational ordering with respect to the walls. Our results show that even a single factor such as the water density distribution has a drastic effect on the character of peptide aggregation near surfaces. A wider diversity of possible scenarios can be expected when specific peptide-surface interactions are taken into account.

Free energy study of Ion permeation through Gramicidin

The pentadecapeptide gramicidin forms a cation-specific ion channel in membrane environment. Two conformations are known up-to-date: the head-to-head helical dimer (HD) and the intertwined double helical form (DH). These two conformations are favored depending on the specific conditions but the biologically active form is still a matter of debate. Nevertheless, due to its small size, the gramicidin serves as an excellent ion channel model for both computational and experimental studies.
In this work, we focus on the energetics of single potassium ion permeation by means of the potential of mean force (PMF) for both gramicidin conformations using molecular dynamics simulations. Our results show that the HD has a significantly higher central barrier than the DH, implying an increased ion conduction of the latter. The barrier to ion passage is found to be closely related to the channel flexibility. In addition, the binding site location of DH is in accord with experiment. Multiple ion permeation appears significantly facilitated for the DH conformation due to its opposing pore water dipole moments at the pore entrances.
While the current force field parameters clearly underestimate the ion-protein interaction, we show that with adjusted Lennard Jones 6-12 parameters the PMF profiles could be improved.

Free-energy based all-atom protein folding using worldwide distributed computational resources

Following Anfinsen's thermodynamic hypothesis we have implemented massively parallel stochastic optimization methods for all-atom de-novo protein folding using our free-energy forcefield PFF02[1]. We have implemented this approach (POEM) using a world-wide volunteer computational grid to predictively and reproducibly fold several proteins with up to 57 amino acids, including the engrailed homeodomain and protein A, from completely unfolded conformations.
POEM identifies the native conformation of the protein as the global minimum of the protein free-energy forcefield PFF02, which stabilized the native conformation of all 32 monomeric proteins (without cofactors) against all decoys in the Rosetta decoy set[2]. In addition we could fold a set of 13 proteins with helical, sheet and mixed secondary structure from completely unfolded conformations to near-native conformations, to an average 2.87 Å resolution[1-3].
In this investigation, we deployed a BOINC server implementing an evolutionary strategy[4], which explores the free-energy landscape in many parallel dynamical processes, which communicate with one another through a central server. The overall computational work is thus segmented into medium size work-units, which can be processed independently. The algorithm evolves a population of conformations towards the global optimum of the free-energy surface by balancing energy improvement with population diversity. POEM@HOME (http://boinc.fzk.de) thus implements a complementary approach to existing distributed computational proteomics initatives, such as Folding@Home or Rosetta@Home, to help analyze structure and function of large, experimentally relevant proteins.

References
1. Verma, A. and Wenzel, W., "Towards a universial free-energy approach to all-atom protein folding and structure prediction" (preprint)
2. Verma, A. and Wenzel, W., "Protein structure prediction by all-atom free-energy refinement" BMC Structural Biology 7, 12 (2007)
3. Gopal, S.M., Wenzel, W., "De Novo Folding of the DNA-Binding ATF-2 Zinc Finger Motif in an All-Atom Free-Energy Forcefiel" Angew. Chemie Intl. 45, 46, p7726 (2006)
4. Schug, A., Wenzel, W., "An Evolutionary Strategy for All-Atom Folding of the 60-Amino-Acid Bacterial Ribosomal Protein L20"; Biophys. J. 90, p4273 (2006)

All atom-simulations of protein unfolding - The role of polarity for the denaturation power of urea

Protein unfolding by denaturants such as urea is a widely used technique to study protein folding and stability. Despite the widespread use of urea, however, the molecular mechanism of urea-induced protein denaturation is not yet fully understood. In particular, two opposing mechanisms are controversially discussed, according to which either hydrophobic or polar interactions are the dominant driving force. To advance understanding of the process at the molecular level, we have performed comprehensive MD simulations following different approaches. First, the interactions of urea with all 20 individual amino acids were investigated independent of sequence or structure. This study revealed a clear profile of interaction preferences with either water or urea. Almost all amino acids showed preference for contacts with urea molecules, whereas charged and polar amino acids were found to have slight preferences for contact with water molecules. Particularly strong preference for urea contacts were seen for aromatic and apolar side-chains, as well as for the protein backbone of all amino acids. These results suggest that favorable apolar contacts between urea and apolar residues reduce the hydrophobic effect and thus promote unfolding. Further, an analysis of hydrogen-bond energies allowed to reassess the role of urea-backbone interactions: whereas hydrogen bonds between urea and the backbone do not drive the denaturation, they do contribute to the overall energetics by helping to avoid unfavorable unsatisfied hydrogen bond sites of the backbone, while at the same time shielding it from entropically unfavorable water contact. These findings motivated a "Gedankenexperiment": if apolar interactions are indeed the driving force for denaturation, urea with decreased polarity should be an even stronger denaturant than "regular" urea; and urea with increased polarity should be a weaker denaturant. To test this hypothesis, multiple simulations of the CI2 protein up to one microsecond each were performed with different urea partial charge scalings. Indeed, protein unfolding was observed in all simulations with decreased urea polarity, whereas increased urea polarity even slightly stabilized the native state. Apart from demonstrating that urea-induced denaturation is indeed driven by hydrophobic interactions, our simulations also allowed an investigation of unfolding pathways and residual structure of the denatured state. In summary, our results suggest a mechanism which is a synthesis of seemingly opposing viewpoints.

Parallel 2d-Wavelet Transform on the Cell/B.E. for fast Calculation of Coulomb Potentials

The calculation of long range interactions is a computationally demanding task in particle based simulations. To reduce the computational complexity from O(N²) to O(N) an algorithm based on a fast 2d-Wavelet transform technique was developed. In this algorithm, a CPU and memory demanding part is to construct the 2d-Wavelet transform of a grid based inverse distance matrix. The 2d-Wavelet transform is thereby calculated via a triple-matrix-multiply and threshold procedure, which results in a sparse representation in Wavelet space. To accelerate the calculation of the triple matrix multiplication the capabilities of the heterogeneous multicore processor Cell Broadband Engine (Cell/B.E.) are explored.
An efficient implementation of the Wavelet transform is developed by considering the architectural requirements of Cell/B.E. Via this implementation, the difficulties and problems in porting code to Cell/B.E. and using sparse linear algebra operations on the processor are assessed.

RedMD - a package for reduced molecular dynamics

We developed an RedMD package to perform molecular dynamics simulations for coarse-grained models of proteins, nucleic acids and its complexes. Simulations can be performed in microcanonical ensemble as well as with Berendsen and Langevin thermostats. We provide tools to generate initial configuration and topology which are based on the elastic network approach. However, they can be easily modified by users and one can write extensions to add for example a new potential type. The code is written in C/C++ language and the structure/topology of a molecule is based on an XML format. The code is distributed under GNU public licence and will be available at http://bionano.icm.edu.pl/.

Insights into the Self-assembly of Phenylalanine Oligopeptides by Replica Exchange MD Simulations with the GBSW Implicit-Solvent Model

The Phenylalanine Dipeptide (FF), the core recognition motif of the Alzheimer’s β-amyloid peptide, self-assembles into tubular structures of high stability [1,2]. We have studied the aggregation properties of FF and the related Phenylalanine Tripeptide (FFF) by 0.4-μs implicit-solvent Replica Exchange MD simulations of aqueous FF and FFF solutions. The FF and FFF peptides form ellipsoidal aggregates with a similar density and shape in the simulations. Within each aggregate, we observe structural features, which are consistent with the properties of L-Phe-L-Phe crystals [3]. In particular, the aromatic planes of interacting sidechains are mainly oriented perpendicular to each other and the backbone moieties of several (2-6) adjacent peptides interact frequently by head (NH₃⁺)-to-tail (^-OOC) hydrogen bonds, forming open or closed (ring-like) linear networks. The ring networks of six peptides observed in the FF simulations are reminiscent of the hexagonal FF rings in the L-Phe-L-Phe crystals. In some cases, two networks are connected by β-bridges, leading to more complex structures which resemble the packing of adjacent rings in the FF crystals. The rings are more stable energetically than the open networks, due to both non-polar and polar interactions. The network propensity is higher in the FFF solution, due to the larger number of non-polar interactions among FFF peptides, and the smaller screening of electrostatic interactions in the FFF aggregate; in line with this observation is the somewhat higher stability of the FFF aggregate, observed in the simulations.

1. M. Reches and E. Gazit. Science, 2003, 300, 625–628.
2. E. Gazit. Chem. Soc. Rev. 36, 2007, 1263–1269.
3. C. H. Gorbitz. Chem. Commun., 2006, 2332–2334.

Analysis and classification of the structural interactome

Experimentally determined structures of protein complexes at atomic detail are deposited in the Brookhaven Protein Data Bank (PDB). Many efforts have been recently made towards the development of tools that allow mining of such enormous amount of three-dimensional data. However, there is still a need for accurate description and clustering of protein interfaces to be used for comparative analysis in large-scale.
We have developed the SCOWLP database for a complete characterization and visualization of protein interfaces. It allows us to extract three-dimensional information of interfacial residues and solvent from all protein-protein and protein-peptide complexes of the PDB at atom, residue and domain level. We use the classification scheme of SCOP for fold and protein family definitions. The inclusion of water enriches the definition of protein interfaces by considering residues interacting exclusively by water, defined as wet spots. Mapping the interacting residues and wet spots into the protein sequences allows us to cluster the domains of a protein family based on Pairwise structural alignments and to catalog SCOP family complexes at binding site (BS) and interface (IF) levels.
The SCOWLP web-server (www.scowlp.org) allows the user to search and display all interfacial information contained in the PDB in an automatic and user-friendly fashion. The user can navigate and query our SCOP-extended hierarchy to search and/or select a particular protein family or PDB Id. The analysis and comparison of the detailed interaction information is performed by using sorting tables and an interactive 3D viewer.

Characterization of the binding surface of the human protein GABARAP

GABA(A) receptors are ligand-gated chloride channels that mediate inhibitory neurotransmission. The GABA(A) receptor associated protein (GABARAP) interacts with the gamma2 subunit of the GABA(A) receptor, modulates channel kinetics and promotes receptor clustering. Two hydrophobic pockets acting as indole binding sites were identified as major determinants of the ligand specificity of GABARAP by two dimensional NMR.
We identified peptide K1 that binds GABARAP with high affinity. Co-crystals of GABARAP and K1 diffract to 1.3 Å resolution. Each hydrophobic pocket of GABARAP is occupied by a tryptophan residue of the peptide.
Recently we found that calreticulin binds GABARAP. Co-crystals of GABARAP and calreticulin (178-188) diffract to a resolution of 2.3 Å. In this case the two hydrophobic pockets are occupied by a tryptophan and a leucine, respectively. This is the first complex structure of GABARAP with a native ligand.

Thermodynamics and kinetics of peptide folding

The current scenario of the protein folding problem can be classified into protein structure prediction and mechanism of protein folding. The first approach aims at predicting the three-dimensional structure from the amino acid sequence, the second at understanding and predicting the driving forces of the folding process of known protein structures. The latter is presently an active research field, with many groups participating with full throw [1-2]. From a computational point of view the study of folding mechanisms is mainly restricted to enhanced-sampling techniques or non-atomistic models [3] due to lack of computational resources.
Hereby we study the folding of a 15 residue beta-hairpin peptide (peptide 1), designed by Jimenez et al. [4], using atomistic-detailed molecular dynamics (MD) simulations. The folding kinetics of this peptide has also been experimentally studied using infrared spectroscopy by Gai et al. [5]. The folding time of the beta-hairpin is 0.8 μs, which makes it the fastest folder known till date.
The thermodynamic parameters and the average folding time evaluated from our all-atom MD simulations using an explicit water-model (~15 μs of simulation time) match extremely well with the available experimental data [4-5]. Additionally, the much debated role of the turn in driving the beta-hairpin folding is addressed and various structural insights are provided.

References:

• Micheal R. Shirts and Vijay Pandey. Screen savers of the world, Unite !. Science 2000
• Muñoz and W.A. Eaton. A simple model for calculating the kinetics of protein folding from three-dimensional structures. Proc Natl Acad Sci USA 1999
• Sichun Yang, José N Onuchic, Angel E García, Herbert Levine. Folding Time Predictions from All-atom Replica Exchange Simulations. J Mol Biol. 2007
• Santiveri, C. M.; Santoro, J.; Rico, M.; Jiménez, M. A. Thermodynamic analysis of beta-hairpin-forming peptides from the thermal dependence of H-1 NMR chemical shifts. J. Am. Chem. Soc. 2002
• Y. Xu, R. Oyola, and F. Gai. Infrared study of the stability and folding kinetics of a 15-residue beta-hairpin. J. Am. Chem. Soc. 2003

Freezing and Collapse of Flexible Polymers

We analyze the crystallization and collapse transition of a simple model for flexible polymer chains on simple cubic and face-centered cubic lattices by means of sophisticated chain-growth methods. In contrast to bond-fluctuation polymer models in certain parameter ranges, where these two conformational transitions were found to merge in the thermodynamic limit, we conclude from our results that the two transitions remain well-separated in the limit of infinite chain lengths. The reason for this qualitatively distinct behavior is presumably due to the ultrashort attractive interaction range in the lattice models considered here. [http://xxx.lanl.gov/abs/0710.4960]

A computational approach to study the energy transduction mechanism in the Na⁺/K⁺-ATPase

The Na⁺/K⁺-ATPase produces an electrical gradient across the cell membrane which is important to maintain the membrane potential of cells. For this active ion transport the required energy has to be transferred from the cytoplasmic nucleotide binding site to the transmembrane domain where the ion transport takes place. The mechanism of the energy transduction is assumed to be common for all P-type ATPases and is generally described by the Post-Albers cycle.
Voltage-clamp-fluorometry experiments apply voltage jumps on the Na⁺/K⁺-ATPase to initiate the transport cycle and indicate a specific activation of some transmembrane helices. We simulated the applied voltage by an "ionic capacitor" and studied the impact of the electric field on the Na⁺/K⁺-ATPase by a combination of multiconformation continuum electrostatics (MCCE) and molecular dynamics. Our calculations show a selective activation of the helices M5 and M6 by the electric field especially when the stalk-regions are included. Those helices are likely to act as energy transduction elements.

Hydrogen Bond REMD: A novel approach to study protein folding in atomic detail with explicit solvent

We add and exchange hydrogen bond potentials in Hamiltonian REMD to study protein folding in atomic detail with explicit solvent. This is motivated by a previous study, where we accelerate in silico folding of a protein by alternating hydrogen bond potentials (AHBP) that result in fast reordering of the backbone hydrogen bonds. Combining AHBP with REMD allows fast folding of the protein with the possibility to extract thermodynamic properties. Our first test system is a 16-residue polyalanine.

Study of the complex DNA-EcoRV by molecular dynamics simulation

EcoRV is a restriction enzyme produced by Escherichia Coli. The function of EcoRV is to destroy invading foreign DNA by cleaving it at a GATATC sequence. It is an important defense mechanism against viral attacks. EcoRV forms a homodimer which binds the DNA. Upon binding to DNA, EcoRV induces a kink of ~50 degrees located at the central base pair of the recognition site. We are interested in knowing how much of the bend is intrinsic to the DNA sequence, and how much is induced by the protein? Our study on the behavior of the DNA alone in water showed that EcoRV cognate DNA sequence has an intrinsic propensity to bend alone without the protein. Work is currently done on comparing the behavior of the protein EcoRV with and without DNA, and comparing the interactions of the protein EcoRV with the cognate and non cognate DNA sequence.